Release Notes =============================================================================== 3.3 Series (2013/07/30 - ) =============================================================================== 3.3.5 (tokakiboshi) 2015/02/05 * Version 3.3.5 This is a bugfix release against pgpool-II 3.3.4. __________________________________________________________________ * Bug fixes - doc: Describe explicitly that the number of slave nodes is not necessarily 1 (Tatsuo Ishii) - Fix occasinal deadlock in failover with watchdog enabled (Muhammad Usama) per bug #105 http://www.pgpool.net/mantisbt/view.php?id=105 - Fix uninitialized variable (Tatsuo Ishii) Per Coverity 1234603. - doc: Fix missing release note entries in the previous release (Tatsuo Ishii) - test: Fix wait_for_pgpool_startup in pgpool_setup to use "postgres" database rather than "test" (Tatsuo Ishii) The test database may or may not exist and if it does not, it will always retry until timeout expired (20 seconds), and regression test takes unnecessary long. - Fix possible segmentation fault in query cache (Tatsuo Ishii) - RPM: Add to support for memcached. (Tatsuo Ishii) - test: Allow pgpool_setup to set configuration file directory by setting environment variable PGPOOLDIR (Tatsuo Ishii) - RPM: Rename RPM file (Yugo Nagata) Rename filename to include the RHEL version number like PostgreSQL's official RPM file. (e.g. pgpool-II-pg93-3.4.0-2pgdg.rhel6.x86_64.rpm) - Fix node id range check bug in trigger_failover_command(). (Tatsuo Ishii) The node id should be lower than NUM_BACKENDS. Probably harmless since callers never pass node ids greater or equal to NUM_BACKENDS. - test: Fix failover script of pgpool_setup to use absolute path to the database cluster (Tatsuo Ishii) - test: Add memory leak regression test (Tatsuo Ishii) - Fix ill signal befavior of SIGCHLD in exec_ifconfig() (Tatsuo Ishii) For some reason SIGCHLD was set to SIG_IGN *and* tries to wait child process which always fails because SIGCHLD is never delivered. Due to this "ifconfig up" fails when switching to watchdog master. See "[pgpool-general: 3310] for more details. - Allow $_IP_$ having address prefix in if_up_cmd and if_down_cmd (Muhammad Usama) Previously, it is assumed that "$_IP_$" keyword will always have a white space at the end, and when "$_IP_$" has address prefix (e.g. $_IP_$/24), this prefix was ignored. Per bug #122. http://www.pgpool.net/mantisbt/view.php?id=122 - Allow "show pool_nodes" to print node ids having more than one digits (Muhammad Usama) Problem reported in [pgpool-general: 3374]. - Fix occasional failure in regression test '004.watchdog' (Yugo Nagata) The patch provided by uehara in bug #127. http://www.pgpool.net/mantisbt/view.php?id=127 - Restore SIGCHLD handler to its original after executing if_up/down_cmd (Yugo Nagata) Previously SIGCHLD was set to SIGIGN always after if_up/down_cmd. This led to a incorrect error handling of ping or other commands execution, and warning messages were output to log even when the command exited successfully. - RPM: Add pgpool.service for systemd in RHEL/CentOS 7 (Nozomi Anzai) - RPM: Split pgpool_regclass and pgpool_recovery as a separate extension package (Tatsuo Ishii) On PostgreSQL server, only the extension package is required, no pgpool full package is necessary. Per Andrew N Golovkov. - RPM: Fix wrong SSL configure option in the RPM spec file (Tatsuo Ishii) SSL enabling option is "--with-openssl=/usr" and this is wrong. This should be "--with-openssl". Per bug #115. http://www.pgpool.net/mantisbt/view.php?id=115 - RPM: Change to use systemd if it is available and adopt to PostgreSQL 9.4 (Nozomi Anzai) - Fix to disable debug mode by reloading config (Yugo Nagata) Per bug #114 http://www.pgpool.net/mantisbt/view.php?id=114 =============================================================================== 3.3.4 (tokakiboshi) 2014/09/05 * Version 3.3.4 This is a bugfix release against pgpool-II 3.3.3. __________________________________________________________________ * Bug fixes - Fix a typo of pgpool.spec (Yugo Nagata) - Fix bug that worker child process keeps failing when there's no primary backend (Tatsuo Ishii) Problem identified and fix contributed by Junegunn Choi. See [pgpool-hackers: 471] for more details. - Close listen socket when smart shutdown request is made (Tatsuo Ishii) When smart shutdown process starts, pgpool children still listen on the port and clients can send further connection requests which fail in the end. Which is not only waste of time, but also prevents a load balancer which sits in front of pgpool from realizing the pgpool is going down. Problem analyzed and patch provided by Junegunn Choi in [pgpool-hackers 474], and enhanced to take care not only inet domain socket but UNIX domain socket by Tatsuo Ishii. - doc: Add cautions that recovery commands are killed by statement_timeout of PostgreSQL. (Tatsuo Ishii) - test: Fix bug that wait_for_pgpool_startup does not work. (Tatsuo Ishii) - test: Fix test driver for watchdog to use wait_for_pgpool_startup correctly (Tatsuo Ishii) - Fix to ignore a harmless error code when executing ifconfig command (Yugo Nagata) When executing command such as ifconfig or ping by execv() in a forked child proess, wait() can fails with an error code ECHILD, although the process exists exactly, if SIGCHLD is set to SIG_IGN in Linux. We should ignore this and not treat as failure of the command. - test: Add undocumented "-j" option to the script comment. (Tatsuo Ishii) - test: Fix unwanted JDBC regression test failure (Tatsuo Ishii) - test: Fix to prevent regression test failures on slow machines (Tatsuo Ishii) - test: Allow to specify PostgreSQL library directory by environment variable (Tatsuo Ishii) - Adopt PostgreSQL 9.4 having to_regclass (Tatsuo Ishii) PostgreSQL 9.4 has built-in function "to_regclass", which has equivalent functionality of pgpool_regclass. Now pgpool searches to_regclass first then try pgpool_regclass. - Fix unportable code related to SO_REUSEPORT (Tatsuo Ishii) There are several places using setsockopt(sock, SOL_SOCKET, SO_REUSEPORT...). SO_REUSEPORT is not available on all Linux kernels and the source code uses ifdef to detect the feature in the compile time. Problem is, the binary created with kernel which provides the feature cannot run on a kernel which does not provide the feature. Fix is, detecting the feature in the run time by checking error code returned by setsockopt() and ignore it if the cause of the error is lacking the feature. - doc: Remove old restriction description which is no longer true (Tatsuo Ishii) - Fix return type of text_to_lsn() function (Yugo Nagata) This caused compile warning. - Fix file descriptor leak when daemonize. (Tatsuo Ishii) Per Coverity 1111471. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111442. - Fix pgpool.init's long-standing bug of stop/restart failure (Yugo Nagata) In previous, pgpool.init uses killproc for stopping pgpool, but there are several problems. In the new version, "pgpool -m fast stop" is used in stop command. Original patch contributed by Ryan DeShone and modified by Yugo Nagata. See [pgpool-hackers: 239][pgpool-hackers: 512]. - Allow to run pgpool_setup and regression test on Ubuntu box (Tatsuo Ishii) - test: Fix -i option (install directory of pgpool) not working (Tatsuo Ishii) - Add missing include file sys/wait.h. (Tatsuo Ishii) Per bug #104. http://www.pgpool.net/mantisbt/view.php?id=104 - Fix an infinite loop of ping error at new active pgpool when virtual IP switching (Yugo Nagata) This bug also caused hangs of the old active pgpool, since this waits response from the new active pgpool, which is in the infinite loop and not able to respond. This problem is reported by Qian Peng in [pgpool-hackers: 520]. - Disbale statement_timeout of PostgreSQL while executing online recovery (Tatsuo Ishii) Online recovery may take very long time and user may enable statement timeout. To prevent online recovery canceled by statement timeout, disable statement timeout in the connection used by online recovery. See [pgpool-general: 2919] for more details. - Fix pool_table_name_to_oid in case new to_regclass used which is introduced in PostgreSQL 9.4 (Tatsuo Ishii) The relcache func used here should return 0 if target table is not found. However to_regclass returns NULL in this case. To fix the problem now COALESCE is used. - test: Fix regression test 057 hung if run against PostgreSQL 9.4 or later (Tatsuo Ishii) PostgreSQL 9.4 or later changed libpq behavior: NULL username for connection string does not cause an error, which was expected in the test. - Fix to close pipe when ping command exits with failure (Yugo Nagata) This bug causes pipe open error: Too many open files, which leads to failure of connecting to trusted servers, and then pgpool goes to down status. - Fix Resource leaks (Muhammad Usama) Coverity issue #1222998 and #1222999. - Add a description that pcp_detach_node command disconnects existing sessions (Tatsuo Ishii) - Fix failover.sh produced by pgpool_setup (Tatsuo Ishii) It did not support there's more than 2 nodes case. It always tries to promoto node0 or node1. - Remove meaningless minus check because of unsigned int variable (Tatsuo Ishii) Coverity issue #1111419, #1111420, #1111422. - replacing strcpy() with much safer and recomended call strlcpy() (Muhammad Usama) Coverity issue #1111480. - Fix pgpool.spec to support openssl (Tatsuo Ishii) - Fix pgpool.spec to use 3.3-stable tree head (Tatsuo Ishii) - Remove unnecessary call to pool_shmem_exit() which removes semaphore when it shouldn't (Tatsuo Ishii) exit_handler checks if the process is parent or not. This is good. However, even if it is a child process, it calls pool_shmem_exit() which removes semaphore and shmem when it should not. It should be called only from parent process. Per bug #102. http://www.pgpool.net/mantisbt/view.php?id=102 - Fix get_insert_command_table_name() function not to use freed pointer (Muhammad Usama) Coverity issue #1223007. - Fix pool_handle_query_cache() to not free NULL pointer (Tatsuo Ishii) - Change connect() timeout longer (10 seconds) for flaky network (Tatsuo Ishii) Change connect() timeout in connect_inet_domain_socket_by_port() from 1 second to 10 seconds for flaky network such as AWS. This should help reducing frequent failover. - Fix reset query stuck problem. (Tatsuo Ishii) It is reported that reset query (DISCARD ALL etc.) occasionally does not finish and pgpool child remain waiting for reply from backend thus client cannot connect to pgpool. The cause of problem is not identified yet but if client suddenly closes connection to pgpool in the middle of query processing, backend may not accept the reset queries because they are not ready for query. The fix is, if frontend closes connection in unexpected way, query process loop immediately returns with new state: POOL_END_WITH_FRONTEND_ERROR and pgpool closes connection to PostgreSQL then goes back to new connection request waiting loop. Also, pgpool closes connections to backend when client_idle_limit is set and the idle limit. Per bug #107. http://www.pgpool.net/mantisbt/view.php?id=107 - Fix other dozens of coverity issues (Muhammad Usama) =============================================================================== 3.3.3 (tokakiboshi) 2014/03/24 * Version 3.3.3 This is a bugfix release against pgpool-II 3.3.2. __________________________________________________________________ * Bug fixes - installer: Fix bug that install of pgpool-II RPM fails (Yugo Nagata) - installer: Fix to set up passwordless SSH from apache@localhost to postgres@localhost (Yugo Nagata) - Fix bug that restarted pgpool's watchdog status is regarded as down (Yugo Nagata) - doc: Add mention about "listen queue" and how to increase the "backlog" in the num_init_children section. (Tatsuo Ishii) - Fix bug that watchdog status goes down even when only one of trusted servers get unpingable (Yugo Nagata) - Fix bad performance of unlogged tables detection code. (Tatsuo Ishii) Pointed out at [pgpool-hackers: 435][[pgpool-general:2325]. - Fix primary node detection logic. (Tatsuo Ishii) There's a possibility that primary node is not detected. This happens in following situation. node 0: primary, node 1: standby. Node 0 goes down. Health checking detects the fact but local status is not updated yet. Primary node finding (find_primary_node) runs. Node 0's status is yet healthy. Because find_primary_node fails to connect to node 0, it immediately returns -1 and fails to find that fact that node 1 is now primary. Fix is just continuing to look for primary node when fails to connect to a node. Per [pgpool-general: 2409]. - Add raw mode support to pgpool_setup. (Tatsuo Ishii) - test: Integrate jdbc test to regression test. (Tatsuo Ishii) - Fix jdbc DML fails when operated in raw mode and auto commit is off. (Tatsuo Ishii) This is reported in bug #92. - Fix to output error message when wd_escalation_command fails with exit code other than 0 (Yugo Nagata) This is suggested by Sergey Arlashin in [pgpool-general: 2457]. - test: Add new regression test for query cache. (Tatsuo Ishii) - Fix query cache bug with extended protocol. (Tatsuo Ishii) SELECT retrieves outdated cache even DML was executed in an expilicit transaction. The bug had been there since in memory query cache was born. Per [pgpool-general-jp: 1252]. - Add missing entries "relcache_size" and "check_temp_table" to show pool_status (Tatsuo Ishii) - Fix to output error message when if_up_cmd fails. (Yugo Nagata) This is suggested by Sergey Arlashin in [pgpool-general: 2457]. - Avoid to send queries to unrelated nodes in streaming replication mode. (Tatsuo Ishii) Pgpool-II sends certain queries, such as BEGIN, END and SET commands to all of DB nodes. However in streaming replication mode, only primary node and at most one standby node are only concerned (if primacy node is selected as the load balance node, only 1 node is concerned). See [pgpool-hackers: 464] for more details. - Fix possible buffer overrun problem and memory leak. (Tatsuo Ishii) Per Coverity 1111465 and 1111482. - Fix possible buffer overrun. (Tatsuo Ishii) Replace strcpy() with strlcpy(). Per Coverity report 1111478, 1111480, 1111481. - Add dubug messages for analyzing errors in if_up_cmd (Yugo Nagata) - Fix freeing NULL. (Tatsuo Ishii) Per Coverity 1111384. - Do not enable query cache on materialized views. (Tatsuo Ishii) Per bug #95. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111446. - Fix a segmentation fault in parallel mode with system_db_hostname is empty (Yugo Nagata) - Fix to output debug messages in processing pgpool.conf with -d option (Yugo Nagata) Previously, some debug messages in pool_get_config() ware not output even when -d option was used. - Fix JDBC exception of prepared statement including now() in replication mode (Yugo Nagata) With JDBC, when a prepared statement is executed more than PrepareThreshold times, the statement is named and Describe message is sent after Parse. With named statement, pgpool rewrite now() to parameter in replication mode. Hence, rewritten query has additonal parameter than original. In this case, ParameterDescription message sent to frontend (response of Describe) should include OIDs of the same number os original query's parameters. Otherwize, JDBC throws ArrayIndexoutOfBoundsException. This is reported in [pgpool-general-jp: 1192]. - Fix backend error of prepared statement about table which has column whoes default value is now() in replication mode (Yugo Nagata) When pgpool parses a named prepared statement with default now(), timestamps are replaced to additional parameters. So, Bind message also should inclued additional parameter format codes. However, when the number of original parameter was one, pgpool didn't handle this. This caused a error like "incorrect binary data format in bind parameter 2". - test: Fix a bug of regression test of JDBC that shows OK even when this fails (Yugo Nagata) - test: Add a new regression test for JDBC's PrepareThreshold (Yugo Nagata) - doc: Add description about parallel mode doesn't support PREPARE (Yugo Nagata) Per bug #93 =============================================================================== 3.3.2 (tokakiboshi) 2013/11/29 * Version 3.3.2 This is a bugfix release against pgpool-II 3.3.1. __________________________________________________________________ * Bug fixes - Fix incorrect time stamp rewriting in replication mode for certain time zones. (Tatsuo Ishii) Time stamp rewriting calls "SELECT now()" to get current time. Unfortunately the buffer for the current time is too small for certain time zones such as "02:30". Note that non-30-minutes-time-zone such as "0900" does not reveal the problem. This explains why we haven't the bug report until today. Bug reported in [pgpool-general: 2113] and fix provided by Sean Hogan. http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002142.html - installer: Fix the way to specify pgpoolAdmin version in redhat/rpm_installer/getsources.sh. (Yugo Nagata) - Makefile: Replace pg_config by $(PG_CONFIG) in Makefiles so it can be overridden at build time when compiling for different PG major versions. (Tatsuo Ishii) Patch contributed by Christoph Berg ([pgpool-general: 2127]). http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002156.html - watchdog: Fix a warning/error when compiling with -Werror=format-security. (Tatsuo Ishii) Patch contributed by Christoph Berg ([pgpool-general: 2127]). http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002156.html - configure: Remove -lcompat because it confuses FreeBSD per bug#15. (Tatsuo Ishii) http://www.pgpool.net/mantisbt/view.php?id=15 - Fix segfault when pgpool.conf does not set log_standby_delay. (Tatsuo Ishii) This is caused by wrong initialization for log_standby_delay in pool_config.l. Per bug#74. http://www.pgpool.net/mantisbt/view.php?id=74 - doc: Modify descriptions about restrictions of parallel mode Muliple rows INSERT using VALUES are not supported in parallel mode. (Yugo Nagata) - Avoid calling find_primary_node_repeatedly() when standby node goes down. (Tatsuo Ishii) This will reduce the time to failover. Per bug #75, patch modified by Tatsuo Ishii. http://www.pgpool.net/mantisbt/view.php?id=75 - Fix data inconsistency problem with native replication mode + extended protocol case. (Tatsuo Ishii) It is reported that concurrent INSERT using JDBC driver causes data difference among database node. This only happens following conditions are all met: 1) Native replication mode 2) Extended protocol used 3) The portal created by parse message is reused by bind message 4) autocommit is on 5) SERIAL (sequence) is used Pgpool-II's parse message function knows it has to lock the target table when INSERT (plus #5) is issued by clients. Unfortunately bind message function did not know it. Once parse/bind/execute finishes, pgpool releases the lock obtained by parse because of #4. JDBC wants to reuse the portal and starts the cycle from bind message, which does not obtain lock. As as result, lock-free INSERT are floating around which causes data inconsistency of course. The solution is, lock the table in bind phase. For this bind needs to issue LOCK in extended protocol. This was a little bit hard because the module (do_command()) to issue internal SQL command (other than SELECT) does not support extended protocol. To solve the problem do_query() is modified so that it accepts other than SELECT because it already accepts extended protocol. The modification is minimum and is only tested for the case called from insert_lock(). I do not recommend to replace every occurrence of do_command () with do_query() at this point. BTW the reason why the bug is not reported is, most users uses JDBC with auto commit = off. In this case, the lock obtained by parse persists until user explicitly issues commit or rollback. Per bug report by Steve Kuekes in [pgpool-general: 2142]. http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002171.html - Fix memory allocation size bug in the code path of query cache. (Tatsuo Ishii) In execute() memory allocation size is too small incertain case. No bug has been reported so far, but certainly this is a bug. - Fix occasional segfault in query cache + extended protocol. (Tatsuo Ishii) When the query is not "cache safe", bind_msg->param_offset was not set in Bind(). However, Execute() unconditionally uses bind_msg->param_offset to convert bind parameters to string so that they can be added to the cached query string and it causes segfault because bind_msg->param_offset is a garbage. Also logic bug to calculate bind_msg->param_offset is corrected. Per bug#76. http://www.pgpool.net/mantisbt/view.php?id=76 - Avoid to run out free query cache hash index entry. (Tatsuo Ishii) If hash index entries are run out, pgpool-II cannot reuse old cache entry because pgpool-II always expects there's at least one empty hash index entry. To mitigate the problem, if free hash index entries are run out, look for victim cache block to reuse the hash index entry. Per bug #70. http://www.pgpool.net/mantisbt/view.php?id=70 - installer: checkEnv() didn't do anything and always returned 0. (Nozomi Anzai) - Fix inappropriate shared memory allocation size for clock hand. (Tatsuo Ishii) The memory for clock hand was allocated as sizeof(pool_fsmm_clock_hand)) which is 8 bytes long because the variable is declared as: static int *pool_fsmm_clock_hand; This is plain wrong. The memory size actually needed is only 4 bytes, which is sizeof(*pool_fsmm_clock_hand)). In other word, the bug allocated unnecessary 4 bytes, which did nothing bd for the execution of program. But a bug is a bug. Per covery report "1111476 Wrong sizeof argument" - Fix "show pool_status" always shows memqcache_auto_cache_invalidation as 0. (Tatsuo Ishii) Per bug #80. http://www.pgpool.net/mantisbt/view.php?id=80 - Fix error message in read_password_packet(). (Tatsuo Ishii) - watchdog: Fix to pass big parameter by pointer instead of by value at some function. (Yugo Nagata) - Fix memory leak when SSL is requested. (Tatsuo Ishii) When SSL is requested, pgpool child retries to read start up packet. However it does not free the memory for previous start up packet. Per Coverity report "1111443 Resource". - Fix memory leak when do_query() fails in timestamp rewriting. (Tatsuo Ishii) For this purpose free_select_result() is changed to accept NULL argument. Per Coverity report "1111454, 1111455 Resource leak". - Fix load balance bug in replication mode. (Tatsuo Ishii) When load_balance_mode = off, SELECTs including writing function should be sent to all the DB nodes. Per [pgpool-general: 2221]. http://www.sraoss.jp/pipermail/pgpool-general/2013-October/002250.html The bug was introduced in: http://git.postgresql.org/gitweb/?p=pgpool2.git;a=commitdiff;h=1ac45a28258074ea4d9d902aca016f970d31f311 (3.3.1). - watchdog: Fix to put null character at end of ping result string used in watchdog. (Yugo Nagata) - Fix target node selection logic when "DEALLOCATE portal|statement". (Tatsuo Ishii) When "DEALLOCATE portal|statement" is used and last prepared statement or portal was not found, target node selection map is not set. Probably this is not actually harmful because prepared statement or portal was not found is an error case. The bug was there since day 0. Per Coverity report "1111491 Structurally dead code". - Fix range check bug of MAX_NUM_BACKENDS in corner case. (Tatsuo Ishii) MAX_NUM_BACKENDS is the allowed max number of DB nodes (128, at this point). In reality, probably no one ever tried more than 128 DB nodes and that's the reason why nobody noticed. Per Coverity report "1111429, 1111430 and 1111431 Out-of-bounds write". - Do not set/unset fronted connection info for dead backend. (Tatsuo Ishii) Per bug #82. http://www.pgpool.net/mantisbt/view.php?id=82 - Fix that the script forgets to allow public access to pgpool_catalog. (Tatsuo Ishii) The bug prevents inserting data into user tables if pgpool_catalog is created in native replication mode. The bug was there from day 1. I wonder why nobody noticed until today. Per [pgpool-general-jp: 1229]. http://www.sraoss.jp/pipermail/pgpool-general-jp/2013-November/001228.html - doc: Add description that it is recommended to specify multiple servers to trusted_servers. (Yugo Nagata) - Fix uninitialized variable in error case in pool_do_auth(). (Tatsuo Ishii) If there's no valid backend, pgpool will return garbage pid to frontend in auth phase. Actually because no backend is available, frontend will be disconnected later on. So this is harmless. Per Coverity report "1127331 Uninitialized scalar variable". - Fix to add node id range check when issue an error message using node id. (Tatsuo Ishii) Per Coverity report #1111433 "Out-of-bounds read". - Fix buffer overrun bug and resource leak bug of parse_copy_data(). (Tatsuo Ishii) Per Coverity report 1111427 "Out-of-bounds write" and 1111453 "Resource leak". - Fix possible segfault in CopyDataRaws(). (Tatsuo Ishii) Coverity pointed out that if pool_get_id() returns an error, VALID_BACKEND will access out of array. Per Coverity report 1111413 "Memory - illegal accesses". - Fix query cache is enabled and protocol version = 2 case. (Tatsuo Ishii) When the protocol version = 2, we assume that session state is "idle". This is not feasible but no way. I recommend to not use query cache with protocol 2. Per Coverity report 1111488 "Uninitialized scalar variable". - Fix strftime() usage in pool_pools(). (Tatsuo Ishii) The buffer is not large enough as expected by the second parameter. This is not harmless because the format string will not produce longer result string than the buffer. Per Coverity report 1111426 "Out-of-bounds access". - RPM: Improved to specify the versions of pgool-II and PostgreSQL in the. spec file. (Nozomi Anzai) - Fix resource leak in make_persistent_db_connection. (Tatsio Ishii) For this pupose, new static function free_persisten_db_connection_memory is added. Per Coverity report #1111468. - Fix a bug that connection check to trusted servers fails when the RTT is very short. (Yugo Nagata) - Fix several small bug fixes detected by Coverity. (Tatsuo Ishii) =============================================================================== 3.3.1 (tokakiboshi) 2013/09/06 * Version 3.3.1 This is a bugfix release against pgpool-II 3.3.0. __________________________________________________________________ * Bug fixes - Fix to add the regression test suite for making tar boll (Tatsuo Ishii) This isn't included in the tar ball of 3.3.0 release. - Fix watchdog test script in regression test (Tatsuo Ishii) - Fix a memory overrun bug (Tatsuo Ishii) This problem is pointed out in [pgpool-general: 1956] by Sean Hogan. [pgpool-general: 1956] memory overrun bug? http://www.sraoss.jp/pipermail/pgpool-general/2013-July/001984.html - Fix a compile error (Yugo Nagata) - Fix child process termination with sig abort when memory query cache is enabled (Tatsuo Ishii) This is due to double free bug that occurs when multiple bind/execute messages come after a parse message. When a parse messages comes, query context is created along with temp cache. The pointer to the temp cache is added to the temp cache array when the query executed. Subsequent bind messages uses the same temp cache pointer. This is the source of double free bug when the cache array discarded. This is reported in Bug track #68 by harukat. #68: child process termination with sigabort when memory_cache_enabled = on http://www.pgpool.net/mantisbt/view.php?id=68 - Add some test cases to regression test (Tatsuo Ishii) - Fix typos of the document in installation of pgpool-recovery section. (Tatsuo Ishii) - Fix a typo of log message (Yugo Nagata) - Fix typos of the japanese document (Yugo Nagata) - Fix to send SELECT to master node only when load_balance_mode is off in replication mode (Tatsuo Ishii) When load_balance_mode is off in replication mode, SELECT queries should be sent to master node only, rather than sent to all nodes in an explicit transaction. This problem is reported in [pgpool-general: 2038] by Rypl Lukas. [pgpool-general: 2038] SELECT sent to both nodes in replication mode http://www.sraoss.jp/pipermail/pgpool-general/2013-August/002066.html - Fix pgpool_setup to work in standalone. (Tatsuo Ishii) =============================================================================== 3.3.0 (tokakiboshi) 2013/07/30 * Version 3.3.0 This is the first version of pgpool-II 3.3 series. That is, a "major version up" from 3.2 series. __________________________________________________________________ * Incompatible changes - All the follwing are about watchdog. See "New features" section below for details of these changes. - Default monitoring method was changed from query mode to heartbeat mode. - Failover/failback commands are executed in only one pgpool-II. - In default, all the query caches on shared memory are cleared when standby pgpool-II escalates to active. - Database name, user name, and password used for monitring other pgpool-II by query are specified by dedicated parameters. Previously, template1, recovery_user, and recovery_password are used. __________________________________________________________________ * New features ** Watchdog - Add a new monitring method using heartbeat signal of UDP packet in lifecheck. (Yugo Nagata) You can select a method from either of two methods, "heartbeat" mode or "query" mode. The heartbeat mode is the new method. In this mode, watchdog monitors other pgpool-II process by using heartbeat signal. Watchdog receives heartbeat signals sent by other pgpool-II periodically. If there are no signal for a certain period, watchdog regards it as failure of the pgpool-II. For redundancy you can use multiple networks for heartbeat exchange between pgpool-IIs. This mode is default and recommended one. The query mode is the conventional method. In this mode, watchdog monitors pgpool-II's service rather than process. Watchdog sends queries to other pgpool-II and checks the response. Note that this method requires connections from other pgpool-IIs, so it would fail motoring if num_init_children isn't large enough. This mode is deprecated and left for backward compatibility. Add these new parameters: - wd_lifecheck_method - wd_heartbeat_port - wd_heartbeat_keepalive - wd_heartbeat_deadtime - heartbeat_destinationX - heartbeat_destination_portX - heartbeat_deviceX - Add interlocking mechanism of exclusive failover/failback command execution. (Yugo Nagata) When using multiple pgpool-IIs with watchdog enabled, failover commands (failover_command, failback_command, and follow_master_command) get executed only at one pgpool-II. Previously, these command got executed at all pgpool-IIs. - Add authentication mechanism for watchdog packet communication. (Yugo Nagata) Watchdog packets (include heartbeat signal) from pgpool-II with wrong authentication key are rejected. All pgpool-IIs must have the same key, which is specified wd_authkey parameter in pgpool.conf. pgpool-II with wrong authkey can't even start watchdog, because the startup packet is rejected by other pgpool-IIs. - Add clear_memqcache_on_escalation parameter. (Yugo Nagata) If this is on, all the query caches on shared memory are cleared when standby pgpool-II escalates to active. This is aimed to prevent the new active pgpool-II from using inconsistent query caches with the previous active. - Add wd_escalation_command parameter. (Yugo Nagata) This specifies command which is executed at escalation on the new active pgpool-II server. The timing is just before virtual IP is brought up. - Add parameters wd_lifecheck_dbname, wd_lifecheck_user, and wd_lifecheck_password. (Yugo Nagata) These parameters specify the database name, the user name, and password used in query mode lifecheck of watchdog . Previously, these are hard coded to use template1, recovery_user, and recovery_password. - When delegate_IP parameter is emply, viertual IP is neither brought up nor switched. (Yugo Nagata) This allows multi-master like configuration without virtual IP. - Add pcp_watchdog_info command (Yugo Nagata) This is PCP command for retrieving the watchdog status. ** Others - Import PostgreSQL 9.2 raw parser. (Nozomi Anzai, Tatsuo Ishii) - Add a tool called pgpool_setup to set up pgpool-II and PostgreSQL temporary installation in current directory for *testing* purpose. (Tatsuo Ishii) usage: pgpool_setup [-m r|s][-n num_clusters][--no-stop] -m s: create an installation as streaming replication mode. (the default) -m r: create an installation as native replication mode. -n num_clusters: create num_clusters PostgreSQL database cluster nodes -p base_port: specify base port. The first PostgreSQL node's port is base_port. Second PostgreSQL node's port is base_port+1 ... nth PostgreSQL node's port is base_port+n-1, pgpool port is base_port+n, pcp port is base_port+n+1. --no-stop: do not stop pgpool and PostgreSQL after the work - Support installation method using CREATE EXTENSION for pgpool-recovery and pgpool-regclass. (Tatsuo Ishii) Older installtion method is still preserved. Note: extension names are "pgpool_recovery" and "pgpool_regclass", not "pgpool-recovery" and "pgpool-regclass" because latters are not convenient in using CREATE EXTENSION command (requires double quotes). - Add a function "pgpool_pgctl()" which enebles to execute pg_ctl stop/restart/reload (except for start) by SQL. (Nozomi Anzai) $ psql sales -c "select pgpool_pgctl('reload', 'fast')"; pgpool_pgctl -------------- t (1 row) This function always ignores the actual result and returns 't', so the user can't know if pg_ctl succeeded or failed. To use this we have to set the custom variable for security which limits the users to execute pg_ctl who has the permission of data directory. - Add shell scripts to install pgpool-II and pgpoolAdmin by RPM. (Nozomi Anzai, Yugo Nagata) To make the installer package execute getsources.sh, and the directory named "work" will be created. And you rpmbuild each spec files in work/, put RPMs into work/installer and make tar ball of work/installer. The installer does not only install RPMs but also edit postgresql.conf, pgpool.conf, pg_hba.conf, recovery.conf and scripts for failover and online recovery. This assumes two-nodes configuration and the install script have to be executed in both nodes. - Add new parameter "search_primary_node_timeout". (Muhammad Usama, Tatsuo Ishii) The parameter specifies the maximum amount of time in seconds to search for a primary node when a failover scenario occurs. Patch contributed by Muhammad Usama. Japanese doc and slight editing of English doc by Tatsuo Ishii. - Chinese tutorials for memqcache and watchdog. (Bambo Huang) - Add regression test suit. (Tatsuo Ishii) __________________________________________________________________ * Bug fixes - Consider timeout waiting for compeletion of failback request in on line recovery. (Tatsuo Ishii) This will prevent the situation that recovery operation continues forever and we cannot even shutdown pgpool-II main process. This could happen especially while executing follow master command. - Fix a bug that %H of follow_master_command is not assigned correctly the new primary node in stream replication mode. (Tatsuo Ishii) - Fix not to execute escalation when the pgpool-II which is already active receives down notification from other pgpool-II. (Yugo Nagata) - Fix wd_create_send_socket() not to execute select() before connect(). (Yugo Nagata) How select() works on an unconnected socket is undefined, and differs between platform. On Linux, this returns 2 and it is eventually harmless. However, on Soraris, this returns 0 and it is indistinguishable from time timeout, so watchdog wouldn't work correctly. - Fix error when pgpool_regclass is not installed. (Tatsuo Ishii) The query used in pool_has_pgpool_regclass() fails if pgpool_regclass does not exist. The bug was introduced in 3.2.4. See [pgpool-general: 1722] for more details. [pgpool-general: 1722] [PgPool-II 3.2.4] pgpool_regclass now mandatory? http://www.sraoss.jp/pipermail/pgpool-general/2013-May/001749.html - Fix do_query() not to hang when PostgreSQL returns an error. (Tatsuo Ishii) The typical symptom is "I see SELECT is keep on running according to pg_stat_activity". To fix this pgpool-II just exits the process and kill the existig connection. This is not gentle but at this point I believe this is the best solution. - Fix possible deadlock during failover with watchdog enabled. (Yugo Nagata) This is reported in Bug track #54 by arshu arora http://www.pgpool.net/mantisbt/view.php?id=54 - Fix unnecessary degeneration caused by error on commit. (Tatsuo Ishii) In master slave mode, if master gets an error at commit, while other slaves are normal at commit, we don't need to degenrate any backend because it is likely that the "kind mismatch error" was caused by a deferred trigger. - Fix bug with do_query which causes hung in extended protocol. (Tatsuo Ishii) This problem could occur when insert lock is enabled and pgpool_catalog.insert_lock exists, See [pgpool-general: 1684] for more details. [pgpool-general: 1684] insert_lock hangs http://www.sraoss.jp/pipermail/pgpool-general/2013-May/001711.html - Fix possible failure of query cache invalidation for DML in transaction. (Tatsuo Ishii) CREATE TABLE t1(i INTEGER); CREATE TABLE t2(i INTEGER); SELECT * FROM t1; BEGIN; DELETE FROM t2 WHERE i = 0; INSERT INTO t1(i) VALUES(1); COMMIT; SELECT * FROM t1; At commit pgpool tries to delete cache for t2 but failes because there's no oid table entry for t2. Problem is, it fails to check oid table for t1. So cahce for t1 remains and the last SELECT incorrectly returns cached data. Fix is, continuing to check oid table entries. This is reported in Bug track #58 by wms http://www.pgpool.net/mantisbt/view.php?id=58 - Fix to register pgpool_regclass in pg_catalog schema (Tatsuo Ishii) This is necessary to deal with clients which restricts schema search path to pg_catalog only. Postgres_fdw is such a client. - Fix a hang of "pgpool -m f stop" (Tatsuo Ishii) This is caused by unmanaged pgpool children remaining. This could happen when multiple PostgreSQLs are going down or even when pgpool is started before PostgreSQL startup. - Fix a potential crash in pg_md5 command (Muhammad Usama) - Fix a segmentation fault that occurs when on memory query cache enabled and the query is issued in extended query mode and the result is too large (Tatsuo Ishii) This is reported in Bug track #63 by harukat. Analysis and a test case are also provided. #63 Child process was terminated by segmentation fault with memcached http://www.pgpool.net/mantisbt/view.php?id=63 - Fix a segmentation fault of a child process that occurs when a startup packet has no PostgreSQL user information (Yugo Nagata) You can reproduce it by $ psql -p 9999 -U '' If enable_pool_hba is on, a child process terminates by segmentation fault. Otherwise if enable_pool_hba is off, the error message is ERROR: pool_discard_cp: cannot get connection pool for user (null) database (null) In both cases, psql terminates with no message on frontend. In the fixed version, if PostgreSQL user is not specified in startup packet, the message as following is output to both log and frontend. This is the same behavior as PostgreSQL. FATAL: no PostgreSQL user name specified in startup packet - Fix memory allocation logic in extended query processing with on-memory query cache enabled (Tatsuo Ishii) When very long query string (> 1024 bytes) supplied in extended query with bind parameters, it fails to allocate enough memory. - Fix to verify the backend node number in pcp_recovery_node (Yugo Nagata) When an invalid number is used, null value is passed as an arguments of recovery script, and this causes a malfunction. In especially, rsync may delete unrelated files in basebackup scripts. - Fix a segmentation fault on main process that could occures after backend error detection (Tatsuo Ishii) This is reported in Bug track #62 by tuomas. #62 Slave network outage causes a segmentation fault on main process http://www.pgpool.net/mantisbt/view.php?id=62 - Fix bug with health check when used with child_life_time (Tatsuo Ishii) Failover could happen even if the backend was running fine. This problem is reported in [pgpool-general: 1892] by larisa sabban. [pgpool-general: 1892] Pgpool is unable to connect backend PostgreSQL http://www.sraoss.jp/pipermail/pgpool-general/2013-July/001920.html - Fix bug in parsing prepared statements with transaction handling in replication mode (Tatsuo Ishii) Parse() automatically starts a transaction for non SELECT query to keep consistency among nodes in replication mode. But this wasn't closed. If wrong query comes in, the transaction goes into an abort state but pgpool does not close the transaction. Thus next query causes error because the transaction is still in abort status. This problem was reported in [pgpool-general: 1877] by Sean Hogan. [pgpool-general: 1877] current transaction is aborted, commands ignored http://www.sraoss.jp/pipermail/pgpool-general/2013-July/001905.html __________________________________________________________________ * Enhancements - Add mention about "-D" option to the man page. (Tatsuo Ishii) - Fix to restart watchdog processes automatically when these exit abnormally. (Yugo Nagata) - Add more error checks and reportings to functions executing ping command with watchdog enabled. (Tatsuo Ishii) - Replace some unsafe functions, sprintf and strncpy, with more safe ones, snprintf and strlcpy respectively. (Yugo Nagata) - Replace "sticky bit" to "setuid bit" in log message, comments and funcation names. (Yugo Nagata) These words were used mistakenly and caused confusion. - Fix description on SSL in pool_hba.conf.sample. (Tatsuo Ishii) - Allow to load balancing in an explicit transaction in replication mode. (Tatsuo Ishii) The condition to allow the load balancing is as follows: 1) replicate_select is off 2) no writing functions are used 3) transaction isolation level is not serializable 4) no DML/DDL are issued in the transaction - Chinese manual is updated to the latest especially about watchdog. (Bambo Huang) - Add mention about "-D" option to the man page. (Tatsuo Ishii) - Move ssl_ca_cert and ssl_ca_cert_dir descriptions to the SSL section in the document (Yugo Nagata) - Add ssl_ca_cert and ssl_ca_cert_dir descriptions to the japanese document (Yugo Nagata) =============================================================================== 3.2 Series (2012/08/03 - ) =============================================================================== 3.2.10 (namameboshi) 2015/02/05 * Version 3.2.10 This is a bugfix release against pgpool-II 3.2.9. __________________________________________________________________ * Bug fixes - doc: Describe explicitly that the number of slave nodes is not necessarily 1 (Tatsuo Ishii) - Fix uninitialized variable (Tatsuo Ishii) Per Coverity 1234603. - doc: Fix missing release note entries in the previous release (Tatsuo Ishii) - Fix possible segmentation fault in query cache (Tatsuo Ishii) - Fix node id range check bug in trigger_failover_command(). (Tatsuo Ishii) The node id should be lower than NUM_BACKENDS. Probably harmless since callers never pass node ids greater or equal to NUM_BACKENDS. - Allow $_IP_$ having address prefix in if_up_cmd and if_down_cmd (Muhammad Usama) Previously, it is assumed that "$_IP_$" keyword will always have a white space at the end, and when "$_IP_$" has address prefix (e.g. $_IP_$/24), this prefix was ignored. Per bug #122. http://www.pgpool.net/mantisbt/view.php?id=122 - Allow "show pool_nodes" to print node ids having more than one digits (Muhammad Usama) Problem reported in [pgpool-general: 3374]. - Fix to disable debug mode by reloading config (Yugo Nagata) Per bug #114 http://www.pgpool.net/mantisbt/view.php?id=114 =============================================================================== 3.2.9 (namameboshi) 2014/09/05 * Version 3.2.9 This is a bugfix release against pgpool-II 3.2.8. __________________________________________________________________ * Bug fixes - Fix a typo of pgpool.spec (Yugo Nagata) - Fix bug that worker child process keeps failing when there's no primary backend (Tatsuo Ishii) Problem identified and fix contributed by Junegunn Choi. See [pgpool-hackers: 471] for more details. - Close listen socket when smart shutdown request is made (Tatsuo Ishii) When smart shutdown process starts, pgpool children still listen on the port and clients can send further connection requests which fail in the end. Which is not only waste of time, but also prevents a load balancer which sits in front of pgpool from realizing the pgpool is going down. Problem analyzed and patch provided by Junegunn Choi in [pgpool-hackers 474], and enhanced to take care not only inet domain socket but UNIX domain socket by Tatsuo Ishii. - doc: Add cautions that recovery commands are killed by statement_timeout of PostgreSQL. (Tatsuo Ishii) - doc: Remove old restriction description which is no longer true (Tatsuo Ishii) - Fix return type of text_to_lsn() function (Yugo Nagata) This caused compile warning. - Fix file descriptor leak when daemonize. (Tatsuo Ishii) Per Coverity 1111471. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111442. - Fix pgpool.init's long-standing bug of stop/restart failure (Yugo Nagata) In previous, pgpool.init uses killproc for stopping pgpool, but there are several problems. In the new version, "pgpool -m fast stop" is used in stop command. Original patch contributed by Ryan DeShone and modified by Yugo Nagata. See [pgpool-hackers: 239][pgpool-hackers: 512]. - Add missing include file sys/wait.h. (Tatsuo Ishii) Per bug #104. http://www.pgpool.net/mantisbt/view.php?id=104 - Disbale statement_timeout of PostgreSQL while executing online recovery (Tatsuo Ishii) Online recovery may take very long time and user may enable statement timeout. To prevent online recovery canceled by statement timeout, disable statement timeout in the connection used by online recovery. See [pgpool-general: 2919] for more details. - Remove unnecessary call to pool_shmem_exit() which removes semaphore when it shouldn't (Tatsuo Ishii) exit_handler checks if the process is parent or not. This is good. However, even if it is a child process, it calls pool_shmem_exit() which removes semaphore and shmem when it should not. It should be called only from parent process. Per bug #102. http://www.pgpool.net/mantisbt/view.php?id=102 - Fix reset query stuck problem. (Tatsuo Ishii) It is reported that reset query (DISCARD ALL etc.) occasionally does not finish and pgpool child remain waiting for reply from backend thus client cannot connect to pgpool. The cause of problem is not identified yet but if client suddenly closes connection to pgpool in the middle of query processing, backend may not accept the reset queries because they are not ready for query. The fix is, if frontend closes connection in unexpected way, query process loop immediately returns with new state: POOL_END_WITH_FRONTEND_ERROR and pgpool closes connection to PostgreSQL then goes back to new connection request waiting loop. Also, pgpool closes connections to backend when client_idle_limit is set and the idle limit. Per bug #107. http://www.pgpool.net/mantisbt/view.php?id=107 =============================================================================== 3.2.8 (namameboshi) 2014/03/24 * Version 3.2.8 This is a bugfix release against pgpool-II 3.2.7. __________________________________________________________________ * Bug fixes - doc: Add mention about "listen queue" and how to increase the "backlog" in the num_init_children section. (Tatsuo Ishii) - Fix bug that watchdog status goes down even when only one of trusted servers get unpingable (Yugo Nagata) - Fix bad performance of unlogged tables detection code. (Tatsuo Ishii) Pointed out at [pgpool-hackers: 435][[pgpool-general:2325]. - Fix primary node detection logic. (Tatsuo Ishii) There's a possibility that primary node is not detected. This happens in following situation. node 0: primary, node 1: standby. Node 0 goes down. Health checking detects the fact but local status is not updated yet. Primary node finding (find_primary_node) runs. Node 0's status is yet healthy. Because find_primary_node fails to connect to node 0, it immediately returns -1 and fails to find that fact that node 1 is now primary. Fix is just continuing to look for primary node when fails to connect to a node. Per [pgpool-general: 2409]. - Fix jdbc DML fails when operated in raw mode and auto commit is off. (Tatsuo Ishii) This is reported in bug #92. - Fix query cache bug with extended protocol. (Tatsuo Ishii) SELECT retrieves outdated cache even DML was executed in an expilicit transaction. The bug had been there since in memory query cache was born. Per [pgpool-general-jp: 1252]. - Add missing entries "relcache_size" and "check_temp_table" to show pool_status (Tatsuo Ishii) - Avoid to send queries to unrelated nodes in streaming replication mode. (Tatsuo Ishii) Pgpool-II sends certain queries, such as BEGIN, END and SET commands to all of DB nodes. However in streaming replication mode, only primary node and at most one standby node are only concerned (if primacy node is selected as the load balance node, only 1 node is concerned). See [pgpool-hackers: 464] for more details. - Fix possible buffer overrun problem and memory leak. (Tatsuo Ishii) Per Coverity 1111465 and 1111482. - Fix possible buffer overrun. (Tatsuo Ishii) Replace strcpy() with strlcpy(). Per Coverity report 1111478, 1111480, 1111481. - Fix freeing NULL. (Tatsuo Ishii) Per Coverity 1111384. - Do not enable query cache on materialized views. (Tatsuo Ishii) Per bug #95. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111446. - Fix a segmentation fault in parallel mode with system_db_hostname is empty (Yugo Nagata) - Fix to output debug messages in processing pgpool.conf with -d option (Yugo Nagata) Previously, some debug messages in pool_get_config() ware not output even when -d option was used. - Fix JDBC exception of prepared statement including now() in replication mode (Yugo Nagata) With JDBC, when a prepared statement is executed more than PrepareThreshold times, the statement is named and Describe message is sent after Parse. With named statement, pgpool rewrite now() to parameter in replication mode. Hence, rewritten query has additonal parameter than original. In this case, ParameterDescription message sent to frontend (response of Describe) should include OIDs of the same number os original query's parameters. Otherwize, JDBC throws ArrayIndexoutOfBoundsException. This is reported in [pgpool-general-jp: 1192]. - Fix backend error of prepared statement about table which has column whoes default value is now() in replication mode (Yugo Nagata) When pgpool parses a named prepared statement with default now(), timestamps are replaced to additional parameters. So, Bind message also should inclued additional parameter format codes. However, when the number of original parameter was one, pgpool didn't handle this. This caused a error like "incorrect binary data format in bind parameter 2". - doc: Add description about parallel mode doesn't support PREPARE (Yugo Nagata) Per bug #93 =============================================================================== 3.2.7 (tokakiboshi) 2013/12/06 * Version 3.2.7 This is a bugfix release against pgpool-II 3.2.6. __________________________________________________________________ * Bug fixes - Fix incorrect time stamp rewriting in replication mode for certain time zones. (Tatsuo Ishii) Time stamp rewriting calls "SELECT now()" to get current time. Unfortunately the buffer for the current time is too small for certain time zones such as "02:30". Note that non-30-minutes-time-zone such as "0900" does not reveal the problem. This explains why we haven't the bug report until today. Bug reported in [pgpool-general: 2113] and fix provided by Sean Hogan. http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002142.html - Makefile: Replace pg_config by $(PG_CONFIG) in Makefiles so it can be overridden at build time when compiling for different PG major versions. (Tatsuo Ishii) Patch contributed by Christoph Berg ([pgpool-general: 2127]). http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002156.html - Remove -lcompat because it confuses FreeBSD per bug#15. (Tatsuo Ishii http://www.pgpool.net/mantisbt/view.php?id=15 - Fix segfault when pgpool.conf does not set log_standby_delay. (Tatsuo Ishii) This is caused by wrong initialization for log_standby_delay in pool_config.l. Per bug#74. http://www.pgpool.net/mantisbt/view.php?id=74 - doc: Modify descriptions about restrictions of parallel mode Muliple rows INSERT using VALUES are not supported in parallel mode. (Yugo Nagata) - Avoid calling find_primary_node_repeatedly() when standby node goes down. (Tatsuo Ishii) This will reduce the time to failover. Per bug #75, patch modified by Tatsuo Ishii. http://www.pgpool.net/mantisbt/view.php?id=75 - Fix data inconsistency problem with native replication mode + extended protocol case. (Tatsuo Ishii) It is reported that concurrent INSERT using JDBC driver causes data difference among database node. This only happens following conditions are all met: 1) Native replication mode 2) Extended protocol used 3) The portal created by parse message is reused by bind message 4) autocommit is on 5) SERIAL (sequence) is used Pgpool-II's parse message function knows it has to lock the target table when INSERT (plus #5) is issued by clients. Unfortunately bind message function did not know it. Once parse/bind/execute finishes, pgpool releases the lock obtained by parse because of #4. JDBC wants to reuse the portal and starts the cycle from bind message, which does not obtain lock. As as result, lock-free INSERT are floating around which causes data inconsistency of course. The solution is, lock the table in bind phase. For this bind needs to issue LOCK in extended protocol. This was a little bit hard because the module (do_command()) to issue internal SQL command (other than SELECT) does not support extended protocol. To solve the problem do_query() is modified so that it accepts other than SELECT because it already accepts extended protocol. The modification is minimum and is only tested for the case called from insert_lock(). I do not recommend to replace every occurrence of do_command () with do_query() at this point. BTW the reason why the bug is not reported is, most users uses JDBC with auto commit = off. In this case, the lock obtained by parse persists until user explicitly issues commit or rollback. Per bug report by Steve Kuekes in [pgpool-general: 2142]. http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002171.html - Fix memory allocation size bug in the code path of query cache. (Tatsuo Ishii) In execute() memory allocation size is too small incertain case. No bug has been reported so far, but certainly this is a bug. - Fix occasional segfault in query cache + extended protocol. (Tatsuo Ishii) When the query is not "cache safe", bind_msg->param_offset was not set in Bind(). However, Execute() unconditionally uses bind_msg->param_offset to convert bind parameters to string so that they can be added to the cached query string and it causes segfault because bind_msg->param_offset is a garbage. Also logic bug to calculate bind_msg->param_offset is corrected. Per bug#76. http://www.pgpool.net/mantisbt/view.php?id=76 - Avoid to run out free query cache hash index entry. (Tatsuo Ishii) If hash index entries are run out, pgpool-II cannot reuse old cache entry because pgpool-II always expects there's at least one empty hash index entry. To mitigate the problem, if free hash index entries are run out, look for victim cache block to reuse the hash index entry. Per bug #70. http://www.pgpool.net/mantisbt/view.php?id=70 - Fix inappropriate shared memory allocation size for clock hand. (Tatsuo Ishii) The memory for clock hand was allocated as sizeof(pool_fsmm_clock_hand)) which is 8 bytes long because the variable is declared as: static int *pool_fsmm_clock_hand; This is plain wrong. The memory size actually needed is only 4 bytes, which is sizeof(*pool_fsmm_clock_hand)). In other word, the bug allocated unnecessary 4 bytes, which did nothing bd for the execution of program. But a bug is a bug. Per covery report "1111476 Wrong sizeof argument" - Fix "show pool_status" always shows memqcache_auto_cache_invalidation as 0. (Tatsuo Ishii) Per bug #80. http://www.pgpool.net/mantisbt/view.php?id=80 - Fix error message in read_password_packet(). (Tatsuo Ishii) - watchdog: Fix to pass big parameter by pointer instead of by value at some function. (Yugo Nagata) - Fix memory leak when SSL is requested. (Tatsuo Ishii) When SSL is requested, pgpool child retries to read start up packet. However it does not free the memory for previous start up packet. Per Coverity report "1111443 Resource". - Fix memory leak when do_query() fails in timestamp rewriting. (Tatsuo Ishii) For this purpose free_select_result() is changed to accept NULL argument. Per Coverity report "1111454, 1111455 Resource leak". - watchdog: Fix to put null character at end of ping result string used in watchdog. (Yugo Nagata) - Fix target node selection logic when "DEALLOCATE portal|statement". (Tatsuo Ishii) When "DEALLOCATE portal|statement" is used and last prepared statement or portal was not found, target node selection map is not set. Probably this is not actually harmful because prepared statement or portal was not found is an error case. The bug was there since day 0. Per Coverity report "1111491 Structurally dead code". - Fix range check bug of MAX_NUM_BACKENDS in corner case. (Tatsuo Ishii) MAX_NUM_BACKENDS is the allowed max number of DB nodes (128, at this point). In reality, probably no one ever tried more than 128 DB nodes and that's the reason why nobody noticed. Per Coverity report "1111429, 1111430 and 1111431 Out-of-bounds write". - Do not set/unset fronted connection info for dead backend. (Tatsuo Ishii) Per bug #82. http://www.pgpool.net/mantisbt/view.php?id=82 - Fix that the script forgets to allow public access to pgpool_catalog. (Tatsuo Ishii) The bug prevents inserting data into user tables if pgpool_catalog is created in native replication mode. The bug was there from day 1. I wonder why nobody noticed until today. Per [pgpool-general-jp: 1229]. http://www.sraoss.jp/pipermail/pgpool-general-jp/2013-November/001228.html - Fix uninitialized variable in error case in pool_do_auth(). (Tatsuo Ishii) If there's no valid backend, pgpool will return garbage pid to frontend in auth phase. Actually because no backend is available, frontend will be disconnected later on. So this is not harmless. Per Coverity report "1127331 Uninitialized scalar variable". - Fix to add node id range check when issue an error message using node id. (Tatsuo Ishii) Per Coverity report #1111433 "Out-of-bounds read". - Fix buffer overrun bug and resource leak bug of parse_copy_data(). (Tatsuo Ishii) Per Coverity report 1111427 "Out-of-bounds write" and 1111453 "Resource leak". - Fix possible segfault in CopyDataRaws(). (Tatsuo Ishii) Coverity pointed out that if pool_get_id() returns an error, VALID_BACKEND will access out of array. Per Coverity report 1111413 "Memory - illegal accesses". - Fix query cache is enabled and protocol version = 2 case. (Tatsuo Ishii) When the protocol version = 2, we assume that session state is "idle". This is not feasible but no way. I recommend to not use query cache with protocol 2. Per Coverity report 1111488 "Uninitialized scalar variable". - Fix strftime() usage in pool_pools(). (Tatsuo Ishii) The buffer is not large enough as expected by the second parameter. This is not harmless because the format string will not produce longer result string than the buffer. Per Coverity report 1111426 "Out-of-bounds access". - Fix resource leak in make_persistent_db_connection. (Tatsio Ishii) For this pupose, new static function free_persisten_db_connection_memory is added. Per Coverity report #1111468. - Fix a bug that connection check to trusted servers fails when the RTT is very short. (Yugo Nagata) - Fix several small bug fixes detected by Coverity. (Tatsuo Ishii) =============================================================================== 3.2.6 (namameboshi) 2013/09/06 * Version 3.2.6 This is a bugfix release against pgpool-II 3.2.5. __________________________________________________________________ * Bug fixes - Fix a segmentation fault on main process that could occures after backend error detection (Tatsuo Ishii) This is reported in Bug track #62 by tuomas. #62 Slave network outage causes a segmentation fault on main process http://www.pgpool.net/mantisbt/view.php?id=62 - Fix a bug with health check when used with child_life_time (Tatsuo Ishii) Failover could happen even if the backend was running fine. This problem is reported in [pgpool-general: 1892] by larisa sabban. [pgpool-general: 1892] Pgpool is unable to connect backend PostgreSQL http://www.sraoss.jp/pipermail/pgpool-general/2013-July/001920.html - Fix "Deploying pgpool-II" section in the document (Tatsuo Ishii) Update descriptions about watchdog use. - Fix a mistake in ssh command of doc/basebackup.sh (Tatsuo Ishii) - Fix a bug in parsing prepared statements with transaction handling in replication mode (Tatsuo Ishii) Parse() automatically starts a transaction for non SELECT query to keep consistency among nodes in replication mode. But this wasn't closed. If wrong query comes in, the transaction goes into an abort state but pgpool does not close the transaction. Thus next query causes error because the transaction is still in abort status. This problem was reported in [pgpool-general: 1877] by Sean Hogan. [pgpool-general: 1877] current transaction is aborted, commands ignored http://www.sraoss.jp/pipermail/pgpool-general/2013-July/001905.html - Fix child process termination with sig abort when memory query cache is enabled (Tatsuo Ishii) This is due to double free bug that occurs when multiple bind/execute messages come after a parse message. When a parse messages comes, query context is created along with temp cache. The pointer to the temp cache is added to the temp cache array when the query executed. Subsequent bind messages uses the same temp cache pointer. This is the source of double free bug when the cache array discarded. This is reported in Bug track #68 by harukat. #68: child process termination with sigabort when memory_cache_enabled = on http://www.pgpool.net/mantisbt/view.php?id=68 - Fix typos of the japanese document (Yugo Nagata) =============================================================================== 3.2.5 (namameboshi) 2013/07/10 * Version 3.2.5 This is a bugfix release against pgpool-II 3.2.4. __________________________________________________________________ * Bug fixes - Add mention about "-D" option to the man page. (Tatsuo Ishii) - Consider timeout waiting for compeletion of failback request in on line recovery (Tatsuo Ishii) This will prevent the situation that recovery operation continues forever and we cannot even shutdown pgpool-II main process. This could happen especially while executing follow master command. - Fix a bug that %H of follow_master_command is not assigned correctly the new primary node in stream replication mode (Tatsuo Ishii) - Fix wd_create_send_socket() to not execute select() before connect() (Yugo Nagata) How select() works on an unconnected socket is undefined, and differs between platform. On Linux, this returns 2 and it is eventually harmless. However, on Soraris, this returns 0 and it is indistinguishable from time timeout, so watchdog wouldn't work correctly. - Fix error when pgpool_regclass is not installed (Tatsuo Ishii) The query used in pool_has_pgpool_regclass() fails if pgpool_regclass does not exist. The bug was introduced in 3.2.4. See [pgpool-general: 1722] for more details. [pgpool-general: 1722] [PgPool-II 3.2.4] pgpool_regclass now mandatory? http://www.sraoss.jp/pipermail/pgpool-general/2013-May/001749.html - Fix do_query() to not hang when PostgreSQL returns an error (Tatsuo Ishii) The typical symptom is "I see SELECT is keep on running according to pg_stat_activity". To fix this pgpool-II just exits the process and kill the existig connection. This is not gentle but at this point I believe this is the best solution. - Fix possible deadlock during failover with watchdog enabled (Yugo Nagata) This is reported in Bug track #54 by arshu arora #54 pgpool-II semaphore lock problem http://www.pgpool.net/mantisbt/view.php?id=54 - Fix bug with do_query which causes hung in extended protocol (Tatsuo Ishii) This problem could occur when insert lock is enabled and pgpool_catalog.insert_lock exists, See [pgpool-general: 1684] for more details. [pgpool-general: 1684] insert_lock hangs http://www.sraoss.jp/pipermail/pgpool-general/2013-May/001711.html - Fix unnecessary degeneration caused by error on commit (Tatsuo Ishii) In master slave mode, if master gets an error at commit, while other slaves are normal at commit, we don't need to degenrate any backend because it is likely that the "kind mismatch error" was caused by a deferred trigger. - Fix possible failure of query cache invalidation for DML in transaction (Tatsuo Ishii) CREATE TABLE t1(i INTEGER); CREATE TABLE t2(i INTEGER); SELECT * FROM t1; BEGIN; DELETE FROM t2 WHERE i = 0; INSERT INTO t1(i) VALUES(1); COMMIT; SELECT * FROM t1; At commit pgpool tries to delete cache for t2 but failes because there's no oid table entry for t2. Problem is, it fails to check oid table for t1. So cahce for t1 remains and the last SELECT incorrectly returns cached data. Fix is, continuing to check oid table entries. This is reported in Bug track #58 by wms #58 query cache invalidation does not fire for multiple DML in transaction http://www.pgpool.net/mantisbt/view.php?id=58 - Fix to register pgpool_regclass in pg_catalog schema (Tatsuo Ishii) This is necessary to deal with clients which restricts schema search path to pg_catalog only. Postgres_fdw is such a client. - Fix a potential crash in pg_md5 command (Muhammad Usama) - Fix a segmentation fault that occurs when on memory query cache enabled and the query is issued in extended query mode and the result is too large (Tatsuo Ishii) This is reported in Bug track #63 by harukat. Analysis and a test case are also provided. #63 Child process was terminated by segmentation fault with memcached http://www.pgpool.net/mantisbt/view.php?id=63 - Fix a segmentation fault of a child process that occurs when a startup packet has no PostgreSQL user information (Yugo Nagata) You can reproduce it by $ psql -p 9999 -U '' If enable_pool_hba is on, a child process terminates by segmentation fault. Otherwise if enable_pool_hba is off, the error message is ERROR: pool_discard_cp: cannot get connection pool for user (null) database (null) In both cases, psql terminates with no message on frontend. In the fixed version, if PostgreSQL user is not specified in startup packet, the message as following is output to both log and frontend. This is the same behavior as PostgreSQL. FATAL: no PostgreSQL user name specified in startup packet - Fix memory allocation logic in extended query processing with on-memory query cache enabled (Tatsuo Ishii) When very long query string (> 1024 bytes) supplied in extended query with bind parameters, it fails to allocate enough memory. - Move ssl_ca_cert and ssl_ca_cert_dir descriptions to the SSL section (Yugo Nagata) - Add ssl_ca_cert and ssl_ca_cert_dir descriptions to the japanese document (Yugo Nagata) - Fix to verify the backend node number in pcp_recovery_node (Yugo Nagata) When an invalid number is used, null value is passed as an arguments of recovery script, and this causes a malfunction. In especially, rsync may delete unrelated files in basebackup scripts. =============================================================================== 3.2.4 (namameboshi) 2013/04/26 * Version 3.2.4 This is a bugfix release against pgpool-II 3.2.3. __________________________________________________________________ * Bug fixes - Fix connect_inet_domain_socket_by_port() to set more appropriate value for timeout parameter of select(2). (Tatsuo Ishii) Some platforms such as Solaris do not allow to specify too large microseconds timeout value (>=1000000). So divide the timeout value to seconds and microseconds. - Fix connect_inet_domain_socket_by_port() to not return as normal when interrupted by alarm. (Tatsuo Ishii) This confuses health checking because connect_inet_domain_socket_by_port() returns unsable fd. This makes detecting errors in health checking longer. See the following for more details: [pgpool-general: 1458] health check timeout in pgpool-II-3.2.3 http://www.pgpool.net/pipermail/pgpool-general/2013-March/001482.html - Fix long standing bug with timestamp rewriting code for processing extended protocol. (Tatsuo Ishii) Parse() allocate memory using palloc() while rewriting the parse message. Problem is, the rewritten message was kept in the data which is managed by pool_create_sent_message() etc. The function assumes that all the data is in session context memory. However, palloc() allocates memory in query context of course, and gets freeed later on when the query context disappears. And the function tries to free the memory as well, which causes various problems, including segfault and double free. To fix this, memory to store rewritten message is allocated using session context. The bug was there since pgpool-II 3.0 was born. Problem analysis and patch contributed by Naoya Anzai. [pgpoolgenera-jp: 1146]. (in Japanese) http://www.pgpool.net/pipermail/pgpool-general-jp/2013-March/001145.html - Fix bug with md5 auth long user name handling. (Tatsuo Ishii) If user name is longer than 32 bytes, md5 authentication doesn't work. Problem reported in [pgpool-general: 1526] by Thomas Martin. [pgpool-general: 1526] [pgPool-II 3.2.3] MD5 authentication and username longer than 32 characters. http://www.pgpool.net/pipermail/pgpool-general/2013-March/001551.html - Fix to calculate replication delay only if standby server is behind from the primay server. (Yugo Nagata) When the primary server is behind from standby server, negative value of delay is calculated and the value is assigned to unsigned variable. It causes a log message informing negative replication delay. And what is worse, it also causes SELECT queries to be sent to the primary in load balance even though there are no replication delay in fact. The problem is reported and analyzed by Saitoh Hidenori in [pgpool-genera-jp: 1145]. [pgpool-general-jp: 1145] (in Japanese) http://www.pgpool.net/pipermail/pgpool-general-jp/2013-March/001144.html - pgpool-recovery adopts PostgreSQL 9.3. (Tatsuo Ishii) Patch contributed by Asif Rehman. Slight editing by Tatsuo Ishii. [pgpool-hackers: 180] compile error in ppool-recovery http://www.pgpool.net/pipermail/pgpool-hackers/2013-April/000179.html - Fix pool_has_pgpool_regclass() to check execute privilege of pgpool_regclass(). (Tatsuo Ishii) Even though pgpool_regclass() exists, if pgpool cannot execute the function, the connection to backend hangs. You can reproduce the problem by just dropping the execute privilege from pgpool_regclass and do some insert in native replication mode. The problem is reported in bugtrack #53. #53 pgpool_regclas hangs all connections Date: 2013-04-04 13:35 Reporter: tmandke http://www.pgpool.net/mantisbt/view.php?id=53 - Fix error message mistakes in detect_postmaster_down_error(). (Tatsuo Ishii) For example, "LOG: detect_stop_postmaster_error: detect_error error" is fixed to "LOG: detect_postmaster_down_error: detect_error error", and so on. - Remove root user check when watchdog is enabled. (Tatsuo Ishii) Per discussion [pgpool-general: 1627] Re: watchdog root requirement. [pgpool-general: 1627] Re: watchdog root requirement. http://www.pgpool.net/pipermail/pgpool-general/2013-April/001654.html - Fix bug with on memory query cache in handling UPDATE/DELETE with table alias. (Tatsuo Ishii) If UPDATE/DELETE is with table alias (UPDATE t1 AS foo...) pgpool thinks the table name is "t1 AS foo" and fails to invalidate query cache. This is caused by _outRangeVar() called from nodeToString() which generates a query string from RangeVar node in raw parse tree. The solution is removing "AS foo" part from the output of the string. Reported in bugtrack #56. #56 UPDATE with alias does not discard cache Date: 2013-04-18 17:33 Reporter: harukat http://www.pgpool.net/mantisbt/view.php?id=56 =============================================================================== 3.2.3 (namameboshi) 2013/02/18 * Version 3.2.3 This is a bug fix release against pgpool-II 3.2.2. Main purpose of this release is to fix fatal problem with pgpool-II 3.2.2's health checking. If all of following conditions are met, pgpool main process disappeared and all client connections to pgpool-II hang forever when failover happens. And the only way to recover from it is, manualy killing the pgpool child process and restart pgpool-II. - health checking is enabled - connecting method to PostgreSQL is TCP/IP, not UNIX domain socket(i.e. "backend_hostnameN" is not empty string) __________________________________________________________________ * Bug fixes - Fix connect_inet_domain_socket_by_port() bug introduced in 3.2.2. (Tatsuo Ishii) When non blocking connect() reports EINPROGRESS or EALREADY, it calls select(2) to wait for read or write fd ready. However it mistakenly checks error condition using getsockopt(). It should be called when select() returns > 0, rather than 0. Because of this, connect_inet_domain_socket_by_port() could return succeeded fd even it actually failed. And what is worse, this health_check() mistakenly believes that backend is alive and tries to write to backend socket, which of course fails. This triggers to call notice_backend_error(), which sends SIGUSR1 signal to pgpool main's parent process. This will result in various weird things: for example, if you start pgpool from a shell, the signal kills the shell. If you start pgpool in background, pgpool's parent is the process #1. As long as you started pgpool as non root, it's ok. Even if you start pgpool as root, init just reopens /dev/initctl by receiving SIGUSR1. These all annoying bugs have been there since pgpool was born. The connect_inet_domain_socket_by_port() bug just reveals it. To fix this, I modified notice_backend_error and child_exit() so that it does nothing when called from pgpool main process itself to prevent pgpool from shooting itself in the foot. - Fix to show pool_passwd in "SHOW pool_status". (Yugo Nagata) - Fix a typo at configure's help in configure.in. (Yugo Nagata) =============================================================================== 3.2.2 (namameboshi) 2013/02/08 * Version 3.2.2 This is a bugfix release against pgpool-II 3.2.1. __________________________________________________________________ * Bug fixes - Fix compile errors on FreeBSD. (Tatsuo Ishii) - Fix pgpool does not recognize VIEWs other than in default schema, which is usually "public". (Tatsuo Ishii) This makes pgpool to create caches for such a VIEW's query results, which of course should not be allowed. Problem reported and patch provided by jgentsch in bug id #30. #30 pgpool 3.2.1 - views in schema other than public are caching Reporter: jgentsch Date: 2012-10-19 23:13 http://www.pgpool.net/mantisbt/view.php?id=30 - Fix race condition when using md5 authentication. (Tatsuo Ishii) The file descriptor to pool_passwd is opened in pgpool main and pgpool child inherits it. When concurrent connections try to authenticate md5 method, they call pool_get_passwd and seek the fd and cause random md5 auth failure because underlying fd is shared. Fix is, let individual pgpool child open the file by calling pool_reopen_passwd_file. Problem reported and analyzed by Jason Slagle in pgpool-general:1141. [pgpool-general: 1141] Possible race condition in pool_get_passwd From: Jason Slagle Date: Sun, 28 Oct 2012 01:12:52 -0400 http://www.sraoss.jp/pipermail/pgpool-general/2012-October/001160.html - Clarify load balance condition information in manual. (Tatsuo Ishii) - Fix segfault due to bug with query cache array handling. (Tatsuo Ishii) The cache arrary is used to keep temporary cache results in a transaction. If there are more than 128 SELECTs in a transaction, the module expands cache_arrary by using realloc. However it does not record the new pointer returned by realloc. So the module keeps on using the old pointer which is absoleted. This problem is reported in bug track #31 by jgentsch. #31 pgpool V3_2_STABLE - segfault in pool_memqcache.c:2529 Reporter:jgentsch Date: 2012-10-23 06:25 http://www.pgpool.net/mantisbt/view.php?id=31 - Fix hung up while repeating pcp_attach_node and pcp_detatch_node (Tatsuo Ishii) When node status is changed by pcp_attach_node and pcp_detatch_node, failover() sends SIGUSR1 to pcp_child process expecting it exits to refresh node status. In this situation lots of pgpool children exit and produce SIGCHLD as well. The SIGCHLD handler reaper() tries catch all SIGCHLD but sometimes it fails depending on the system load and timing. If SIGCHLD produced by pcp child is not caught, the process becomes zombie and never restarted. This problem is reported in bug track #32 (by oleg_myrk) etc. #32 PGPool hangs on pcp_attach/detach Reporter: oleg_myrk Date: 2012-10-24 00:01 http://www.pgpool.net/mantisbt/view.php?id=32 - Fix pool_send_severity_message() not to use uninitialized memory. (Tatsuo Ishii) It cause a segmentaion fault. Reported in Bug #33's attached valgrind output by dudee. #33 pgpool-II 3.2.1 segfault Reporter: dudee Date: 2012-10-30 19:16 http://www.pgpool.net/mantisbt/view.php?id=33 - Fix bug with query cache returning incorrect data in some cases when a persistent table and temp table have same name. (Tatsuo Ishii) Here is a sequence to trigger the bug: 1) CREATE TABLE t1(i int); -- create a persistent table 2) INSERT INTO t1 VALUES(1); 3) SELECT * FROM t1; -- query cache entry created 4) CREATE TEMP TABLE t1(i int); -- create a temp table 5) SELECT * FROM t1; -- query cache entry mistakenly created! Problem is #3 creates relcache entry for t1, and #5 incorrecly uses it and believes that temp table t1 is not a temp table. - Add a description about "-f" to help message. (Tatsuo Ishii) - Fix reaper() not to exit wait3() loop when catches pcp or worker child exit event. (Tatsuo Ishii) Otherwise reaper() mistakenly ignore some process exit event and make a risk of creating zombie process and forgetting to create new process. Problem reported and fix suggested by Goto in [pgpool-general-jp: 1123]. http://www.sraoss.jp/pipermail/pgpool-general-jp/2012-November/001122.html (in Japanese) - Fix a typo of configure help message. (Yugo Nagata) - Add wd_hostname to pool_process_reporting.c. (Yugo Nagata) Otherwise, wd_hostname is not contained in results of SHOW pool_status and cp_pool_status. - Fix connect_inet_domain_socket_by_port() to not error out when connect(2) returns EISCONN (Socket is already connected) error. (Tatsuo Ishii) This could happen with non blocking socket and should be treated as normal. Per bug track #29 (by spork) and pgpool-general 1218 (by Mikola Rose). #29 pgpool 3.2.1 cannot connect to db hosts Reporter: spork Date: 2012-10-18 15:03 http://www.pgpool.net/mantisbt/view.php?id=29 [pgpool-general: 1218] pgpool 3.2.1 - Health check failing to connect From: Mikola Rose Date: Tue, 4 Dec 2012 20:21:55 +0000 http://www.sraoss.jp/pipermail/pgpool-general/2012-December/001237.html - Fix health_check() to check the health check timer before retrying with template1 database. (Tatuo Ishii) Without this, the retry with node 0 always fails because health check timer may be already expired. - Fix pool_search_relcache() to use MASTER or MASTER_NODE_ID macro, rather than REAL_MASTER_NODE_ID. (Tatsuo Ishii) In case node 0 fail back in streaming replication mode, pgpool does not restart child process. So REAL_MASTER_NODE_ID looks for node 0 con info, which is not present until new connection to backend made. Thus referring to node con info results in segfault. MASTER or MASTER_NODE_ID are safe in this situation because they look at cached former master node id. - Fix long standing bug "portal not found" error when replication delay is too much in streaming replication mode. (Tatsuo Ishii) The bug had been there since the delay threshold was introduced. We changed destination DB node if delay threshold exceeds in bind, describe and execute. However, if parse sends to different node, bind, describe or execute will fail because no parsed statement or portal exists. Solution is, not to send to different parse node even if delay threshold is too much. - Fix pg_md5 to output "\n" after user inputs password. (Yugo Nagata) - Fix to print error message when the port number for watchdog is already used. (Yugo Nagata) This issue was reported by Will Ferguson in [pgpool-general: 1167]. [pgpool-general: 1167] Re: Watchdog error - wd_init: delegate_IP already exists From: Will Ferguson Date: Tue, 6 Nov 2012 13:03:36 +0000 http://www.sraoss.jp/pipermail/pgpool-general/2012-November/001186.html - Fix child_exit() to not call send_frontend_exits() if there's no connection pool. (Tatsuo Ishii) Otherwise, it segfaults because send_frontend_exits() referes to objects pointed to by pool_connection_pool. Per bug track #44 by tuomas. #44 pgpool went haywire after slave shutdown triggering master failover Reporter: tuomas Date: 2012-12-11 00:33 http://www.pgpool.net/mantisbt/view.php?id=44 - Fix bug that only tables in white_memqcache_table_list was cached when black_memqcache_table_list has any tables. (Yugo Nagata) - Fix read_startup_packet() to reset alarm and free StartupPacket when pool_read() returns 0 which means incorrect packet length. (Nozomi Anzai) Previously, authentication timeout occurs when connected by a program monitoring the pgpool port.It is reported in bug track #35. #35 Authentication is timeout Reporter: tuomas Date: 2012-11-20 11:54 http://www.pgpool.net/mantisbt/view.php?id=35 - Fix long standing bug with pool_open(). (Tatsuo Ishii) It initializes wrong buffer pointer. Actually this is harmless because the pointer is initialized by prior memset() call, though. - Log that failover is avoided because "fail_over_on_backend_error" is turned off. (Tatsuo Ishii) - Fix LISTEN/NOTIFY handling bugs. (Tatsuo Ishii) 1) In streaming replication mode: Session 1: LISTEN aaa; Session 2: NOTIFY aaa; Session 1: LISTEN aaa; --- hangs (If LISTEN and NOTIFY are issued in a same session, it works fine.) We assume that packets come from all nodes. However in streaming replication mode, notification message only comes from primary node and we should avoid reading from standby nodes. 2) In streaming replication mode: If primary node is not node 0, it hangs like #1 even if fix applies. This is because MASTER_NODE_ID macro (actually pool_virtual_master_db_node_id()) always returns REAL_MASTER_NODE_ID, which is node 0 (if it is alive). The function should return PRIMARY_NODE_ID in master/slave mode. 3) In replication mode, LISTEN/NOTIFY simply does not work. In the mode, NOTIFY is sent to all backends. However the order of arrival of 'Notification response' is not necessarily the master first and then slaves. So if it arrives slave first, we should try to read from master, rather than just discard it. Fixed in pool_process_query(). 4) In replication mode, if LISTEN and NOTIFY are issued in a same session, the session is disconnected because do_command() may receive other than 'N', 'E', 'S' and 'C'. The solution is, put 'A' packet on a stack and pop out when it is convenient. For this purpose, new functions pool_push(), pool_pop() and pool_stacklen() are added. This probmel is reported in but grack #45 by rpashin. #45 LISTEN/NOTIFY doesn't work if cluster contains more then 1 node in streaming replication mode Reporter: rpashin Date: 2012-12-12 00:09 http://www.pgpool.net/mantisbt/view.php?id=45 Considering the size of the patch, I do not back patch to 3.1 or before(so far, we have not heard any complaints on 3.1 or before). - Fix connect_inet_domain_socket_by_port() to call select(2) rather than error out when connect(2) returns EINPROGESS or EALREADY error. (Tatsuo Ishii) When using non-blocking socket, despite the errors like "Connection timed out", actually connection has been established. To solve the problem we should use select(2) to wait for connection establishing when connect(2) reports EINPROGRESS or EALREADY, instead of doing a retry tight loop. This problem is reported in bug track #46 by mcousin. #46 Watchdog failing to connect sometimes Reporter: mcousin Date: 2012-12-15 01:01 http://www.pgpool.net/mantisbt/view.php?id=46 - Add caution to increase num_init_children if watchdog enabled in manual. (Tatsuo Ishii) See [pgpool-general: 1330] for more details. [pgpool-general: 1330] WatchDog and pgool sudden stop working From: Tomas Halgas Date: Fri, 18 Jan 2013 14:47:23 +0100 http://www.sraoss.jp/pipermail/pgpool-general/2013-January/001350.html - Fix segmentation fault while pgpool-II stating up or fail over when watchdog is enabled. (Yugo Nagata) This is caused by wrong usage of pthread, namely pthread_detach and pthread_join are mixed together. Solution is to use pthread_join only if we need to get status of child thread. BTW, the problem could occu on moderately modern OS such as Fedora 17, but the reason why the problem is not observed on other OSs is, just we were lucky. Problem reported in [pgpool-general: 1179] by Lonni J Friedman. [pgpool-general: 1179] 3.2.1 segfaults at startup on Fedora17. From: Lonni J Friedman Date: Mon, 12 Nov 2012 15:58:29 -0800 http://www.sraoss.jp/pipermail/pgpool-general/2012-November/001198.html Patch provided by chads in bug track #48. pthread_detach is being used wrong; causes pgpool to segfault. Reporter: chads Date: 2013-01-16 05:44 http://www.pgpool.net/mantisbt/view.php?id=48 - Avoid split-brain situation reported in [pgpool-general: 1046] (Yugo Nagata) After all backend nodes are detached from pgpools and then some backend node are attached to these, multiple active pgpools could exist simultaneously, that is to say, split-brain situation occurs. In this fix, when once all backend DB nodes are detached from pgpool, the pgpool stays DOWN status until this is restarted. The pgpool in DOWN status cannot escalate to active (delegate IP holder), so split-brain situation is avoided. [pgpool-general: 1046] watchdog enabled delegate_IP on multiple nodes simultaneously From: Lonni J Friedman Date: Wed, 26 Sep 2012 09:05:09 -0700 http://www.sraoss.jp/pipermail/pgpool-general/2012-September/001064.html - Avoid a possible hang during the active pgpool exits. (Yugo Nagata) When exiting, the active pgpool brings down the virtual IP and then sends a packet to other pgpools. However, the packet sometimes is sent before the virtual IP is brought down completely. In this case the packet sender is set to this IP. When the IP is brought down before other pgpools respond, the active pgpool can not recieve the response, and hang up. In this fix, the active pgpool confirms that the virtual IP is brought down before sending the packet. - Modify descriptions of restrictions on watchdog. (Yugo Nagata) - Modify pgpool.conf.sample* and documents to correct information of whether a certain parameter change requires restart. (Yugo Nagata) - Add pool_passwd option to pgpool.conf.sample*, pool_process_reporting.c, and documents. (Yugo Nagata) Otherwise, wd_hostname is not contained in results of SHOW pool_status and pcp_pool_status. =============================================================================== 3.2.1 (namameboshi) 2012/10/12 * Version 3.2.1 This is a bugfix release against pgpool-II 3.2.0. __________________________________________________________________ * Bug fixes - Fix send_cached_messages(). (Tatsuo Ishii) Before it had 8192 bytes fix length buffer for each row data and if data exceeded 8192 bytes, it just crashed. To fix this, eliminate copying raw data which is passed as an argument to buffer and pass the pointer to send_message. - Fix that extended queries failed due to query cache. (Nozomi Anzai) - Fix read_startup_packet(). (Tatsuo Ishii) If packet length is lower than 0, it should have returned immediately. Otherwise it would cause memory allocation error later on. per pgpool-general:886. Also add canceling alarm. [pgpool-general: 886] read_startup_packet: out of memory From: Lonni J Friedman Date: Wed, 8 Aug 2012 10:18:15 -0700 http://www.sraoss.jp/pipermail/pgpool-general/2012-August/000896.html - Fix too watchdog process's aggressively kill other processes when pgpool shuts down. (Tatsuo Ishii) watchdog process calls kill(0,SIG) to kill all processes related to watchdog. Unfortunately this will kill not only watchdog related processes but parent pgpool and even httpd in case when pgpool was invoked from pgpoolAdmin because they are in the same process group. So for now, fix is removing call to the kill() and setpgid() because setpgid() does nothing useful. In the future, we should call setsid() to establish new process group in any case. - Fix query cache to regist such queries that start with "-- comment" or have comments more than one. (Nozomi Anzai) - Fix query cache to ignore multi statement. (Nozomi Anzai) In previous, queries like "SELECT 1;UPDATE..." were cached, too, but it was wrong. - Add a watchdog restriction to documents. (Yugo Nagata) - Add NOTICE message handling to s_do_auth(). (Tatsuo Ishii) Without this, health check responses false alarm and causes failover. per bug track: #25 s_do_auth doesn't handle NoticeResponse (N) message Date: 2012-08-28 03:57 Reporter: singh.gurjeet Date: http://www.pgpool.net/mantisbt/view.php?id=25 - Remove unnecessary/confusing debug log from s_do_auth().(Tatsuo Ishii) - Fix buffer overrun in Execute when memory cache enabled. (Tatsuo Ishii) If one of bind parameter < 0, it was possible to produce more than 2 byte string for "%02X" due to sign extention. Also use snprintf, rather than sprintf to prevent from possible buffer overrun in the future. - Fix long standing memory leak bug with free_select_result() since pgpool-II 2.3 was born in December 2009. (Tatsuo Ishii) Actually this bug only appears when operated in replication mode (triggered by timestamp rewriting process by coincidence). Per bug track #24: #24 Severe memory leak in an OLTP environment Date: 2012-08-28 03:43 Reporter: singh.gurjeet Date: http://www.pgpool.net/mantisbt/view.php?id=24 - Fix typo in cache_reporting(). (Tatsuo Ishii) - Fix inifinit loop in SSL mode. (Tatsuo Ishii) When there's pending data in SSL layer of frontend, pool_process_query() checks pending data in backend. If there's non, it loops again and checks frontend/backend receive buffer by using is_cache_empty(). Unfortunately it first checks pending data in SSL layer of frontend, thus goes to backend data and checks again (infinite loop). The solution is, if there's pending data in SSL layer of frontend and query is not in progress, call ProcessFrontendResponse() to process new request from frontend. - Fix is_system_catalog() to use pgpool_regclass if available. (Tatsuo Ishii) - Fix memory leak in pool_get_insert_table_name(). (Tatsuo Ishii) If session context's memory contex is used for nodeToString(), memory is not freed until session ends. See bug track #24 for more details. #24 Severe memory leak in an OLTP environment Date: 2012-08-28 03:43 Reporter: singh.gurjeet http://www.pgpool.net/mantisbt/view.php?id=24 - Use fcntl(2) rather than flock(2) to lock oid map files. (Tatsuo Ishii) flock(2) is not portable and cannot be used on Solaris. Patch contributed by Ibrar Ahmed. - Fix oversight of get_next_master_node() when operated in raw mode. (Tatsuo Ishii) If master node goes down, it always returned master node id 0. See [pgpool-general: 1039] for more details. [pgpool-general: 1039] Raw failover not working as expected on pgpool-II v3.2.0 From: Quentin White Date: Tue, 25 Sep 2012 07:45:34 +0000 http://www.sraoss.jp/pipermail/pgpool-general/2012-September/001057.html - Fix segfault of do_query(). (Tatsuo Ishii) When memqcache enabled and extended protocol is used, do_query() accesses system catalog and use pool_read2(). Unfortunately parse message packet is given to Parse() and the packet contents is on pool_read2's buffer. Thus do_query could break the packet contents, and it leads to segfault. Solution is, allocate memory and copies the packet contents and pass to Parse(). Note that query context holds query string, which is in the packet as well. So we need to copy it and save the pointer in the query context. We think the problem is not only with Parse() but with other protocol modules. So this fix is not Parse() only, rather for other modules. For this purpose ProcessFrontendResponse() is changed. See bugtrack in detai. #21 pgpool-II 3.2.0 cannot execute sql through jdbc Date: 2012-08-17 16:31 Reporter: elisechiang http://www.pgpool.net/mantisbt/view.php?id=21 - Fix to set unix domain socket path for pgpool PCP communication before setting up signal handlers. (Yugo Nagata) Previously, unlink of the socket in exitting process was failed because the path of the socket was not set. Patch contributed by Gilles Darold [pgpool-hackers: 131] Found bug with watchdog resulting in pgpool segmentation fault From: Gilles Darold Date: Thu, 13 Sep 2012 18:54:42 +0200 http://www.sraoss.jp/pipermail/pgpool-hackers/2012-September/000130.html - Fix to output the message when, in watchdog, ifup/down or arping command does not exist. (Yugo Nagata) - Fix long standing problem with do_query(). (Tatsuo Ishii) When 1) extended protocol used and 2)unnamed portal is used and 3) no explicit transaction is used, user's unnamed portal is removed by Sync message. This is because Sync message closes transaction and unnamed portal is removed. This leads to "portal "" does not exist" error. Fix is, use "Flush" message instead of Sync. Main difference between using Sync and Flush is, Flush does not return Ready for Query message. So do_query() does not return until all expected responses are returned. It seems the order of messages returned from backend is random, and do_query () manages it by using state bits. =============================================================================== 3.2.0 (namameboshi) 2012/08/03 * Version 3.2.0 This is the first version of pgpool-II 3.2 series. That is, a "major version up" from 3.1 series. __________________________________________________________________ * Incompatible changes - The new query cache "On memory query cache" took the place of the old one. - Now the parameter "enable_query_cache" is deleted. __________________________________________________________________ * New features ** Memory based query cache Original author is Masanori Yamazaki, improved by Development Group. (Tatsuo Ishii, Nozomi Anzai, Yugo Nagata) Overview: On memory query cache is faster because cache storage is on memory. Moreover you don't need to restart pgpool-II when the cache is outdated because the underlying table gets updated. On memory cache saves pair of SELECT statements (with its Bind parameters if the SELECT is an extended query). If the same SELECTs comes in, it returns the value from cache. Since no SQL parsing nor access to PostgreSQL are involed, it's extremely fast. On the other hand, it might be slower than the normal path because it adds some overhead to store cache. Moreover when a table is updated, pgpool automatically deletes all the caches related to the table. So the prformace will be degraded by a system with a lot of updates. If the cache_hit_ratio is lower than 70%, you might want to disable onl memory cache. Choosing cache storage: You can choose a cache storage: shared memory or memcached (you can't use the both). Query cache with shared memory is fast and easy because you don't have to install and config memcached, but restricted the max size of cache by the one of shared memory. Query cache with memcached needs a overhead to access network, but you can set the size as you like. Restrictions: - On memory query cache deletes the all cache of an updated table automatically with monitoring if the executed query is UPDATE, INSERT, ALTER TABLE and so on. But pgpool-II isn't able to recognize implicit updates due to trigers, foreign keys and DROP TABLE CASCADE. You can avoid this problem with memqcache_expire by which pgpool deletes old cache in a fixed time automatically, or with black_memqcache_table_list by which pgpool's memory cache flow ignores the tables. - If you want to use multiple instances of pgpool-II with online memory cache which uses shared memory, it could happen that one pgpool deletes cache, and the other one doesn't do it thus finds old cached result when a table gets updated. Memcached is the better cache storage in this case. New parameters: - Add parameters for on memoey query cache as follows: memory_cache_enabled, memqcache_method, memqcache_expire, memqcache_maxcache, memqcache_oiddir. (Tatsuo Ishii) - Add parameters about shared memory for on memory query cache as follows: memqcache_total_size, memqcache_max_num_cache, memqcache_cache_block_size. (Tatsuo Ishii) - Add parameters about memcached for on memory query cache as follows: memqcache_memcached_host, memqcache_memcached_port. (Tatsuo Ishii) - Add parameters about relation cache for on memory query cache as follows: relcache_expire, relcache_size. (Tatsuo Ishii) - Add a parameter "check_temp_table" to check if the SELECTed table is temp table. (Tatsuo Ishii) - Add the parameters of white_memqcache_table_list, black_memqcache_table_list that check if the SELECTed tables, temp tables and views are to be cached or not. (Nozomi Anzai) - Add a parameter "memqcache_auto_cache_invalidation" of the flag if query cache is triggered by corresponding DDL/DML/DCL (and memqcache_expire). (Yugo Nagata) New commands: - Add "SHOW pool cache" which shows hit ratio of query cache and the status of cache storage. - Add "--with-memcached" option to configure. (Nozomi Anzai) - Add "-C, --clear-oidmaps" option to "pgpool" command. (Nozomi Anzai) If pgpool with memcached starts / restarts with -C option, discard oid maps, if not, it can reuse old oid maps and query caches. ** Watchdog The author is Atsushi Mitani, tested by Yugo Nagata. Overview: "Watchdog" is a sub process of pgpool-II aiming for adding high availability feature to it. Features added by watchdog include: - Life checking of pgpool service Watchdog monitors responses of pgpool service rather than process. It sends queries to PostgreSQL via pgpool which is being monitored by watchdog and watchdog checks the response. Also watchdog monitors connections to up stream servers (application servers etc.) from the pgpool. The connection between the up stream servers and the pgpool is monitored as service of pgpool. - Mutual monitoring of watchdog processes Watchdog processes exchange information on the monitored servers to keep the information up to date, and to allow watchdog processes to mutually monitor each other. - Changing active/standby state in case of certain faults detected When a fault is detected in the pgpool service, watchdog notifies the other watchdogs of it. Watchdogs decide the new active pgpool if previous active pgpool is broken by voting and change active/standby state. - Automatic virtual IP address assigning synchronous to server switching When a standby pgpool server promotes to active, the new active server brings up virtual IP interface. Meanwhile, the previous active server brings down the virtual IP interface. This enables the active pgpool to work using the same IP address even when servers is switched over. - Automatic registration of a server as standby in recovery When broken server recovers or new server is attached, the watchdog process notifies the other watchdog process along with information of the new server, and the watchdog process receives information on the active server and other servers. Then, the attached server is registered as standby. New parameters: - Add a parameter to enable watchdog: use_watchdog. (Atsushi Mitani) - Add the parameters about life checking of pgpool service as follows: wd_interval,other_pgpool_port wd_life_point, wd_lifecheck_query, other_pgpool_hostname, other_pgpool_port. (Atsushi Mitani) - Add the parameters about up to stream connection (e.g. to application servers) as follows: trusted_servers, ping_path. (Atsushi Mitani) - Add the parameters about mutual monitoring of watchdog processes as follows: wd_hostname, wd_port, other_wd_port. (Atsushi Mitani) - Add the parameters about virtual IP as follows: ifconfig_path, if_up_cmd, if_down_cmd, arping_path, arping_cmd. (Atsushi Mitani) __________________________________________________________________ * Enhancements - Retry if health check faied rather than immediately do failover. For this purpose new directives "health_check_max_retries" and "health_check_retry_delay" were added. (Tatsuo Ishii) Patch contributed by Matt Solnit. Subject: [Pgpool-hackers] Health check retries (patch) From: Matt Solnit Date: Fri, 18 Nov 2011 16:28:44 -0500 - Log client IP and port number when pgpool failed to parse given query. (Tasuo) This is usefull to identify which client gives wrong query without enabling log_connections, which produces too many log entries on busy web systems. - Refactor pool_process_query(). (Tatsuo Ishii) Deal with the case when: 1) query not in progress 2) other than master node has pending data. It is reported that pgpool goes into infinite loop in this case. [pgpool-general: 43] Re: [Pgpool-general] seemingly hung pgpool process consuming 100% CPU http://www.pgpool.net/pipermail/pgpool-general/2011-December/000042.html - Add "role" which represents "primary" or "standby" iin streaming replication mode for example to "SHOW pool_nodes" command.(Tatsuo Ishii) - Add params to the result of "SHOW pool_status": backend_data_directory, ssl_ca_cert, ssl_ca_cert_dir, and sort by orders in pgpool.conf. (Anzai) - Commentout params about system db from pgpool.conf.sample-*. (Nozomi Anzai) - Add new parameter to failover/failback/followmaster command. (Tatsuo Ishii) %r: new master port number %R: new master database cluster path - Allow to reload to recognize md5 password change. (Tatsuo Ishii) Before, the only way to recognize md5 password change was restarting pgpool-II. Patch contributed by Gurjeet Singh. - Remove unused parameter "query" from is_set_transaction_serializable(). (Tastuo Ishii) - Fix on memory query cache section's comments in pgpool.com in the way which other sections employ. (Tastuo Ishii) - Add missing health_check_max_retries and health_check_retry_delay directives to pgpool.conf.sample-master-slave, pgpool.conf.sample-replication pgpool.conf.sample-stream. (Tastuo Ishii) - Fix pool_ssl_write. It seems someone forgot to do retrying when SSL returns SSL_ERROR_WANT_READ or SSL_ERROR_WANT_WRITE. Per [pgpool-general: 797] [pgpool-general: 797] Re: Problem with pgpool when using SSL, for client/pgpool communication http://www.sraoss.jp/pipermail/pgpool-general/2012-July/000807.html - Improve the design of manuals. (Nozomi Anzai) __________________________________________________________________ * Bug Fix - Fix memory leak in Raw mode.(Tatsuo Ishii) - Fix failover/failback in Raw mode. (Tatsuo Ishii) - Simply does not failover if target node is not master. - Fail to select master node if the node is in CON_UP. - Allow health check retry while connect(). (Tatsuo Ishii) It is reported that connect() blocks sigalarm under some conditions, for example: When system is configured for security reasons not to return destination host unreachable messages ([pgpool-general: 131]) Part of changes are contributed by Stevo Slavic. [pgpool-general: 131] Healthcheck timeout not always respected http://www.pgpool.net/pipermail/pgpool-general/2012-January/000131.html - Fix pool_send_and_wait() to send or not to send COMMIT / ABORT depending on the transaction state on each node. (Tatsuo Ishii) It is possible that only primary is in an explicit transaction but standby is not in it if multi statement query has been sent. Per bug report [pgpool-general-jp: 1049]. - Fix load balance in Solaris. (Tatsuo Ishii) Problem is, random() in using random() in Solaris results in strange load balancing calculation. Use srand()/rand() instead although they produce lesser quality random Problem reported at [pgpool-general: 396]. [pgpool-general: 396] strange load balancing issue in Solaris http://www.pgpool.net/pipermail/pgpool-general/2012-April/000397.html - Fix segfault of pcp_systemdb_info not in parallel mode. (Nozomi Anzai) - Fix "unnamed prepared statment does not exist" error. (Tatsuo Ishii) This is caused by pgpool's internal query, which breaks client's unnamed statements. To fix this, if extended query is used, named statement/portal for internal are used for internal query. - Fix hangup when query conflict occurs in Hot-Standby mode. (Yugo Nagata) Query example to reproduce: (S1) BEGIN; (S1) SELECT * FROM t; (S2) DELETE FROM t; (S2) VACUUM t; - Fix pool_process_query() bug reported in [pgpool-general: 672]. (Tatsuo Ishii) This is caused by the function waits for primary node which does not have pending data, while standbys have pending data. [pgpool-general: 672] Transaction never finishes http://www.pgpool.net/pipermail/pgpool-general/2012-June/000676.html - Fix wait_for_query_response() not to send param status to frontend if frontend is NULL. (Tatsuo Ishii) This could happen while processing reset_query_list and occur crash. - Fix bug with treatment of BEGIN TRANSACTION in master/slave mode. (Tatsuo Ishii) Original complain is [pgpool-general: 714]. From 3.1, pgpool-II sends BEGIN.. to all DB nodes. Of course we cannot send BEGIN TRANSACTION READ WRITE to standby nodes. Problem is, we did not check BEGIN WORK ISOLATION LEVEL SERIALIZABLE; and sent to standby nodes. Of course this is wrong, since it's not allowed to run transactions in serializable mode on standby nodes. So added check for BEGIN WORK ISOLATION LEVEL SERIALIZABLE case. [pgpool-general: 714] Load Balancing / Streaming Replication / Isolation Level serializable http://www.pgpool.net/pipermail/pgpool-general/2012-July/000719.html - Fix send_to_where() to send the query to only primary if the it is like SET TRANSACTION ISOLATION LEVELSERIALIZABLE etc. (Tatsuo Ishii) Case in streaming replication mode. Previously, it was sent to not only primary but also to standby and of course this causes an error. Similar SQLs are: SET SESSION CHARACTERISTICS AS TRANSACTION ISOLATION LEVEL SERIALIZABLE or SET transaction_isolation TO 'serializable' SET default_transaction_isolation TO 'serializable' Original complain is [pgpool-general: 715]. [pgpool-general: 715] Re: Load Balancing / Streaming Replication / Isolation Level serializable http://www.pgpool.net/pipermail/pgpool-general/2012-July/000720.html - Prevent errors even if memcached is not available with on memory query cache enabled and cache storage is memcached. (Tatsuo Ishii) This is mainly achieved by modifying pool_fetch_cache to pretend "cache not found" if memcached_get returns error other than MEMCACHED_NOTFOUND. Also we set pool_config->memory_cache_enabled to 0 to prevent future errors by accessing memcached. - Run rerun libtoolize with --copy and --force option. (Tatsuo Ishii) This should prevent build problems in different environment. Following commands were executed: libtoolize --copy --force aclocal autoheader automake -a autoconf - Fix pool_ssl_read. (Tatsuo Ishii) When SSL_read returns unknown error, treat as if EOF detected and returns 0 to caller (pool_read). This is same as libpq's behavior. Also this will avoid unwanted failover in pool_read. pool_read triggers failover if underlying I/O functions, read(2) or pool_ssl_read returns -1. - Fix pool_process_query. (Tatsuo Ishii) When other than primary sends packet, pgpool tries to terminate session gracefully and hung because now ssl_read returns EOF, rather than error and triggers failover. For example [pgpool-general: 766] reported this: 2012-07-17 00:11:03 NZST [15692]: [257-1] ERROR: canceling statement due to conflict with recovery 2012-07-17 00:11:03 NZST [15692]: [258-1] DETAIL: User query might have needed to see row versions that must be removed. 2012-07-17 00:11:03 NZST [15692]: [259-1] STATEMENT: 2012-07-17 00:11:03 NZST [15696]: [366-1] FATAL: terminating connection due to conflict with recovery 2012-07-17 00:11:03 NZST [15696]: [367-1] DETAIL: User query might have needed to see row versions that must be removed. In this case pool_process_query should return POOL_ERROR, rather than POOL_END. [pgpool-general: 766] Re: pgpool dropping backends too much http://www.pgpool.net/pipermail/pgpool-general/2012-July/000774.html - If frontend terminates while reading large number of query results from backend, pgpool continues to read from backend and write to frontend until all query results consumued. (Tatsuo Ishii) This will take very long time if query results is huge. To finish pgpool session as soon as possible, modify pool_flush_it to returns error if fail to write to frontend when operated in other than replication mode. In replication mode, we need to keep the previous behavior to save consistency among backends). Also fix SimpleForwardToFrontend to return pool_write_and_flush error status. Before it was ignored. =============================================================================== 3.1 Series (2011/09/08 - ) =============================================================================== 3.1.13 (hatsuiboshi) 2015/02/05 * Version 3.1.13 This is a bugfix release against pgpool-II 3.1.12. __________________________________________________________________ * Bug fixes - doc: Describe explicitly that the number of slave nodes is not necessarily 1 (Tatsuo Ishii) - Fix uninitialized variable (Tatsuo Ishii) Per Coverity 1234603. - doc: Fix missing release note entries in the previous release (Tatsuo Ishii) - Fix node id range check bug in trigger_failover_command(). (Tatsuo Ishii) The node id should be lower than NUM_BACKENDS. Probably harmless since callers never pass node ids greater or equal to NUM_BACKENDS. - Fix to disable debug mode by reloading config (Yugo Nagata) Per bug #114 http://www.pgpool.net/mantisbt/view.php?id=114 =============================================================================== 3.1.12 (hatsuiboshi) 2014/09/05 * Version 3.1.12 This is a bugfix release against pgpool-II 3.1.11. __________________________________________________________________ * Bug fixes - Fix a typo of pgpool.spec (Yugo Nagata) - Fix bug that worker child process keeps failing when there's no primary backend (Tatsuo Ishii) Problem identified and fix contributed by Junegunn Choi. See [pgpool-hackers: 471] for more details. - Close listen socket when smart shutdown request is made (Tatsuo Ishii) When smart shutdown process starts, pgpool children still listen on the port and clients can send further connection requests which fail in the end. Which is not only waste of time, but also prevents a load balancer which sits in front of pgpool from realizing the pgpool is going down. Problem analyzed and patch provided by Junegunn Choi in [pgpool-hackers 474], and enhanced to take care not only inet domain socket but UNIX domain socket by Tatsuo Ishii. - doc: Add cautions that recovery commands are killed by statement_timeout of PostgreSQL. (Tatsuo Ishii) - doc: Remove old restriction description which is no longer true (Tatsuo Ishii) - Fix return type of text_to_lsn() function (Yugo Nagata) This caused compile warning. - Fix file descriptor leak when daemonize. (Tatsuo Ishii) Per Coverity 1111471. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111442. - Fix pgpool.init's long-standing bug of stop/restart failure (Yugo Nagata) In previous, pgpool.init uses killproc for stopping pgpool, but there are several problems. In the new version, "pgpool -m fast stop" is used in stop command. Original patch contributed by Ryan DeShone and modified by Yugo Nagata. See [pgpool-hackers: 239][pgpool-hackers: 512]. - Disbale statement_timeout of PostgreSQL while executing online recovery (Tatsuo Ishii) Online recovery may take very long time and user may enable statement timeout. To prevent online recovery canceled by statement timeout, disable statement timeout in the connection used by online recovery. See [pgpool-general: 2919] for more details. - Remove unnecessary call to pool_shmem_exit() which removes semaphore when it shouldn't (Tatsuo Ishii) exit_handler checks if the process is parent or not. This is good. However, even if it is a child process, it calls pool_shmem_exit() which removes semaphore and shmem when it should not. It should be called only from parent process. Per bug #102. http://www.pgpool.net/mantisbt/view.php?id=102 - Fix reset query stuck problem. (Tatsuo Ishii) It is reported that reset query (DISCARD ALL etc.) occasionally does not finish and pgpool child remain waiting for reply from backend thus client cannot connect to pgpool. The cause of problem is not identified yet but if client suddenly closes connection to pgpool in the middle of query processing, backend may not accept the reset queries because they are not ready for query. The fix is, if frontend closes connection in unexpected way, query process loop immediately returns with new state: POOL_END_WITH_FRONTEND_ERROR and pgpool closes connection to PostgreSQL then goes back to new connection request waiting loop. Also, pgpool closes connections to backend when client_idle_limit is set and the idle limit. Per bug #107. http://www.pgpool.net/mantisbt/view.php?id=107 =============================================================================== 3.1.11 (hatsuiboshi) 2014/03/24 * Version 3.1.11 This is a bugfix release against pgpool-II 3.1.10. __________________________________________________________________ * Bug fixes - doc: Add mention about "listen queue" and how to increase the "backlog" in the num_init_children section. (Tatsuo Ishii) - Fix bad performance of unlogged tables detection code. (Tatsuo Ishii) Pointed out at [pgpool-hackers: 435][[pgpool-general:2325]. - Fix primary node detection logic. (Tatsuo Ishii) There's a possibility that primary node is not detected. This happens in following situation. node 0: primary, node 1: standby. Node 0 goes down. Health checking detects the fact but local status is not updated yet. Primary node finding (find_primary_node) runs. Node 0's status is yet healthy. Because find_primary_node fails to connect to node 0, it immediately returns -1 and fails to find that fact that node 1 is now primary. Fix is just continuing to look for primary node when fails to connect to a node. Per [pgpool-general: 2409]. - Fix jdbc DML fails when operated in raw mode and auto commit is off. (Tatsuo Ishii) This is reported in bug #92. - Avoid to send queries to unrelated nodes in streaming replication mode. (Tatsuo Ishii) Pgpool-II sends certain queries, such as BEGIN, END and SET commands to all of DB nodes. However in streaming replication mode, only primary node and at most one standby node are only concerned (if primacy node is selected as the load balance node, only 1 node is concerned). See [pgpool-hackers: 464] for more details. - Fix possible buffer overrun problem and memory leak. (Tatsuo Ishii) Per Coverity 1111465 and 1111482. - Fix possible buffer overrun. (Tatsuo Ishii) Replace strcpy() with strlcpy(). Per Coverity report 1111478, 1111480, 1111481. - Fix freeing NULL. (Tatsuo Ishii) Per Coverity 1111384. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111446. - Fix a segmentation fault in parallel mode with system_db_hostname is empty (Yugo Nagata) - Fix to output debug messages in processing pgpool.conf with -d option (Yugo Nagata) Previously, some debug messages in pool_get_config() ware not output even when -d option was used. - Fix JDBC exception of prepared statement including now() in replication mode (Yugo Nagata) With JDBC, when a prepared statement is executed more than PrepareThreshold times, the statement is named and Describe message is sent after Parse. With named statement, pgpool rewrite now() to parameter in replication mode. Hence, rewritten query has additonal parameter than original. In this case, ParameterDescription message sent to frontend (response of Describe) should include OIDs of the same number os original query's parameters. Otherwize, JDBC throws ArrayIndexoutOfBoundsException. This is reported in [pgpool-general-jp: 1192]. - Fix backend error of prepared statement about table which has column whoes default value is now() in replication mode (Yugo Nagata) When pgpool parses a named prepared statement with default now(), timestamps are replaced to additional parameters. So, Bind message also should inclued additional parameter format codes. However, when the number of original parameter was one, pgpool didn't handle this. This caused a error like "incorrect binary data format in bind parameter 2". - doc: Add description about parallel mode doesn't support PREPARE (Yugo Nagata) Per bug #93 =============================================================================== 3.1.10 (hatsuiboshi) 2013/12/06 * Version 3.1.10 This is a bugfix release against pgpool-II 3.1.9. __________________________________________________________________ * Bug fixes - Fix incorrect time stamp rewriting in replication mode for certain time zones. (Tatsuo Ishii) Time stamp rewriting calls "SELECT now()" to get current time. Unfortunately the buffer for the current time is too small for certain time zones such as "02:30". Note that non-30-minutes-time-zone such as "0900" does not reveal the problem. This explains why we haven't the bug report until today. Bug reported in [pgpool-general: 2113] and fix provided by Sean Hogan. http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002142.html - Makefile: Replace pg_config by $(PG_CONFIG) in Makefiles so it can be overridden at build time when compiling for different PG major versions. (Tatsuo Ishii) Patch contributed by Christoph Berg ([pgpool-general: 2127]). http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002156.html - configure: Remove -lcompat because it confuses FreeBSD per bug#15. (Tatsuo Ishii) http://www.pgpool.net/mantisbt/view.php?id=15 - Fix segfault when pgpool.conf does not set log_standby_delay. (Tatsuo Ishii) This is caused by wrong initialization for log_standby_delay in pool_config.l. Per bug#74. http://www.pgpool.net/mantisbt/view.php?id=74 - doc: Modify descriptions about restrictions of parallel mode Muliple rows INSERT using VALUES are not supported in parallel mode. (Yugo Nagata) - Avoid calling find_primary_node_repeatedly() when standby node goes down. (Tatsuo Ishii) This will reduce the time to failover. Per bug #75, patch modified by Tatsuo Ishii. http://www.pgpool.net/mantisbt/view.php?id=75 - Fix data inconsistency problem with native replication mode + extended protocol case. (Tatsuo Ishii) It is reported that concurrent INSERT using JDBC driver causes data difference among database node. This only happens following conditions are all met: 1) Native replication mode 2) Extended protocol used 3) The portal created by parse message is reused by bind message 4) autocommit is on 5) SERIAL (sequence) is used Pgpool-II's parse message function knows it has to lock the target table when INSERT (plus #5) is issued by clients. Unfortunately bind message function did not know it. Once parse/bind/execute finishes, pgpool releases the lock obtained by parse because of #4. JDBC wants to reuse the portal and starts the cycle from bind message, which does not obtain lock. As as result, lock-free INSERT are floating around which causes data inconsistency of course. The solution is, lock the table in bind phase. For this bind needs to issue LOCK in extended protocol. This was a little bit hard because the module (do_command()) to issue internal SQL command (other than SELECT) does not support extended protocol. To solve the problem do_query() is modified so that it accepts other than SELECT because it already accepts extended protocol. The modification is minimum and is only tested for the case called from insert_lock(). I do not recommend to replace every occurrence of do_command () with do_query() at this point. BTW the reason why the bug is not reported is, most users uses JDBC with auto commit = off. In this case, the lock obtained by parse persists until user explicitly issues commit or rollback. Per bug report by Steve Kuekes in [pgpool-general: 2142]. http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002171.html - Fix error message in read_password_packet(). (Tatsuo Ishii) - Fix memory leak when do_query() fails in timestamp rewriting. (Tatsuo Ishii) For this purpose free_select_result() is changed to accept NULL argument. Per Coverity report "1111454, 1111455 Resource leak". - Fix target node selection logic when "DEALLOCATE portal|statement". (Tatsuo Ishii) When "DEALLOCATE portal|statement" is used and last prepared statement or portal was not found, target node selection map is not set. Probably this is not actually harmful because prepared statement or portal was not found is an error case. The bug was there since day 0. Per Coverity report "1111491 Structurally dead code". - Fix range check bug of MAX_NUM_BACKENDS in corner case. (Tatsuo Ishii) MAX_NUM_BACKENDS is the allowed max number of DB nodes (128, at this point). In reality, probably no one ever tried more than 128 DB nodes and that's the reason why nobody noticed. Per Coverity report "1111429, 1111430 and 1111431 Out-of-bounds write". - Fix that the script forgets to allow public access to pgpool_catalog. (Tatsuo Ishii) The bug prevents inserting data into user tables if pgpool_catalog is created in native replication mode. The bug was there from day 1. I wonder why nobody noticed until today. Per [pgpool-general-jp: 1229]. http://www.sraoss.jp/pipermail/pgpool-general-jp/2013-November/001228.html - Fix uninitialized variable in error case in pool_do_auth(). (Tatsuo Ishii) If there's no valid backend, pgpool will return garbage pid to frontend in auth phase. Actually because no backend is available, frontend will be disconnected later on. So this is not harmless. Per Coverity report "1127331 Uninitialized scalar variable". - Fix to add node id range check when issue an error message using node id. (Tatsuo Ishii) Per Coverity report #1111433 "Out-of-bounds read". - Fix buffer overrun bug and resource leak bug of parse_copy_data(). (Tatsuo Ishii) Per Coverity report 1111427 "Out-of-bounds write" and 1111453 "Resource leak". - Fix possible segfault in CopyDataRaws(). (Tatsuo Ishii) Coverity pointed out that if pool_get_id() returns an error, VALID_BACKEND will access out of array. Per Coverity report 1111413 "Memory - illegal accesses". - Fix resource leak in make_persistent_db_connection. (Tatsio Ishii) For this pupose, new static function free_persisten_db_connection_memory is added. Per Coverity report #1111468. =============================================================================== 3.1.9 (hatsuiboshi) 2013/09/06 * Version 3.1.9 This is a bugfix release against pgpool-II 3.1.8. __________________________________________________________________ * Bug fixes - Fix a mistake in ssh command of doc/basebackup.sh (Tatsuo Ishii) - Fix a bug in parsing prepared statements with transaction handling in replication mode (Tatsuo Ishii) Parse() automatically starts a transaction for non SELECT query to keep consistency among nodes in replication mode. But this wasn't closed. If wrong query comes in, the transaction goes into an abort state but pgpool does not close the transaction. Thus next query causes error because the transaction is still in abort status. This problem was reported in [pgpool-general: 1877] by Sean Hogan. [pgpool-general: 1877] current transaction is aborted, commands ignored http://www.sraoss.jp/pipermail/pgpool-general/2013-July/001905.html - Fix typos of the japanese document (Yugo Nagata) =============================================================================== 3.1.8 (hatsuiboshi) 2013/07/10 * Version 3.1.8 This is a bugfix release against pgpool-II 3.1.7. __________________________________________________________________ * Bug fixes - Add mention about "-D" option to the man page. (Tatsuo Ishii) - Consider timeout waiting for compeletion of failback request in on line recovery (Tatsuo Ishii) This will prevent the situation that recovery operation continues forever and we cannot even shutdown pgpool-II main process. This could happen especially while executing follow master command. - Fix a bug that %H of follow_master_command is not assigned correctly the new primary node in stream replication mode (Tatsuo Ishii) - Fix do_query() to not hang when PostgreSQL returns an error (Tatsuo Ishii) The typical symptom is "I see SELECT is keep on running according to pg_stat_activity". To fix this pgpool-II just exits the process and kill the existig connection. This is not gentle but at this point I believe this is the best solution. - Fix bug with do_query which causes hung in extended protocol (Tatsuo Ishii) This problem could occur when insert lock is enabled and pgpool_catalog.insert_lock exists, See [pgpool-general: 1684] for more details. [pgpool-general: 1684] insert_lock hangs http://www.sraoss.jp/pipermail/pgpool-general/2013-May/001711.html - Fix unnecessary degeneration caused by error on commit (Tatsuo Ishii) In master slave mode, if master gets an error at commit, while other slaves are normal at commit, we don't need to degenrate any backend because it is likely that the "kind mismatch error" was caused by a deferred trigger. - Fix to register pgpool_regclass in pg_catalog schema (Tatsuo Ishii) This is necessary to deal with clients which restricts schema search path to pg_catalog only. Postgres_fdw is such a client. - Fix a potential crash in pg_md5 command (Muhammad Usama) - Fix a segmentation fault of a child process that occurs when a startup packet has no PostgreSQL user information (Yugo Nagata) You can reproduce it by $ psql -p 9999 -U '' If enable_pool_hba is on, a child process terminates by segmentation fault. Otherwise if enable_pool_hba is off, the error message is ERROR: pool_discard_cp: cannot get connection pool for user (null) database (null) In both cases, psql terminates with no message on frontend. In the fixed version, if PostgreSQL user is not specified in startup packet, the message as following is output to both log and frontend. This is the same behavior as PostgreSQL. FATAL: no PostgreSQL user name specified in startup packet - Move ssl_ca_cert and ssl_ca_cert_dir descriptions to the SSL section (Yugo Nagata) - Add ssl_ca_cert and ssl_ca_cert_dir descriptions to the japanese document (Yugo Nagata) - Fix to verify the backend node number in pcp_recovery_node (Yugo Nagata) When an invalid number is used, null value is passed as an arguments of recovery script, and this causes a malfunction. In especially, rsync may delete unrelated files in basebackup scripts. =============================================================================== 3.1.7 (hatsuiboshi) 2013/4/26 * Version 3.1.7 This is a bugfix release against pgpool-II 3.1.6 __________________________________________________________________ * Bug fixes - Fix to show pool_passwd in "SHOW pool_status". (Yugo Nagata) - Fix long standing bug with timestamp rewriting code for processing extended protocol. (Tatsuo Ishii) Parse() allocate memory using palloc() while rewriting the parse message. Problem is, the rewritten message was kept in the data which is managed by pool_create_sent_message() etc. The function assumes that all the data is in session context memory. However, palloc() allocates memory in query context of course, and gets freeed later on when the query context disappears. And the function tries to free the memory as well, which causes various problems, including segfault and double free. To fix this, memory to store rewritten message is allocated using session context. The bug was there since pgpool-II 3.0 was born. Problem analysis and patch contributed by Naoya Anzai. [pgpoolgenera-jp: 1146]. (in Japanese) http://www.pgpool.net/pipermail/pgpool-general-jp/2013-March/001145.html - Fix bug with md5 auth long user name handling. (Tatsuo Ishii) If user name is longer than 32 bytes, md5 authentication doesn't work. Problem reported in [pgpool-general: 1526] by Thomas Martin. [pgpool-general: 1526] [pgPool-II 3.2.3] MD5 authentication and username longer than 32 characters. http://www.pgpool.net/pipermail/pgpool-general/2013-March/001551.html - Fix to calculate replication delay only if standby server is behind from the primay server. (Yugo Nagata) When the primary server is behind from standby server, negative value of delay is calculated and the value is assigned to unsigned variable. It causes a log message informing negative replication delay. And what is worse, it also causes SELECT queries to be sent to the primary in load balance even though there are no replication delay in fact. The problem is reported and analyzed by Saitoh Hidenori in [pgpool-genera-jp: 1145]. [pgpool-general-jp: 1145] (in Japanese) http://www.pgpool.net/pipermail/pgpool-general-jp/2013-March/001144.html - pgpool-recovery adopts PostgreSQL 9.3. (Tatsuo Ishii) Patch contributed by Asif Rehman. Slight editing by Tatsuo Ishii. [pgpool-hackers: 180] compile error in ppool-recovery http://www.pgpool.net/pipermail/pgpool-hackers/2013-April/000179.html - Fix pool_has_pgpool_regclass() to check execute privilege of pgpool_regclass(). (Tatsuo Ishii) Even though pgpool_regclass() exists, if pgpool cannot execute the function, the connection to backend hangs. You can reproduce the problem by just dropping the execute privilege from pgpool_regclass and do some insert in native replication mode. The problem is reported in bugtrack #53. #53 pgpool_regclas hangs all connections Date: 2013-04-04 13:35 Reporter: tmandke http://www.pgpool.net/mantisbt/view.php?id=53 - Fix error message mistakes in detect_postmaster_down_error(). (Tatsuo Ishii) For example, "LOG: detect_stop_postmaster_error: detect_error error" is fixed to "LOG: detect_postmaster_down_error: detect_error error", and so on. =============================================================================== 3.1.6 (hatsuiboshi) 2013/2/8 * Version 3.1.6 This is a bugfix release against pgpool-II 3.1.5 __________________________________________________________________ * Bug fixes - Fix race condition when using md5 authentication. (Tatsuo Ishii) The file descriptor to pool_passwd is opened in pgpool main and pgpool child inherits it. When concurrent connections try to authenticate md5 method, they call pool_get_passwd and seek the fd and cause random md5 auth failure because underlying fd is shared. Fix is, let individual pgpool child open the file by calling pool_reopen_passwd_file. Problem reported and analyzed by Jason Slagle in pgpool-general:1141. [pgpool-general: 1141] Possible race condition in pool_get_passwd From: Jason Slagle Date: Sun, 28 Oct 2012 01:12:52 -0400 http://www.sraoss.jp/pipermail/pgpool-general/2012-October/001160.html - Fix hung up while repeating pcp_attach_node and pcp_detatch_node (Tatsuo Ishii) When node status is changed by pcp_attach_node and pcp_detatch_node, failover() sends SIGUSR1 to pcp_child process expecting it exits to refresh node status. In this situation lots of pgpool children exit and produce SIGCHLD as well. The SIGCHLD handler reaper() tries catch all SIGCHLD but sometimes it fails depending on the system load and timing. If SIGCHLD produced by pcp child is not caught, the process becomes zombie and never restarted. This problem is reported in bug track #32 (by oleg_myrk) etc. #32 PGPool hangs on pcp_attach/detach Reporter: oleg_myrk $B!!(BDate: 2012-10-24 00:01 http://www.pgpool.net/mantisbt/view.php?id=32 - Fix pool_send_severity_message() not to use uninitialized memory. (Tatsuo Ishii) It cause a segmentaion fault. Reported in Bug #33's attached valgrind output by dudee. #33 pgpool-II 3.2.1 segfault Reporter: dudee Date: 2012-10-30 19:16 http://www.pgpool.net/mantisbt/view.php?id=33 - Add a description about "-f" to help message. (Tatsuo Ishii) - Fix reaper() not to exit wait3() loop when catches pcp or worker child exit event. (Tatsuo Ishii) Otherwise reaper() mistakenly ignore some process exit event and make a risk of creating zombie process and forgetting to create new process. Problem reported and fix suggested by Goto in [pgpool-general-jp: 1123]. http://www.sraoss.jp/pipermail/pgpool-general-jp/2012-November/001122.html (in Japanese) - Fix pool_search_relcache() to use MASTER or MASTER_NODE_ID macro, rather than REAL_MASTER_NODE_ID. (Tatsuo Ishii) In case node 0 fail back in streaming replication mode, pgpool does not restart child process. So REAL_MASTER_NODE_ID looks for node 0 con info, which is not present until new connection to backend made. Thus referring to node con info results in segfault. MASTER or MASTER_NODE_ID are safe in this situation because they look at cached former master node id. - Fix long standing bug "portal not found" error when replication delay is too much in streaming replication mode. (Tatsuo Ishii) The bug had been there since the delay threshold was introduced. We changed destination DB node if delay threshold exceeds in bind, describe and execute. However, if parse sends to different node, bind, describe or execute will fail because no parsed statement or portal exists. Solution is, not to send to different parse node even if delay threshold is too much. - Fix pg_md5 to output "\n" after user inputs password. (Yugo Nagata) - Fix child_exit() to not call send_frontend_exits() if there's no connection pool. (Tatsuo Ishii) Otherwise, it segfaults because send_frontend_exits() referes to objects pointed to by pool_connection_pool. Per bug track #44 by tuomas. #44 pgpool went haywire after slave shutdown triggering master failover Reporter: tuomas Date: 2012-12-11 00:33 http://www.pgpool.net/mantisbt/view.php?id=4 - Fix read_startup_packet() to reset alarm and free StartupPacket when pool_read() returns 0 which means incorrect packet length. (Nozomi Anzai) Previously, authentication timeout occurs when connected by a program monitoring the pgpool port.It is reported in bug track #35 by tuomas. #35 Authentication is timeout Reporter: tuomas Date: 2012-11-20 11:54 http://www.pgpool.net/mantisbt/view.php?id=3 - Fix long standing bug with pool_open(). (Tatsuo Ishii) It initializes wrong buffer pointer. Actually this is harmless because the pointer is initialized by prior memset() call, though. - Modify documents to correct information of whether a certain parameter change requires restart. (Yugo Nagata) - Add pool_passwd option to pgpool.conf.sample*, and documents. (Yugo Nagata) =============================================================================== 3.1.5 (hatsuiboshi) 2012/10/12 * Version 3.1.5 This is a bugfix release against pgpool-II 3.1.4 __________________________________________________________________ * Bug fixes - Fix read_startup_packet. (Tatsuo Ishii) If packet length is lower than 0, it should have returned immediately. Otherwise it would cause memory allocation error later on. per pgpool-general:886. Also add canceling alarm. - Add NOTICE message handling to s_do_auth. (Tatsuo Ishii) Without this, health check responses false alarm and causes failover. per bug track: http://www.pgpool.net/mantisbt/view.php?id=25 Also allow to receive ready for query packet *not* right after backend keydata. I'm not sure if this could happen in the real world but the protocol seems to allow this. - Remove unnecessary/confusing debug log from s_do_auth.(Tatsuo Ishii) - Fix inifinit loop in SSL mode. When there's pending data in SSL layer of frontend, pool_process_query() checks pending data in backend. (Tatsuo Ishii) If there's non, it loops again and checks frontend/backend receive buffer by using is_cache_empty(). Unfortunately it first checks pending data in SSL layer of frontend, thus goes to backend data and checks again (infinite loop). The solution is, if there's pending data in SSL layer of frontend and query is not in progress, call ProcessFrontendResponse() to process new request from frontend. - Fix is_system_catalog to use pgpool_regclass if available. (Tatsuo Ishii) - Fix memory leak in pool_get_insert_table_name(). (Tatsuo Ishii) If session context's memory contex is used for nodeToString(), memory is not freed until session ends. - Fix possible memory leak in nodeToString(). (Tatsuo Ishii) Since the memory context could be session context, memory used for String object may be considerable if user session continues for hours or days. See bug track #24 for more details. - Fix long standing problem with do_query(). (Tatsuo Ishii) When 1) extended protocol used and 2)unnamed portal is used and 3) no explicit transaction is used, user's unnamed portal is removed by Sync message. This is because Sync message closes transaction and unnamed portal is removed. This leads to "portal "" does not exist" error. Fix is, use "Flush" message instead of Sync. Main difference between using Sync and Flush is, Flush does not return Ready for Query message. So do_query() does not return until all expected responses are returned. It seems the order of messages returned from backend is random, and do_query () manages it by using state bits. =============================================================================== 3.1.4 (hatsuiboshi) 2012/08/06 * Version 3.1.4 This is a bugfix release against pgpool-II 3.1.3. __________________________________________________________________ * General - Adopt PostgreSQL 9.2. (Tatsuo Ishii) __________________________________________________________________ * Bug fixes - Fix pool_send_and_wait() to send or not to send COMMIT / ABORT depending on the transaction state on each node. (Tatsuo Ishii) It is possible that only primary is in an explicit transaction but standby is not in it if multi statement query has been sent. Per bug report [pgpool-general-jp: 1049]. - Fix load balance in Solaris. (Tatsuo Ishii) Problem is, random() in using random() in Solaris results in strange load balancing calculation. Use srand()/rand() instead although they produce lesser quality random Problem reported at [pgpool-general: 396]. [pgpool-general: 396] strange load balancing issue in Solaris http://www.pgpool.net/pipermail/pgpool-general/2012-April/000397.html - Add params to the result of "SHOW pool_status": backend_data_directory, ssl_ca_cert, ssl_ca_cert_dir. (Nozomi Anzai) - Fix segfault of pcp_systemdb_info not in parallel mode. (Nozomi Anzai) - Fix "unnamed prepared statment does not exist" error. (Tatsuo Ishii) This is caused by pgpool's internal query, which breaks client's unnamed statements. To fix this, if extended query is used, named statement/portal for internal are used for internal query. - Fix is_system_catalog(). Its relcach was accidently defined as "session local". (Tatsuo Ishii) - Fix hangup when query conflict occurs in Hot-Standby mode. (Yugo Nagata) Query example to reproduce: (S1) BEGIN; (S1) SELECT * FROM t; (S2) DELETE FROM t; (S2) VACUUM t; - Improve reading and writing pid_file. (Tatsuo Ishii) - Fix pool_process_query() bug reported in [pgpool-general: 672]. (Tatsuo Ishii) This is caused by the function waits for primary node which does not have pending data, while standbys have pending data. [pgpool-general: 672] Transaction never finishes http://www.pgpool.net/pipermail/pgpool-general/2012-June/000676.html - Fix wait_for_query_response() not to send param status to frontend if frontend is NULL. (Tatsuo Ishii) This could happen while processing reset_query_list and occur crash. - Fix bug with treatment of BEGIN TRANSACTION in master/slave mode. (Tatsuo Ishii) Original complain is [pgpool-general: 714]. From 3.1, pgpool-II sends BEGIN.. to all DB nodes. Of course we cannot send BEGIN TRANSACTION READ WRITE to standby nodes. Problem is, we did not check BEGIN WORK ISOLATION LEVEL SERIALIZABLE; and sent to standby nodes. Of course this is wrong, since it's not allowed to run transactions in serializable mode on standby nodes. So added check for BEGIN WORK ISOLATION LEVEL SERIALIZABLE case. [pgpool-general: 714] Load Balancing / Streaming Replication / Isolation Level serializable http://www.pgpool.net/pipermail/pgpool-general/2012-July/000719.html - Fix send_to_where() to send the query to only primary if the it is like SET TRANSACTION ISOLATION LEVELSERIALIZABLE etc. (Tatsuo Ishii) Case in streaming replication mode. Previously, it was sent to not only primary but also to standby and of course this causes an error. Similar SQLs are: SET SESSION CHARACTERISTICS AS TRANSACTION ISOLATION LEVEL SERIALIZABLE or SET transaction_isolation TO 'serializable' SET default_transaction_isolation TO 'serializable' Original complain is [pgpool-general: 715]. [pgpool-general: 715] Re: Load Balancing / Streaming Replication / Isolation Level serializable http://www.pgpool.net/pipermail/pgpool-general/2012-July/000720.html =============================================================================== 3.1.3 (hatsuiboshi) 2012/04/23 * Version 3.1.3 This is a bugfix release against pgpool-II 3.1.2. __________________________________________________________________ * Bug fixes - Add m4 files. This should prevent compiling problem on older OS's. (Tatsuo Ishii) - Fix to handle failover properly in detect_postmaster_down_error(). (Tatsuo Ishii) It is possible that it fails to read backend socket after detecting backend errors and before actually detaching the backend. - Fix bug that the process exits before unlocking semaphore by a signal interrupt. (Tatsuo Ishii) - Fix a memory leak in case of reset_query. (Tatsuo Ishii) - Fix pool_ssl_read() to deal with large data reading. (Tatsuo Ishii) Original complain is here: http://www.pgpool.net/pipermail/pgpool-general/2012-March/000299.html - Fix deadlock by enabling log_destination = syslog. (Tatsuo Ishii) Reported in bug tracker http://www.pgpool.net/mantisbt/view.php?id=9. Patch provided by Gilles Darold. - Allow to use multi statement in master/slave mode. (Tatsuo Ishii) From 3.1 transactional statements such as "BEGIN" are sent to not only primary but also standbys. This brings an unfortunate side effect: if the multi statement is "BEGIN;DELETE FROM table;END", this will be sent to standy as well and cause an error because standby does notallow write SQL. So fix is, if a query is a multi statement one, then send it to primary only. - Allow to have private cache of master node id. (Tatsuo Ishii) When master goes down running pgpool children look into NULL pointer because the process does not connect to the new master yet. The private cache returns the old master node id and will avoid the situation. Per bug reported in http://www.pgpool.net/mantisbt/view.php?id=51 - Fix pool_start_query() so that it uses my_master_node_id instead of REAL_MASTER_NODE_ID for initial virtual_master_node_id. (Tatsuo Ishii) Real master node could be changed while pgpool is running in streaming replication mode. Previously it is assigned REAL_MASTER_NODE_ID, which could have no connections after fail over and cause segafult. - Fix pool_setall_node_to_be_sent() to use private_backend_status instead of BACKEND_INFO macro. (Tatsuo Ishii) It is possible that while processing, the node returned by BACKEND_INFO is not usable any more. - Fix failover(). (Tatsuo Ishii) Before it only restarts worker child only when pgpool child do not need restart. This is wrong. We need to restart worker child in any case. Otherwise it continues to send replication time lag check request to down the node. - Fix debug log message so that it does not contain null characters. (Toshihiro Kitagawa) - Fix SimpleQuery() so that it restores parser memory context when: 1) Builtin show commands are used 2) Parallel query mode 3) Query cache is used (Tatsuo Ishii) - Fix hangup when PREPARE statement causes error. (Toshihiro Kitagawa) This issue was reported by Tomonari Katsumata: Subject: [pgpool-general: 121] question of pgpool's behavior - Add doc/pgpool-fr.html to Makefile.am. (Tatsuo Ishii) This had been forgotten when the doc was added. - Fix hangup during md5 authentication that occurs in daemon mode and log_destination is 'syslog'. (Yugo Nagata) Reported in bug tracker http://www.pgpool.net/mantisbt/view.php?id=2. =============================================================================== 3.1.2 (hatsuiboshi) 2012/01/31 * Version 3.1.2 This is a bugfix release against pgpool-II 3.1.1. __________________________________________________________________ * Bug fixes - Fix to recognize READ UNCOMMITTED and REPEATABLE READ of transaction isolation levels. (Tatsuo Ishii) - Fix infinite loop reported in this thread (Tatsuo Ishii): http://www.pgpool.net/pipermail/pgpool-general/2011-December/000099.html It was not considered the case that, when received buffer in primary was empty but the one in a standby was not, the standby spontaneously sent packet to pgpool. This could happen when, for example, reloading postgresql.conf. The fix is that such buffer in standby is discarded. =============================================================================== 3.1.1 (hatsuiboshi) 2011/12/06 * Version 3.1.1 This is a bugfix release against pgpool-II 3.1. __________________________________________________________________ * Bug fixes - Fix add_regex_pattern(). It does not allocate enough memory for each black/white_function_list items. The function adds "^" and "$" to each function items which do not contain those characters. Unfortunately the function forgot to add extra 2 bytes for those characters. This may lead to memory corruption errors when pgpool starting up. - Fix error message of check_replication_time_lag(Tatsuo Ishii). It emitted wrong error message when it failed to connect to PostgreSQL while checking streaming replication delay. Because it does not use health_check_user anymore. - Fix memory leak(Toshihiro Kitagawa). This is essentially same as the fix made for 3.0.5(commit 19a4ea9215da0b61728741fc0da2271958b09238). - Major cleanup for strncpy(Tatsuo Ishii). There are several places where strncpy() is used. Problem is some of them do not consider the case when copy lengh == buffer size. In this case copied buffer is not null terminated and may cause tons of problems later. To fix this, most of them are replaced by strlcpy(). - Update cached backend status whenever possible(Tatsuo Ishii). This solves the problem of follow_master_command not being able to lookup backend status correctly which was reported by Jeff Frost: Subject: [Pgpool-general] diagnosing BackendError from pcp_recovery_node To: pgpool-general@pgfoundry.org Date: Wed, 05 Oct 2011 15:15:07 -0700 - Fix buffer overrun problem when pcp password is longer than 32(Tatsuo Ishii). - Remove PGDLLIMPORTI which is only neccessary for Windows and cause a problem for non gcc. Patch contributed by Ibrar Ahmed. =============================================================================== 3.1 (hatsuiboshi) 2011/09/08 * Version 3.1 This is the first version of pgpool-II 3.1 series. That is, a "major version up" from 3.0 series. __________________________________________________________________ * Incompatible changes - Change the lock method of insert_lock. The previous insert_lock uses row locking against the sequence relation, but the current one uses row locking against pgpool_catalog.insert_lock table. The reason is that PostgreSQL core developers decided to disallow row locking against the sequence relation to avoid an internal error which it leads. So creating insert_lock table in all databases which are accessed via pgpool-II beforehand is required. If does not exist insert_lock table, pgpool-II locks the insert target table. This behavior is same as pgpool-II 2.2 and 2.3 series. If you want to use insert_lock which is compatible with older releases, you can specify lock method by configure options: --enable-sequence-lock, --enable-table-lock(Toshihiro Kitagawa) - Deprecate backend_socket_dir. Instead, if backend_hostname starts with '/' it is regarded the as path to Unix domain. If backend_hostname is left empty, then default Unix domain path(/tmp) is used. This follows the convention of libpq interface. Patch contributed by Jehan-Guillaume (ioguix) de Rorthais. - Now "pgpool_walrecrunning()" was not used. pgpool-II used to consider the node that is promoted a primary node using the function. Now, pgpool-II waits for completing of the promotion to primary node because it did not work as we intended. But we still have a problem that pgpool-II waits while recovery_timeout, when there is no primary node(Toshihiro Kitagawa) - Add node_id to each PostgreSQL DB node info in the output of show pool_nodes(Jean-Paul Argudo) - Change the handling of sequence functions(nextval, setval) so that they completely obey setting of black/white_function_list. They were always handled as write functions before(Toshihiro Kitagawa) __________________________________________________________________ * New features - Add syslog support. Patch contributed by Gilles Darold. Review and editing by Guillaume Lelarge. - Adapt application_name introduced in PostgreSQL 9.0. When reusing connection, send application_name in the startup packet to backend and send parameter status to frontend(Tatsuo Ishii) - Add relcache_expire directive to control the expiration of the internal system catalog cache. ALTER TABLE might make these cache values obsoleted and the new directive will make the risk lower(Tatsuo Ishii). - Add follow_master_command directive. This directive specifies a command to run in master/slave streaming replication only after a master failover. Patch contributed by Gilles Darold. - Add pcp_promote_node command. This command promotes a new master node to pgpool-II. This can use in master/slave streaming replication only. Patch contributed by Gilles Darold. - Add pcp_pool_status command which produces similar output of show pool_status. Also C API for this command is added. Patch contributed by Jehan-Guillaume (ioguix) de Rorthais. - Add new per backend directive "backend_flag". This controls per backend behavior. Currently "ALLOW_TO_FAILOVER" or "DISALLOW_TO_FAILOVER" are allowed(Tatsuo Ishii) - Add health_check_password directive, but is not yet implemented (Nicolas Thauvin) - Add sr_check_period, sr_check_user and sr_check_password directives. These are used for streaming replication delay checking and determining primary node(Tatsuo Ishii) - Add --username(or -u) option to pg_md5 command. This allows to manage users which do not have UNIX accounts. Japanese document change by Tatsuo Ishii(Nicolas Thauvin) - Add pgpool_adm functions to pgpool_adm/. These are user-defined functions written in C language which work like pcp commands (Jehan-Guillaume (ioguix) de Rorthais) - Add Simplified Chinese version of documents(Huang Jian, Sun Peng) - Add SQL files to uninstall functions to sql/(Nicolas Thauvin) - In master/slave mode, SELECTs to unlogged table execute only on master/primary(Toshihiro Kitagawa) __________________________________________________________________ * Bug fixes - Fix bug which cannot use the cursors of JDBC driver on standby node. The transaction commands come to be sent to all nodes by this fix in master/slave mode(Toshihiro Kitagawa) - Fix bug with the handling of empty queries. The empty queries come to be handled the same as SELECT queries. This fix allows load-balance after the empty query(Toshihiro Kitagawa) - Fix insert_lock so that it works correctly even if the column definition such as "DEFAULT nextval(('"x_seq"'::text)::regclass)" (Toshihiro Kitagawa) - Fix pcp_attach_node command so that it emits error message while doing failover(Toshihiro Kitagawa) - Fix log message which is emitted when pgpool-II cannot parse the query in the extended query protocol so that it shows the query (Toshihiro Kitagawa) - Fix description about backend_weight inpgpool-II manual. It can be changed by reloading pgpool.conf(Tatsuo Ishii) - Fix and enhance wording in English tutorial document. Fix suggested by Huang Jian(Tatsuo Ishii) - Fix bug which does not update the node status when reattaching the node in raw mode(Guillaume Lelarge) - Fix incorrect calculation of the replication delay in streaming replication mode(Tatsuo Ishii) - Replace wrong function name "notice_backend_error" with correct one "degenerate_backend_set" in the failover log message(Tatsuo Ishii) - Remove unnecessary logging at the end of pgpool.conf parsing(Tatsuo Ishii) - Fix possible crash of pgpool/worker child after attaching new backend. Fix suggested by Gurjeet Singh(Tatsuo Ishii) - Fix bug that SELECTs which have subquery with FOR SHARE/UPDATE clause are sent to slave/standby(Tatsuo Ishii) - Fix bug which rewriting timestamp of default value fails in PREPARE statements. This used to work but was broken in 3.0(Toshihiro Kitagawa) - Fix fail to compile pcp commands on the environment without getopt_long()(Tatsuo Ishii) - Fix crash of pgpool child when frontend connects if in raw mode, enable_hba is off and more than 2 backends(Toshihiro Kitagawa) - Fix some memory leaks(Toshihiro Kitagawa) __________________________________________________________________ * Enhancements - Enhance online recovery in streaming replication mode. Now restarting pgpool-II children is avoided when recovery finished. So existing sessions can be continued while doing online recovery(Tatsuo Ishii) - pcp_attach_node does not diconnect existing sessions in streaming replication mode. In other mode, pcp_attache_node still disconnects existing sessions(Tatsuo Ishii). - Import PostgreSQL 9.0 parser. This allows to use CREATE INDEX with implicit index name, which is new in 9.0. Patch contributed by Akio Ishida. - Allow to use regular expressions in black and white function list. Patch contributed by Gilles Darold. Patch reviewed by Guillaume Lelarge. - Reorganize pgpool.conf sample files so that they are easier to read (Guillaume Lelarge) - Add tags into all parameters in the pgpool-II user manual(Haruka Takatsuka) - Enhance online recovery documents in streaming replication(Tatsuo Ishii) - Change the function to check the replication delay in streaming replication mode. Currently, pgpool uses pg_last_xlog_replay_location() instead of pg_last_xlog_receive_location(). Fix suggested by Anton Yuzhaninov (Tatsuo Ishii) - Allow time stamp rewriting to work with arbitrary expression in default value of a column. Before we detected anything including now() then simply replaced it to now(). This will lead to wrong rewriting of default value. for example, timezone('utc'::text, now()). Note that, however, this only adopts to simple queries. Extended protocols(for example Java, PHP PDO) or SQL "PREPARE" still remain same(Tatsuo Ishii) - Enhance the error message which is emitted when failed to check replication delay(Nicolas Thauvin) - Change error message "do_md5: read_password_packet failed" into debug level(Toshihiro Kitagawa) - Allow to compile pgpool-regclass() against PostgreSQL 9.1(Tatsuo Ishii) - Update and sync pgpool-II manuals of English version and Japanese version(Tatsuo Ishii) =============================================================================== 3.0 Series (2013/07/30 - ) =============================================================================== 3.0.17 (umiyameboshi) 2015/02/05 * Version 3.0.17 This is a bugfix release against pgpool-II 3.0.16. __________________________________________________________________ * Bug fixes - doc: Describe explicitly that the number of slave nodes is not necessarily 1 (Tatsuo Ishii) - Fix uninitialized variable (Tatsuo Ishii) Per Coverity 1234603. - doc: Fix missing release note entries in the previous release (Tatsuo Ishii) - Fix node id range check bug in trigger_failover_command(). (Tatsuo Ishii) The node id should be lower than NUM_BACKENDS. Probably harmless since callers never pass node ids greater or equal to NUM_BACKENDS. - Fix to disable debug mode by reloading config (Yugo Nagata) Per bug #114 http://www.pgpool.net/mantisbt/view.php?id=114 =============================================================================== 3.0.16 (umiyameboshi) 2014/09/05 * Version 3.0.16 This is a bugfix release against pgpool-II 3.0.15. __________________________________________________________________ * Bug fixes - Fix a typo of pgpool.spec (Yugo Nagata) - Fix bug that worker child process keeps failing when there's no primary backend (Tatsuo Ishii) Problem identified and fix contributed by Junegunn Choi. See [pgpool-hackers: 471] for more details. - Close listen socket when smart shutdown request is made (Tatsuo Ishii) When smart shutdown process starts, pgpool children still listen on the port and clients can send further connection requests which fail in the end. Which is not only waste of time, but also prevents a load balancer which sits in front of pgpool from realizing the pgpool is going down. Problem analyzed and patch provided by Junegunn Choi in [pgpool-hackers 474], and enhanced to take care not only inet domain socket but UNIX domain socket by Tatsuo Ishii. - doc: Add cautions that recovery commands are killed by statement_timeout of PostgreSQL. (Tatsuo Ishii) - doc: Remove old restriction description which is no longer true (Tatsuo Ishii) - Fix return type of text_to_lsn() function (Yugo Nagata) This caused compile warning. - Fix file descriptor leak when daemonize. (Tatsuo Ishii) Per Coverity 1111471. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111442. - Fix pgpool.init's long-standing bug of stop/restart failure (Yugo Nagata) In previous, pgpool.init uses killproc for stopping pgpool, but there are several problems. In the new version, "pgpool -m fast stop" is used in stop command. Original patch contributed by Ryan DeShone and modified by Yugo Nagata. See [pgpool-hackers: 239][pgpool-hackers: 512]. - Disbale statement_timeout of PostgreSQL while executing online recovery (Tatsuo Ishii) Online recovery may take very long time and user may enable statement timeout. To prevent online recovery canceled by statement timeout, disable statement timeout in the connection used by online recovery. See [pgpool-general: 2919] for more details. - Remove unnecessary call to pool_shmem_exit() which removes semaphore when it shouldn't (Tatsuo Ishii) exit_handler checks if the process is parent or not. This is good. However, even if it is a child process, it calls pool_shmem_exit() which removes semaphore and shmem when it should not. It should be called only from parent process. Per bug #102. http://www.pgpool.net/mantisbt/view.php?id=102 =============================================================================== 3.0.15 (umiyameboshi) 2014/03/24 * Version 3.0.15 This is a bugfix release against pgpool-II 3.0.14. __________________________________________________________________ * Bug fixes - doc: Add mention about "listen queue" and how to increase the "backlog" in the num_init_children section. (Tatsuo Ishii) - Fix bad performance of unlogged tables detection code. (Tatsuo Ishii) Pointed out at [pgpool-hackers: 435][[pgpool-general:2325]. - Fix primary node detection logic. (Tatsuo Ishii) There's a possibility that primary node is not detected. This happens in following situation. node 0: primary, node 1: standby. Node 0 goes down. Health checking detects the fact but local status is not updated yet. Primary node finding (find_primary_node) runs. Node 0's status is yet healthy. Because find_primary_node fails to connect to node 0, it immediately returns -1 and fails to find that fact that node 1 is now primary. Fix is just continuing to look for primary node when fails to connect to a node. Per [pgpool-general: 2409]. - Fix jdbc DML fails when operated in raw mode and auto commit is off. (Tatsuo Ishii) This is reported in bug #92. - Avoid to send queries to unrelated nodes in streaming replication mode. (Tatsuo Ishii) Pgpool-II sends certain queries, such as BEGIN, END and SET commands to all of DB nodes. However in streaming replication mode, only primary node and at most one standby node are only concerned (if primacy node is selected as the load balance node, only 1 node is concerned). See [pgpool-hackers: 464] for more details. - Fix possible buffer overrun problem and memory leak. (Tatsuo Ishii) Per Coverity 1111465 and 1111482. - Fix freeing NULL. (Tatsuo Ishii) Per Coverity 1111384. - Fix memory leak. (Tatsuo Ishii) Per Coverity 1111446. - Fix a segmentation fault in parallel mode with system_db_hostname is empty (Yugo Nagata) - Fix to output debug messages in processing pgpool.conf with -d option (Yugo Nagata) Previously, some debug messages in pool_get_config() ware not output even when -d option was used. - Fix JDBC exception of prepared statement including now() in replication mode (Yugo Nagata) With JDBC, when a prepared statement is executed more than PrepareThreshold times, the statement is named and Describe message is sent after Parse. With named statement, pgpool rewrite now() to parameter in replication mode. Hence, rewritten query has additonal parameter than original. In this case, ParameterDescription message sent to frontend (response of Describe) should include OIDs of the same number os original query's parameters. Otherwize, JDBC throws ArrayIndexoutOfBoundsException. This is reported in [pgpool-general-jp: 1192]. - Fix backend error of prepared statement about table which has column whoes default value is now() in replication mode (Yugo Nagata) When pgpool parses a named prepared statement with default now(), timestamps are replaced to additional parameters. So, Bind message also should inclued additional parameter format codes. However, when the number of original parameter was one, pgpool didn't handle this. This caused a error like "incorrect binary data format in bind parameter 2". - doc: Add description about parallel mode doesn't support PREPARE (Yugo Nagata) Per bug #93 =============================================================================== 3.0.14 (umiyameboshi) 2013/12/06 * Version 3.0.14 This is a bugfix release against pgpool-II 3.0.13. __________________________________________________________________ * Bug fixes - Fix incorrect time stamp rewriting in replication mode for certain time zones. (Tatsuo Ishii) Time stamp rewriting calls "SELECT now()" to get current time. Unfortunately the buffer for the current time is too small for certain time zones such as "02:30". Note that non-30-minutes-time-zone such as "0900" does not reveal the problem. This explains why we haven't the bug report until today. Bug reported in [pgpool-general: 2113] and fix provided by Sean Hogan. http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002142.html - Makefile: Replace pg_config by $(PG_CONFIG) in Makefiles so it can be overridden at build time when compiling for different PG major versions. (Tatsuo Ishii) Patch contributed by Christoph Berg ([pgpool-general: 2127]). http://www.sraoss.jp/pipermail/pgpool-general/2013-September/002156.html - configure: Remove -lcompat because it confuses FreeBSD per bug#15. (Tatsuo Ishii) http://www.pgpool.net/mantisbt/view.php?id=15 - Fix segfault when pgpool.conf does not set log_standby_delay. (Tatsuo Ishii) This is caused by wrong initialization for log_standby_delay in pool_config.l. Per bug#74. http://www.pgpool.net/mantisbt/view.php?id=74 - doc: Modify descriptions about restrictions of parallel mode Muliple rows INSERT using VALUES are not supported in parallel mode. (Yugo Nagata) - Avoid calling find_primary_node_repeatedly() when standby node goes down. (Tatsuo Ishii) This will reduce the time to failover. Per bug #75, patch modified by Tatsuo Ishii. http://www.pgpool.net/mantisbt/view.php?id=75 - Fix error message in read_password_packet(). (Tatsuo Ishii) - Fix memory leak when do_query() fails in timestamp rewriting. (Tatsuo Ishii) For this purpose free_select_result() is changed to accept NULL argument. Per Coverity report "1111454, 1111455 Resource leak". - Fix target node selection logic when "DEALLOCATE portal|statement". (Tatsuo Ishii) When "DEALLOCATE portal|statement" is used and last prepared statement or portal was not found, target node selection map is not set. Probably this is not actually harmful because prepared statement or portal was not found is an error case. The bug was there since day 0. Per Coverity report "1111491 Structurally dead code". - Fix range check bug of MAX_NUM_BACKENDS in corner case. (Tatsuo Ishii) MAX_NUM_BACKENDS is the allowed max number of DB nodes (128, at this point). In reality, probably no one ever tried more than 128 DB nodes and that's the reason why nobody noticed. Per Coverity report "1111429, 1111430 and 1111431 Out-of-bounds write". - Fix that the script forgets to allow public access to pgpool_catalog. (Tatsuo Ishii) The bug prevents inserting data into user tables if pgpool_catalog is created in native replication mode. The bug was there from day 1. I wonder why nobody noticed until today. Per [pgpool-general-jp: 1229]. http://www.sraoss.jp/pipermail/pgpool-general-jp/2013-November/001228.html - Fix uninitialized variable in error case in pool_do_auth(). (Tatsuo Ishii) If there's no valid backend, pgpool will return garbage pid to frontend in auth phase. Actually because no backend is available, frontend will be disconnected later on. So this is not harmless. Per Coverity report "1127331 Uninitialized scalar variable". - Fix to add node id range check when issue an error message using node id. (Tatsuo Ishii) Per Coverity report #1111433 "Out-of-bounds read". - Fix buffer overrun bug and resource leak bug of parse_copy_data(). (Tatsuo Ishii) Per Coverity report 1111427 "Out-of-bounds write" and 1111453 "Resource leak". - Fix possible segfault in CopyDataRaws(). (Tatsuo Ishii) Coverity pointed out that if pool_get_id() returns an error, VALID_BACKEND will access out of array. Per Coverity report 1111413 "Memory - illegal accesses". - Fix strftime() usage in pool_pools(). (Tatsuo Ishii) The buffer is not large enough as expected by the second parameter. This is not harmless because the format string will not produce longer result string than the buffer. Per Coverity report 1111426 "Out-of-bounds access". - Fix resource leak in make_persistent_db_connection. (Tatsio Ishii) For this pupose, new static function free_persisten_db_connection_memory is added. Per Coverity report #1111468. =============================================================================== 3.0.13 (umiyameboshi) 2013/09/06 * Version 3.0.13 This is a bugfix release against pgpool-II 3.0.12. __________________________________________________________________ * Bug fixes - Fix a mistake in ssh command of doc/basebackup.sh (Tatsuo Ishii) - Fix a bug in parsing prepared statements with transaction handling in replication mode (Tatsuo Ishii) Parse() automatically starts a transaction for non SELECT query to keep consistency among nodes in replication mode. But this wasn't closed. If wrong query comes in, the transaction goes into an abort state but pgpool does not close the transaction. Thus next query causes error because the transaction is still in abort status. This problem was reported in [pgpool-general: 1877] by Sean Hogan. [pgpool-general: 1877] current transaction is aborted, commands ignored http://www.sraoss.jp/pipermail/pgpool-general/2013-July/001905.html - Fix typos of the japanese document (Yugo Nagata) =============================================================================== 3.0.12 (umiyameboshi) 2013/07/10 * Version 3.0.12 This is a bugfix release against pgpool-II 3.0.11. __________________________________________________________________ * Bug fixes - Add mention about "-D" option to the man page. (Tatsuo Ishii) - Consider timeout waiting for compeletion of failback request in on line recovery (Tatsuo Ishii) This will prevent the situation that recovery operation continues forever and we cannot even shutdown pgpool-II main process. This could happen especially while executing follow master command. - Fix do_query() to not hang when PostgreSQL returns an error (Tatsuo Ishii) The typical symptom is "I see SELECT is keep on running according to pg_stat_activity". To fix this pgpool-II just exits the process and kill the existig connection. This is not gentle but at this point I believe this is the best solution. - Fix bug with do_query which causes hung in extended protocol (Tatsuo Ishii) This problem could occur when insert lock is enabled and pgpool_catalog.insert_lock exists, See [pgpool-general: 1684] for more details. [pgpool-general: 1684] insert_lock hangs http://www.sraoss.jp/pipermail/pgpool-general/2013-May/001711.html - Fix unnecessary degeneration caused by error on commit (Tatsuo Ishii) In master slave mode, if master gets an error at commit, while other slaves are normal at commit, we don't need to degenrate any backend because it is likely that the "kind mismatch error" was caused by a deferred trigger. - Fix to register pgpool_regclass in pg_catalog schema (Tatsuo Ishii) This is necessary to deal with clients which restricts schema search path to pg_catalog only. Postgres_fdw is such a client. - Fix a potential crash in pg_md5 command (Muhammad Usama) - Fix a segmentation fault of a child process that occurs when a startup packet has no PostgreSQL user information (Yugo Nagata) You can reproduce it by $ psql -p 9999 -U '' If enable_pool_hba is on, a child process terminates by segmentation fault. Otherwise if enable_pool_hba is off, the error message is ERROR: pool_discard_cp: cannot get connection pool for user (null) database (null) In both cases, psql terminates with no message on frontend. In the fixed version, if PostgreSQL user is not specified in startup packet, the message as following is output to both log and frontend. This is the same behavior as PostgreSQL. FATAL: no PostgreSQL user name specified in startup packet - Move ssl_ca_cert and ssl_ca_cert_dir descriptions to the SSL section (Yugo Nagata) - Add ssl_ca_cert and ssl_ca_cert_dir descriptions to the japanese document (Yugo Nagata) - Fix to verify the backend node number in pcp_recovery_node (Yugo Nagata) When an invalid number is used, null value is passed as an arguments of recovery script, and this causes a malfunction. In especially, rsync may delete unrelated files in basebackup scripts. - Fix reset query stuck problem. (Tatsuo Ishii) It is reported that reset query (DISCARD ALL etc.) occasionally does not finish and pgpool child remain waiting for reply from backend thus client cannot connect to pgpool. The cause of problem is not identified yet but if client suddenly closes connection to pgpool in the middle of query processing, backend may not accept the reset queries because they are not ready for query. The fix is, if frontend closes connection in unexpected way, query process loop immediately returns with new state: POOL_END_WITH_FRONTEND_ERROR and pgpool closes connection to PostgreSQL then goes back to new connection request waiting loop. Also, pgpool closes connections to backend when client_idle_limit is set and the idle limit. Per bug #107. http://www.pgpool.net/mantisbt/view.php?id=107 =============================================================================== 3.0.11 (umiyameboshi) 2013/04/26 * Version 3.0.11 This is a bugfix release against pgpool-II 3.0.10. __________________________________________________________________ * Buf fixes - Fix to show pool_passwd in "SHOW pool_status". (Yugo Nagata) - Fix long standing bug with timestamp rewriting code for processing extended protocol. (Tatsuo Ishii) Parse() allocate memory using palloc() while rewriting the parse message. Problem is, the rewritten message was kept in the data which is managed by pool_create_sent_message() etc. The function assumes that all the data is in session context memory. However, palloc() allocates memory in query context of course, and gets freeed later on when the query context disappears. And the function tries to free the memory as well, which causes various problems, including segfault and double free. To fix this, memory to store rewritten message is allocated using session context. The bug was there since pgpool-II 3.0 was born. Problem analysis and patch contributed by Naoya Anzai. [pgpoolgenera-jp: 1146]. (in Japanese) http://www.pgpool.net/pipermail/pgpool-general-jp/2013-March/001145.html - Fix bug with md5 auth long user name handling. (Tatsuo Ishii) If user name is longer than 32 bytes, md5 authentication doesn't work. Problem reported in [pgpool-general: 1526] by Thomas Martin. [pgpool-general: 1526] [pgPool-II 3.2.3] MD5 authentication and username longer than 32 characters. http://www.pgpool.net/pipermail/pgpool-general/2013-March/001551.html - Fix to calculate replication delay only if standby server is behind from the primay server. (Yugo Nagata) When the primary server is behind from standby server, negative value of delay is calculated and the value is assigned to unsigned variable. It causes a log message informing negative replication delay. And what is worse, it also causes SELECT queries to be sent to the primary in load balance even though there are no replication delay in fact. The problem is reported and analyzed by Saitoh Hidenori in [pgpool-genera-jp: 1145]. [pgpool-general-jp: 1145] (in Japanese) http://www.pgpool.net/pipermail/pgpool-general-jp/2013-March/001144.html - pgpool-recovery adopts PostgreSQL 9.3. (Tatsuo Ishii) Patch contributed by Asif Rehman. Slight editing by Tatsuo Ishii. [pgpool-hackers: 180] compile error in ppool-recovery http://www.pgpool.net/pipermail/pgpool-hackers/2013-April/000179.html - Fix pool_has_pgpool_regclass() to check execute privilege of pgpool_regclass(). (Tatsuo Ishii) Even though pgpool_regclass() exists, if pgpool cannot execute the function, the connection to backend hangs. You can reproduce the problem by just dropping the execute privilege from pgpool_regclass and do some insert in native replication mode. The problem is reported in bugtrack #53. #53 pgpool_regclas hangs all connections Date: 2013-04-04 13:35 Reporter: tmandke http://www.pgpool.net/mantisbt/view.php?id=53 - Fix error message mistakes in detect_postmaster_down_error(). (Tatsuo Ishii) For example, "LOG: detect_stop_postmaster_error: detect_error error" is fixed to "LOG: detect_postmaster_down_error: detect_error error", and so on. =============================================================================== 3.0.10 (umiyameboshi) 2013/02/08 * Version 3.0.10 This is a bugfix release against pgpool-II 3.0.9. __________________________________________________________________ * General - Fix race condition when using md5 authentication. (Tatsuo Ishii) The file descriptor to pool_passwd is opened in pgpool main and pgpool child inherits it. When concurrent connections try to authenticate md5 method, they call pool_get_passwd and seek the fd and cause random md5 auth failure because underlying fd is shared. Fix is, let individual pgpool child open the file by calling pool_reopen_passwd_file. Problem reported and analyzed by Jason Slagle in pgpool-general:1141. [pgpool-general: 1141] Possible race condition in pool_get_passwd From: Jason Slagle Date: Sun, 28 Oct 2012 01:12:52 -0400 http://www.sraoss.jp/pipermail/pgpool-general/2012-October/001160.html - Fix pool_send_severity_message() not to use uninitialized memory. (Tatsuo Ishii) It cause a segmentaion fault. Reported in Bug #33's attached valgrind output by dudee. #33 pgpool-II 3.2.1 segfault Reporter: dudee Date: 2012-10-30 19:16 http://www.pgpool.net/mantisbt/view.php?id=33 - Fix reaper() not to exit wait3() loop when catches pcp or worker child exit event. (Tatsuo Ishii) Otherwise reaper() mistakenly ignore some process exit event and make a risk of creating zombie process and forgetting to create new process. Problem reported and fix suggested by Goto in [pgpool-general-jp: 1123]. http://www.sraoss.jp/pipermail/pgpool-general-jp/2012-November/001122.html (in Japanese) - Fix pool_search_relcache() to use MASTER or MASTER_NODE_ID macro, rather than REAL_MASTER_NODE_ID. (Tatsuo Ishii) In case node 0 fail back in streaming replication mode, pgpool does not restart child process. So REAL_MASTER_NODE_ID looks for node 0 con info, which is not present until new connection to backend made. Thus referring to node con info results in segfault. MASTER or MASTER_NODE_ID are safe in this situation because they look at cached former master node id. - Fix long standing bug "portal not found" error when replication delay is too much in streaming replication mode. (Tatsuo Ishii) The bug had been there since the delay threshold was introduced. We changed destination DB node if delay threshold exceeds in bind, describe and execute. However, if parse sends to different node, bind, describe or execute will fail because no parsed statement or portal exists. Solution is, not to send to different parse node even if delay threshold is too much. - Fix pg_md5 to output "\n" after user inputs password. (Yugo Nagata) - Fix child_exit() to not call send_frontend_exits() if there's no connection pool. (Tatsuo Ishii) Otherwise, it segfaults because send_frontend_exits() referes to objects pointed to by pool_connection_pool. Per bug track #44 by tuomas. #44 pgpool went haywire after slave shutdown triggering master failover Reporter: tuomas Date: 2012-12-11 00:33 http://www.pgpool.net/mantisbt/view.php?id=44 - Fix read_startup_packet() to reset alarm and free StartupPacket when pool_read() returns 0 which means incorrect packet length. (Nozomi Anzai) Previously, authentication timeout occurs when connected by a program monitoring the pgpool port.It is reported in bug track #35 by tuomas. #35 Authentication is timeout Reporter: tuomas Date: 2012-11-20 11:54 http://www.pgpool.net/mantisbt/view.php?id=35 - Fix long standing bug with pool_open(). (Tatsuo Ishii) It initializes wrong buffer pointer. Actually this is harmless because the pointer is initialized by prior memset() call, though. - Add a description about "-f" to help message. (Tatsuo Ishii) - Modify documents to correct information of whether a certain parameter change requires restart. (Yugo Nagata) - Add pool_passwd option to pgpool.conf.sample*, and documents. (Yugo Nagata) =============================================================================== 3.0.9 (umiyameboshi) 2012/10/12 * Version 3.0.9 This is a bugfix release against pgpool-II 3.0.8. __________________________________________________________________ * General - Fix read_startup_packet. (Tatsuo Ishii) If packet length is lower than 0, it should have returned immediately. Otherwise it would cause memory allocation error later on. per pgpool-general:886. Also add canceling alarm. - Add NOTICE message handling to s_do_auth. (Tatsuo Ishii) Without this, health check responses false alarm and causes failover. per bug track: http://www.pgpool.net/mantisbt/view.php?id=25 Also allow to receive ready for query packet *not* right after backend keydata. I'm not sure if this could happen in the real world but the protocol seems to allow this. - Remove unnecessary/confusing debug log from s_do_auth.(Tatsuo Ishii) - Fix inifinit loop in SSL mode. When there's pending data in SSL layer of frontend, pool_process_query() checks pending data in backend. (Tatsuo Ishii) If there's non, it loops again and checks frontend/backend receive buffer by using is_cache_empty(). Unfortunately it first checks pending data in SSL layer of frontend, thus goes to backend data and checks again (infinite loop). The solution is, if there's pending data in SSL layer of frontend and query is not in progress, call ProcessFrontendResponse() to process new request from frontend. - Fix is_system_catalog to use pgpool_regclass if available. (Tatsuo Ishii) - Fix memory leak in pool_get_insert_table_name(). (Tatsuo Ishii) If session context's memory contex is used for nodeToString(), memory is not freed until session ends. - Fix possible memory leak in nodeToString(). (Tatsuo Ishii) Since the memory context could be session context, memory used for String object may be considerable if user session continues for hours or days. See bug track #24 for more details. - Fix long standing problem with do_query(). (Tatsuo Ishii) When 1) extended protocol used and 2)unnamed portal is used and 3) no explicit transaction is used, user's unnamed portal is removed by Sync message. This is because Sync message closes transaction and unnamed portal is removed. This leads to "portal "" does not exist" error. Fix is, use "Flush" message instead of Sync. Main difference between using Sync and Flush is, Flush does not return Ready for Query message. So do_query() does not return until all expected responses are returned. It seems the order of messages returned from backend is random, and do_query () manages it by using state bits. =============================================================================== 3.0.8 (umiyameboshi) 2012/08/06 * Version 3.0.8 This is a bugfix release against pgpool-II 3.0.7. __________________________________________________________________ * General - Adopt PostgreSQL 9.2. (Tatsuo Ishii) __________________________________________________________________ * Bug fixes - Fix load balance in Solaris. (Tatsuo Ishii) Problem is, random() in using random() in Solaris results in strange load balancing calculation. Use srand()/rand() instead although they produce lesser quality random Problem reported at [pgpool-general: 396]. [pgpool-general: 396] strange load balancing issue in Solaris http://www.pgpool.net/pipermail/pgpool-general/2012-April/000397.html - Fix segfault of pcp_systemdb_info not in parallel mode. (Nozomi Anzai) - Fix "unnamed prepared statment does not exist" error. (Tatsuo Ishii) This is caused by pgpool's internal query, which breaks client's unnamed statements. To fix this, if extended query is used, named statement/portal for internal are used for internal query. - Fix is_system_catalog(). Its relcach was accidently defined as "session local". (Tatsuo Ishii) - Improve reading and writing pid_file. (Tatsuo Ishii) - Fix pool_process_query() bug reported in [pgpool-general: 672]. (Tatsuo Ishii) This is caused by the function waits for primary node which does not have pending data, while standbys have pending data. [pgpool-general: 672] Transaction never finishes http://www.pgpool.net/pipermail/pgpool-general/2012-June/000676.html - Fix wait_for_query_response() not to send param status to frontend if frontend is NULL. (Tatsuo Ishii) This could happen while processing reset_query_list and occur crash. =============================================================================== 3.0.7 (umiyameboshi) 2012/04/23 * Version 3.0.7 This is a bugfix release against pgpool-II 3.0.6. __________________________________________________________________ * Bug fixes - Add m4 files. This should prevent compiling problem on older OS's. (Tatsuo Ishii) - Fix bug that the process exits before unlocking semaphore by a signal interrupt. (Tatsuo Ishii) - Fix a memory leak in case of reset_query. (Tatsuo Ishii) - Fix SimpleQuery() so that it restores parser memory context when: 1) Builtin show commands are used 2) Parallel query mode 3) Query cache is used (Tatsuo Ishii) - Fix pool_ssl_read() to deal with large data reading. (Tatsuo Ishii) Original complain is here: http://www.pgpool.net/pipermail/pgpool-general/2012-March/000299.html - Fix hangup when PREPARE statement causes error. (Toshihiro Kitagawa) This issue was reported by Tomonari Katsumata: Subject: [pgpool-general: 121] question of pgpool's behavior =============================================================================== 3.0.6 (umiyameboshi) 2012/01/31 * Version 3.0.6 This version fixes various bugs since 3.0.5. __________________________________________________________________ * Bug fixes - Fix infinite loop reported in this thread (Tatsuo Ishii): http://www.pgpool.net/pipermail/pgpool-general/2011-December/000099.html It was not considered the case that, when received buffer in primary was empty but the one in a standby was not, the standby spontaneously sent packet to pgpool. This could happen when, for example, reloading postgresql.conf. The fix is that such buffer in standby is discarded. =============================================================================== 3.0.5 (umiyameboshi) 2011/10/31 * Version 3.0.5 This version fixes various bugs since 3.0.4. __________________________________________________________________ * Bug fixes - Fix bug with the handling of empty queries. The empty queries come to be handled the same as SELECT queries. This fix allows load-balance after the empty query(Toshihiro Kitagawa) - Fix insert_lock so that it works correctly even if the column definition such as "DEFAULT nextval(('"x_seq"'::text)::regclass)" (Toshihiro Kitagawa) - Fix log message which is emitted when pgpool-II cannot parse the query in the extended query protocol so that it shows the query (Toshihiro Kitagawa) - Fix description about backend_weight inpgpool-II manual. It can be changed by reloading pgpool.conf(Tatsuo Ishii) - Fix bug which does not update the node status when reattaching the node in raw mode(Guillaume Lelarge) - Fix bug that SELECTs which have subquery with FOR SHARE/UPDATE clause are sent to slave/standby(Tatsuo Ishii) - Fix bug which rewriting timestamp of default value fails in PREPARE statements. This used to work but was broken in 3.0(Toshihiro Kitagawa) - Fix crash of pgpool child when frontend connects if in raw mode, enable_hba is off and more than 2 backends(Toshihiro Kitagawa) - Fix some memory leaks(Toshihiro Kitagawa) __________________________________________________________________ * Enhancements - Allow time stamp rewriting to work with arbitrary expression in default value of a column. Before we detected anything including now() then simply replaced it to now(). This will lead to wrong rewriting of default value. for example, timezone('utc'::text, now()). Note that, however, this only adopts to simple queries. Extended protocols(for example Java, PHP PDO) or SQL "PREPARE" still remain same(Tatsuo Ishii) - Change error message "do_md5: read_password_packet failed" into debug level(Toshihiro Kitagawa) =============================================================================== 3.0.4 (umiyameboshi) 2011/06/01 * Version 3.0.4 This version fixes various bugs since 3.0.3. __________________________________________________________________ * Incompatible changes - In streaming replication, if delay_threshold is 0 or health checking is disabled, the delay checking is not performed. This is the behaviour according to a description of the pgpool-II manual. But, so far the delay checking was performed even if health checking was disabled(Guillaume Lelarge) __________________________________________________________________ * Bug fixes - Fix pgpool-regclass() to be compiled in PostgreSQL 8.0 or later. 7.4 still produces errors(Tatsuo Ishii) - Fix possible hangup when using /*NO LOAD BALANCE*/ comment in streaming replication(Toshihiro Kitagawa) - Fix hangup when received Flush(H) message or CloseComplete(C) message(Toshihiro Kitagawa) - Fix possible hangup that happen for the receiving timing of ReadyForQuery(Z) message after pgpool-II connects to backends(Toshihiro Kitagawa) - Add description about parameters for recovery_1st_stage_command and recovery_2nd_stage_command(Tatsuo Ishii) - Increase size of the internal system catalog cache from 32 to 128. This has the effect of reducing "unnamed prepared statement does not exist" error(Tatsuo, Kitagawa) - Fix bug with pcp_connect() which causes double free. Patch contributed by Jehan-Guillaume (ioguix) de Rorthais(Tatsuo Ishii) - Fix bug with start_recoery() which is apparently wrong usage of PQfinish()(Tatsuo Ishii) - Fix incorrect error message which is sent to the frontend when client idle time reached client_idle_limit(Tatsuo Ishii) - Fix "backend status" variable name correctly in pool_status. Replace the space with a '_'(Guillaume Lelarge) - Fix hangup when using md5 authentication method and running as daemon. Patch contributed by Nicolas Thauvin(Tatsuo Ishii) - Fix log_per_node_statement so that it prints statements in the extended query protocol. This used to work but was broken in 3.0 (Toshihiro Kitagawa) __________________________________________________________________ * Enhancements - Add currval() and lastval() to black_function_list of sample configuration files. If they are load balanced, currval() or lastval() may be called before the result of nextval() or setval() is propagated to slaves(Tatsuo Ishii) =============================================================================== 3.0.3 (umiyameboshi) 2011/02/23 * Version 3.0.3 This version fixes various bugs since 3.0.1. Please note that 3.0.2 was canceled due to a packaging problem. __________________________________________________________________ * Incompatible changes - Now installing C function "pgpool_walrecrunning()" is recommended if you plan to use streaming replication mode. This is necessary for better use of online recovery in the mode. Also new variable "%P" can be used in the online recovery script. If you do not install the function, these functionalities cannot be used(Tatsuo Ishii). - In raw mode if there's only one DB node and if a problem arises with the DB node, it will be brought to down status. However if the DB node goes into good condition again, you can use the DB node without restarting pgpool. This change has been included in 3.0, but did not work(Tatsuo, Kitagawa) __________________________________________________________________ * Bug fixes - Fix non portable code in password authentication. Bug report from a FreeBSD user(Tatsuo Ishii) - Fix bug that insert_lock locks all rows in user table (Tatsuo, Kitagawa) - Fix bug with password authentication. If user name is 32 bytes long, pgpool child segfaults.(Tatsuo Ishii) - Fix bug with md5 authentication. If raw mode or number of backend is 1, pgpool child segfaults. Patch contributed by Rob Shepherd(Tatsuo Ishii) - Fix long standing bug with timestamp rewriting against array and complex types. Patch contributed by Akio Ishida(Tatsuo Ishii) - Fix bug that debug_level directive doesn't work. Patch contributed by Gilles Darold(Tatsuo Ishii) - Fix possible crash of pgpool child while doing failover(Toshihiro Kitagawa) - Fix white/black_function_list so that it works correctly when user calls function with schema name(Tatsuo Ishii) - Fix bug that DROP DATABASE fails by connection cache(Toshihiro Kitagawa) - Fix bug that failover fails in raw mode(Toshihiro Kitagawa) - Fix possible termination of pgpool child when both simple query protocol and extended query protocol are used in one session (Toshihiro Kitagawa) - Fix possible hang up when an error occurs while using extended query protocol(Toshihiro Kitagawa) - Fix pgpool-regclass() so that it doesn't use PG_TRY/CATCH. It appeared that using PG_TRY/CATCH is not safe, sometimes backend dies with PANIC: ERRORDATA_STACK_SIZE exceeded.(Tatsuo Ishii) - Fix bug that select query isn't sent to master node when it meets the following conditions(Toshihiro Kitagawa) - in MASTER/SLAVE mode - use extended query protocol - started transaction explicitly - after write queries - Fix bug with load_balance that JDBC driver sends BEGIN to master node many times(Toshihiro Kitagawa) - Fix pool_status so that failback_command and fail_over_on_backend_error show correct values(Toshihiro Kitagawa) - Remove parameters from pool_status: recovery_password, system_db_password(Toshihiro Kitagawa) - Fix online recovery problem in the streaming replication mode(Tatsuo Ishii). Consider following scenario. Suppose node 0 is the initial primary server and 1 is the initial standby server. 1) Node 0 going down and node 1 promotes to new primary. 2) Recover node 0 as new standby. 3) pgpool-II assumes that node 0 is the new primary. This problem happens because pgpool-II regarded unconditionally the youngest node to be the primary. pgpool-II 3.0.3 now checks each node by using pgpool_walrecrunning() to see if it is a actually primary or not and is able to avoid the problem and regards node as standby correctly. Also you can use new variable "%P" to be used in the recovery script. If you do not install the function, the above problem is not resolved. - Fix backend complaining "unexpected EOF on client connection" while doing failover in streaming replication mode(Tatsuo Ishii) - Fix pgpool crashes when all backends go down(Tatsuo Ishii) - Fix replication delay checking so that it does not keep persistent connection to backends. Because the persistent connection may become bogus if a node down and then wake up between replication delay checking period(Tatsuo Ishii) - Rewrite and review english document(Marc Cousin, Gleu) __________________________________________________________________ * Enhancements - Emit log if particular backend is down status while reading the status file(Tatsuo Ishii) - Emit error message if an error occurred by the query that pgpool executed(Tatsuo Ishii) - Add sql directories main Makefile(Tatsuo Ishii) =============================================================================== 3.0.1 (umiyameboshi) 2010/10/19 * Version 3.0.1 This version fixes various bug in 3.0. __________________________________________________________________ * Bug fixes - Fix bug with md5 auth. If there's more than 1 servers to be authenticated, it segfaults(Tatsuo Ishii) - Fix bug that a child process crashes when clients execute a query contains syntax error in extended query protocol(Toshihiro Kitagawa) - Fix bug with handling of portal information, it terminates a child process(Toshihiro Kitagawa) - Fix hungup when a query sent to one node caused an error in extended query protocol(Toshihiro Kitagawa) - Fix typo in English doc. Patch contributed by Asaf Ohaion(Tatsuo Ishii) =============================================================================== 3.0 (umiyameboshi) 2010/09/10 * Version 3.0 This is the first version of pgpool-II 3.0 series. That is, a "major version up" from 2.2 or 2.3 series. The biggest news is, this version adapts to PostgreSQL 9.0's new feature: Streaming Replication/Hot Standby. Streaming replication can be used as a sub mode of master slave mode. Master slave mode itself heavily enhanced: - SELECTs in explicit transactions can be load balanced - In extended protocol, PARSE/BIND/DESCRIBE messages are sent to the node which execute EXECUTE message, not all node. This will reduce lock contentions. - Auto start of transaction happens only when it needed. - Temporary tables can be used safely. - SELECT which calls functions possibly write to database executes on master(primary) Also many new features are added and major refactoring has been made to the internal structure of pgpool-II. For example, in replication mode, SELECTs calling functions possibly write to database will not allow to load balance. __________________________________________________________________ * New features - Online recovery can be used with master/slave/streaming replication mode(Tatsuo Ishii) - New directive "delay_threshold" is added to monitor replication delay in master/slave/streaming replication mode. If replication delay is too much, SELECTs are not load balanced(Tatsuo Ishii) - show pool_status shows replication delay in master/slave/streaming replication mode(Tatsuo Ishii) - New directive "log_standby_delay" is added to control logging of replication delay in master/slave/streaming replication mode(Tatsuo Ishii) - When insert_lock is enabled and the table includes SERIAL data type, issue row lock on the sequence table. Before we issues table lock. Problem is, the table lock conflicts with auto vacuum and sometimes caused excessive lock waiting(Tatsuo Ishii) - Add support for more "SHOW" commands: pool_nodes, pool_processes, pool_pools, and pool_version(Guillaume Lelarge) - Backend process id and whether frontend connects to this connection pool or not are added to pcp_proc_info's output(Tatsuo Ishii) - "Gracefuly detach" option is added to pcp_detatch_node. With this option, pcp_detatch_node waits until all frontends disconnected(Tatsuo Ishii) - New directive "white_function_list" and "black_function_list" are added to register functions those do not or do write to database(Tatsuo Ishii) - In master/slave mode, SELECTs to system catalogs executes only on master/primary(Tatsuo Ishii) - In master/slave mode, SELECTs to temporary table executes only on master/primary(Tatsuo Ishii) - In master/slave mode, write queries outside of explicit transactions no longer trigger to start internal transaction(Tatsuo Ishii) - In master/slave mode, SELECTs inside explicit transactions are load balanced(Tatsuo, Kitagawa) - In master/slave mode, commands are no longer sent to all DB nodes. This will prevent unnecessary locking(Tatsuo, Kitagawa) - New command option adds to ignore the status file when starting up(Tatsuo Ishii) - Supports PostgreSQL 9.0's new VACUUM syntax(Tatsuo Ishii) - New directive "failover_if_affected_tuples_mismatch" controls the behavior when number of result rows of INSERT/UPDATE/DELETE are differ(Tatsuo Ishii) - When number of result rows of INSERT/UPDATE/DELETE are differ, each number are logged(Tatsuo Ishii) - md5 authentication is supported in replication mode and master/slave mode(Tatsuo Ishii) - Allow to force to move to online recovery second stage even there are connecting frontends(Tatsuo Ishii) - If there's only one DB node and it triggers failover, pgpool-II will automatically connects if the DB node coming up(Tatsuo Ishii) - Pcp commands supports long options(Guillaume Lelarge) - New directive "debug_level" added to control the debug message logging(Tatsuo Ishii) - Allow to use various boolean representations as PostgreSQL in pgpool.conf(Toshihiro Kitagawa) - New C language function pgpool_switch_xlog for online recovery added(Toshihiro Kitagawa) - New C language function pgpool_regclass added to avoid a trouble about handling of duplicate table names in different schema(Tatsuo Ishii) __________________________________________________________________ * Bug fixes - Do not rewrite statement which accesses columns having now() etc. as the default value but the data type are not timestamp etc. Otherwise we have an error in DMLs(Tatsuo Ishii) - Fix timestamp rewriting not to omit schema qualification(Tatsuo Ishii) - Fix bug with timeout handling in pcp commands(Tatsuo Ishii) - Fix SSL hang when large amount of data transfered(Tatsuo Ishii) - Fix failover when there's only one DB node(Tatsuo Ishii) - Fix bug with postmaster start check in online recovery. Before it continued infinitely to try to connect to postmaster if the first attempt failed(Tatsuo Ishii) =============================================================================== 2.3 Series (2009/12/07 - 2012/08/06) =============================================================================== 2.3.4 (tomiteboshi) 2012/08/06 * Version 2.3.4 This version fixes various bugs since 2.3.3. __________________________________________________________________ * Bug fixes - Fix do_error_execute_command() so that it discards responses from backend until ErrorResponse received. (Toshihiro Kitagawa) Note that we need to preserve non-error responses (i.e. ReadyForQuery) and put back them using pool_unread(). Otherwise, ReadyForQuery response of DEALLOCATE. This could happen if PHP PDO used. - Fix SimpleForwardToFrontend() so that it reset select_in_transaction flag and execute_select flag when bind error occurred while executing SELECT. (Toshihiro Kitagawa) - Fix SSL connection sometimes hung if lots of data read from backend. This is caused by the buffering in OpenSSL layer. To fix the problem, (Tatsuo Ishii) we check the buffer has any pending data by using SSL_pending() before calling select(2). See thread [Pgpool-general] Fwd: PGPOOL II 2.3.3 hang in ssl mode for more details. - Fix bug with pcp_check_fd()'s timeout handling. (Tatsuo Ishii) Per erboles. Subject: [Pgpool-general] question about pcp_check_fd Date: Sun, 23 May 2010 18:21:41 -0500 To: pgpool - Do not force replication of DEALLOCATE if operated in master/slave mode. Reported by Jan Kantert. (Toshihiro Kitagawa) See: Subject: [Pgpool-hackers] Problems with PgPool 2.3.3 Prepare / Deallocation handling in Master/Slave mode Date: Fri, 28 May 2010 20:59:47 +0200 For more details. - Make timestamp rewriting schema aware. (Tatsuo Ishii) - Do not rewrite statement which accesses columns having now() etc. as the default value but the data type are not timestamp etc. (Tatsuo Ishii) Otherwise we have an error in DMLS. See: Subject: [Pgpool-general] function epoch seems to be causing error To: pgpool-general@pgfoundry.org Date: Mon, 16 Aug 2010 21:48:31 +0000 (UTC) For more details. - Fix insert_lock to be schema aware. (Tatsuo Ishii) - Fix long standing bug with timestamp rewriting against array and complex types. (Tatsuo Ishii) Failed examples are: INSERT INTO r1(col[1], col2.foo) VALUES (1, 2); -- insert_column_item UPDATE r1 SET col1[1] = 1, col2.foo = 1; -- set_target PREPARE "p" (int4[]) AS SELECT $1[1]; -- c_expr SELECT (ARRAY[1,2,3])[1]; SELECT (ARRAY[ARRAY[1]])[1][1]; SELECT ('{1,2,3}'::int[])[1]; SELECT ('{1,2,3}'::int[3])[1]; SELECT r1.col[1], (r1.col1).bar, (r1.col1).* FROM r1; -- columnref SELECT (r1.col1).baz[1], (r1.col1).baz[1][2] FROM r1; Patch provided by Akio Ishida. - Fix buffer overrun problem when pcp password is longer than 32. (Tatsuo Ishii) - Fix wait_for_query_response() not to send param status to frontend if frontend is NULL. This could happen while processing reset_query_list. (Tatsuo Ishii) =============================================================================== 2.3.3 (tomiteboshi) 2010/04/23 * Version 2.3.3 This version fixes various bugs since 2.3.2.2. __________________________________________________________________ * Incompatible changes - Please note that this version of pgpool consumes more shared memory than before. If you encounter problems when starting up pgpool, please look into pgpool log. If you find messages something like "could not create shared memory segment: Cannot allocate memory", please increase the shared memory on your system. - Now parallel mode requires replication mode turned on. parallel mode without replication mode turned on has been broken since pgpool-II was born anyway(Toshihiro Kitagawa) - Change default value for insert_lock to false since there's no point in turning this on in master/slave mode. Fix suggested by Fujii Masao(Tatsuo Ishii) __________________________________________________________________ * Newly added documents - README.online-recovery, which is an internal document of online recovery, added(Tatsuo Ishii) __________________________________________________________________ * Bug fixes - Fix long standing bug since pgpool-II 1.0 which causes segfault of pgpool child process. This was caused by miscalculation of shmem size in pgpool parent. Bug analysis by Kitagawa patch created by Tatsuo - Add restriction for parallel mode(Toshihiro Kitagawa) - The Natural Join is not supported - The USING CLAUSE is converted to ON CLAUSE by query rewrite process - Fix possible crash during rewriting JOIN syntax that have USING in parallel query mode(Toshihiro Kitagawa) This fix is supporting such as following JOIN syntax. example: - SELECT * FROM a JOIN b USING (aid) JOIN c USING (cid); - SELECT * FROM a JOIN b USING (aid) JOIN c USING (cid) JOIN d USING (did) - Fix parallel query so that it can parse INSERT statements that have current_time before a partitioning key column (Toshihiro Kitagawa) - Fix SimpleForwardToBackend() so that pgpool doesn't keep waiting for reply from a backend, when clients using the extended query protocol cause a command error such as bind error. This bug occur in master/slave, raw, and connection pool mode. This fix is, sending SYNC message to recover error after command error(Toshihiro Kitagawa) - Fix SIGINT/SIGQUIT is ignored if pgpool child is in select(). In this case pgpool retries select() thus the signal is ignored(Tatsuo Ishii) - Fix connect_inet_domain_socket_by_port/ connect_unix_domain_socket_by_port so that they check if SIGTERM/SIGINT/SIGQUIT signal has been delivered. Per bug report from Daniel Codina(Tatsuo Ishii) - Fix possible crash during creating "kind mismatch" error message. This used to work but was broken in 2.3.2(Tatsuo Ishii) - Fix bug with healh checking. If a network problem such as unplugged wire happens while calling connect(), health checking does not work since connect_unix_domain_socket()/connect_inet_domain_socket() do retry if connect() is interrupted by ALARM signal. Added new retry argument which controls the retrying behavior to those functions. Per bug report and problem analysis by Daniel Codina(Tatsuo Ishii) - Fix enbug in 2.3.2.2 with time stamp rewriting in SimpleForwardToBackend. Per bug report from Bugtrack #1010771. Report from Peter Pramberge(Tatsuo Ishii) - Fix rewriting "*" in paralell query. Patch conributed by sho-san(Toshihiro Kitagawa) - Fix connect_inet_domain_socket_by_port() so that it print out error message by using hstrerror(), rather than strerror()(Tatsuo Ishii) =============================================================================== 2.3.2.2 (tomiteboshi) 2010/02/22 * Version 2.3.2.2 This version fixes various bug in 2.3.x. - When rewriting timestamp in extended query, it allocated less than what it actually needed. This causes random problematic behavior, typically "message: invalid string in message" error in backend(Tatsuo Ishii). - In extended query, if one of parameters contains NULL, ppool crashes(Tatsuo Ishii). - If in previous status file all nodes were marked down status, it is regarded that this file is bogus. This will prevent "all node down" syndrome(Tatsuo Ishii). =============================================================================== 2.3.2.1 (tomiteboshi) 2010/02/11 * Version 2.3.2.1 This version fixes bug in 2.3.x. It is identfied that pgpool-II 2.3.x has a problem with erroneous query processing(Akio Ishida). =============================================================================== 2.3.2 (tomiteboshi) 2010/02/07 * Version 2.3.2 This version includes various fixes for bugs in 2.3.1 or before. All 2.3.x users are encouraged to upgrade to 2.3.2 as soon as possible. Also this version enables long awaited SSL support and large object replication support. __________________________________________________________________ * Enhancements - SSL support(Sean Finney) - Large object replication support. You need PostgreSQL 8.1 or later, however(Tatsuo Ishii) - Emit statement log when error or notice message comes from query parsing process. This is usefull because PostgreSQL does not log particular statement if the error was detected *before* raw parser get executed. This typlically happens when encoding error was found(Tatsuo Ishii) - Display original query to log if kind mismatch error was caused by DEALLOCATE(Tatsuo Ishii) - While health checking and recovery use postgres database if possible. If postgres database does not exist, use template1 as it stands now. While connecting template1, certain commands, for example DROP DATABASE cannot used. Using postgres database allows to use these commands while recovery(Tatsuo Ishii) __________________________________________________________________ * Bug fixes - Fix errors in timestamp rewriting which occasionaly cause broken packet sentto slave nodes(Tatsuo Ishii) - Fix errors when timestamp rewriting is used with V2 protocol(Toshihiro Kitagawa) - Propagate Bind, Describe and Close messages to only master node if running in master/slave and in transaction (Tatsuo Ishii) - Fix do_child() so that check_stop_request() exits immediately when smart shutdown signal has been sent. This has been used to work in 2.2(Toshihiro Kitagawa) - Fix ProcessFrontendResponse not to accept invalid frotend packet(Xavier Noguer) - Use %dz for sizeof in fprintf for more portability(Tatsuo Ishii) - Fix compiler warnings(Tatsuo Ishii) - Do not force replication of DEALLOCATE if operated in master/slave mode. Now that pgpool do not execute PARSE in all nodes, this was pointless and caused problem (kind mismatch when executing DEALLOCATE)(Tatsuo Ishii) =============================================================================== 2.3.1 (tomiteboshi) 2009/12/18 * Version 2.3.1 This version includes various fixes for bugs in 2.3 or before. All 2.3 users are encouraged to upgrade to 2.3.1 as soon as possible. __________________________________________________________________ * Bug fixes - If all of following conditions met, incorrect data is written to DB(Tatsuo Ishii) - pgpool is running in replication mode - pgpool is running on 64bit OS - INSERT or UPDATE are used which explicitly include now(), - CURRENT_TIMESTAMP, CURRENT_DATE, CURRENT_TIME. Or the target - Table's default value use above functions - The SQL statement inclludes out of 32bit integer value(-2147483648 to 2147483647 in decimal) Example SQL: INSERT INTO t1(id, regdate) VALUES(98887776655, NOW()); - Fix crush in case of more than 18 DB nodes are used(Tatsuo Ishii) - Enhance kind mismatch error message. If kind is ERROR or NOTICE, the error or notice message from PostgreSQL will be printed (Tatsuo Ishii) =============================================================================== 2.3 (tomiteboshi) 2009/12/07 * Version 2.3 This version enhances replication, especially CURRENT_TIMESTAMP, CURRENT_DATE, now() etc. now can be properly replicated. Also perforance of replication when num_init_children == 1 is enhanced. Pgpool-II now records the status of down nodes, and remember when it restarts to ensure that keep the node status as before. Also some logs are enhanced and more fine fail over controls are added. Please note that pgpool-II 2.3 includes all of enhancements and fixes of pgpool-II 2.2.1 to 2.2.6. __________________________________________________________________ * Incompatibilities from 2.2.x - pgpool_status file is created under logdir. So you need to give write permission to the directory so that pgpool-II can read/write pgpool_status file. __________________________________________________________________ * Enhancements - Enable proper replication of results of temporal functions (CURRENT_TIMESTAMP, CURRENT_DATE, now() etc.). Now applications can execute INSERT/UPDATE tables which include defaut values using those temporal functions. There are small limitations. See restriction sections of docs for more details (Akio Ishida) - Use PostgreSQL 8.4's SQL parser(Akio Ishida) - Enhance the performance 20% to 100% of replication when num_init_children == 1(Tatsuo Ishii) - Add new directive log_per_node_statement which is similar to log_statement except that prints info for each DB node to make it easier which query is sent to which DB node(Tatsuo Ishii) - Add new directive fail_over_on_backend_error to control the behaviro of fail over(Tatsuo Ishii) - Record DB node status and remember when pgpool-II restarts(Tatsuo Ishii) - EXPLAIN and EXPLAIN ANALYZE for SELECT query are now load balanced. This will prevent unwanted kind mismatch errors when EXPLAIN produces slightly different plan(Tatsuo Ishii) - Enhance CSS of pgpool-ja.html(Tatsuo Ishii) - Add sample configuration file for replication mode and master/slave mode(Tatsuo Ishii) - Add test for temporal data(Akio Ishida) =============================================================================== 2.2 Series (2009/02/08 - 2012/08/06) =============================================================================== 2.2.8 (urukiboshi) 2012/08/06 * Version 2.2.8 It includes various fixes for bugs various bug fixes in 2.2.7. __________________________________________________________________ * Bug fixes - Fix do_error_execute_command() so that it discards responses from backend until ErrorResponse received. (Toshihiro Kitagawa) Note that we need to preserve non-error responses (i.e. ReadyForQuery) and put back them using pool_unread(). Otherwise, ReadyForQuery response of DEALLOCATE. This could happen if PHP PDO used. - Fix SimpleForwardToFrontend() so that it reset select_in_transaction flag and execute_select flag when bind error occurred while executing SELECT. (Toshihiro Kitagawa) - Fix bug with pcp_check_fd()'s timeout handling. (Tatsuo Ishii) Per erboles. Subject: [Pgpool-general] question about pcp_check_fd Date: Sun, 23 May 2010 18:21:41 -0500 To: pgpool - Fix buffer overrun problem when pcp password is longer than 32. (Tatsuo Ishii) - Fix wait_for_query_response() not to send param status to frontend if frontend is NULL. This could happen while processing reset_query_list. (Tatsuo Ishii) =============================================================================== 2.2.7 (urukiboshi) 2010/04/15 * Version 2.2.7 This version enhances displaying error messages when kind mismatch error occurs. Also it includes various fixes for bugs in 2.2.6 or before as usual. __________________________________________________________________ * Bug fixes - Fix occasional hangup in extend protocol + master/slave mode, row mode or connection pool mode. Back patch from 2.3 tree(Toshihiro Kitagawa) - Fix long standing bug since pgpool-II 1.0 which causes segfault of pgpool child process. This was caused by miscalculation of shmem size in pgpool parent. Bug analysis by Kitagawa patch created by Tatsuo - Send Parse, Bind, Describe and Close message to only master node if operated in explicit transaction and master/slave mode(Tatsuo Ishii). - Emit a log when postmaster goes down(Tatsuo Ishii) - Fix memory leak in make_persistent_db_connection(Xavier Noguer) - Do not force replication of DEALLOCATE if operated in master/slave mode. Now that pgpool do not execute PARSE in all nodes, this was pointless and caused problem (kind mismatch when executing DEALLOCATE) (Tatsuo Ishii) - Fix crash with show pool_status when there many (more than 18) DB nodes(Tatsuo Ishii) - Enhance "kind mismatch" message. If kind is ERROR or NOTICE, print the ERROR/NOTICE message to help users to find what's going on(Tatsuo Ishii) =============================================================================== 2.2.6 (urukiboshi) 2009/12/01 * Version 2.2.6 This version enhances handling of backend weight. Also it allows to use temp tables in master/slave mode. It includes various fixes for bugs in 2.2.5 or before as usual. __________________________________________________________________ * Bug fixes - Do not allow to load balance DECLARE, CLOSE, FETCH and MOVE. If data gets updated and CLOSE issued after transaction commit(i.e. holdbale cursor), this will cause data inconsistency since CLOSE is executed one of the severs, rather than all(Tatsuo Ishii) - In master/slave mode, execute Parse message on only master node. In previous versions Parse executed on all nodes, which grabbed unneccessary lock(Tatsuo Ishii) - Remove init script from all runlevels before uninstall(Devrim) - Fix pgpoo.spec(Devrim) - Do not execute REINDEX DATABASE or SYSTEM, CREATE/DROP TABLE SPACE inside a transaction block(Tatsuo Ishii) __________________________________________________________________ * Enhancements - Allow to change weight by reloading pgool.conf. This will take effect for next client sessions(Tatsuo Ishii) - Reply with usefull error messages, rather than "server closed the connection unexpectedly" when authentication fails(Glyn Astill) - Add info to logs when writing to sockets falis to know if it was for backend or frontend(Tatsuo Ishii) - Do not complain when writing to socket of frontend fails(Tatsuo Ishii) - Allow to use temp tables in master/slave mode. INSERT/UPDATE/DELETE will automatically be sent to master only. Still SELECT you need to add /*NO LOAD BALANCE*/ comment(Tatsuo Ishii) - Add temp table test to test/jdbc(Tatsuo Ishii) =============================================================================== 2.2.5 (urukiboshi) 2009/10/4 * Version 2.2.5 This version fixes various bugs in 2.2.4 or before. __________________________________________________________________ * Bug fixes - Fix connection_count_down(). It decrements the connection counter too much in some corner cases and causes online recover never completes(Tatsuo Ishii) - Detect frontend exiting while waiting for commands complete in other cases such as internal locks are issued and Parse (Tatsuo Ishii) - Fix inifinit loop in reset_backend(Xavier Noguer, Tatsuo) - Fix Parse() to print actual query when it detects kind mismatch error(Tatsuo Ishii) - Document enhancements(Tatsuo Ishii) =============================================================================== 2.2.4 (urukiboshi) 2009/8/24 * Version 2.2.4 This version fixes various bugs in 2.2.3 or before. __________________________________________________________________ * Bug fixes - Fix possible bug introduced in pgpool-II 2.2.2. Load balance control variables may remain not be restored and subsequent DML/DDL call might sent to only master node(Tatsuo Ishii) - Send NOTICE message to frontend periodically if V2 protocol is used. This is ifdef out since it affectes visible change to applications. 2.2.3 unconditionaly sends param packet to client even it uses version 2 protocol, which is apparentlt wrong. Also tweak checking period from 1 second to 30 seconds since 1 second seems too aggressive(Tatsuo Ishii) - Block signals before forking children rather after. Otherwise parent will be killed by failover signal if it receives a signal before establishing signal handler(Tatsuo Ishii) - Remove unnecessary spaces and tabs at the end of line(Jun Kuriyama) =============================================================================== 2.2.3 (urukiboshi) 2009/8/11 * Version 2.2.3 This version fixes various bugs in 2.2.2 or before. __________________________________________________________________ * Bug fixes - Fix child process death if one of backends is not available(Tatsuo Ishii). - Fix various parallel query bugs(Yoshiharu Mori) - Fix message corruption for kid mismatch error(Akio Ishida) - Now stetmemt_time works(Tatsuo Ishii) - Enhance health checking to detect postmaster stopping by SIGSTOP(Tatsuo Ishii) - Detect frontend abnormal exiting while waiting for reply from backend. This only works with V3 protocol(Tatsuo Ishii) - Do not start internal transaction if command is CLUSTER without arguments(Tatsuo Ishii) - Fix bug with COPY FROM in that backend process remains after COPY failed(Tatsuo Ishii) __________________________________________________________________ * Enhancements - Show last query for extended protocol(Akio Ishida) - Allow to compile sql/pgpool-recovery/pgpool-recovery.c with PostgreSQL 8.4(Tatsuo Ishii) =============================================================================== 2.2.2 (urukiboshi) 2009/5/5 * Version 2.2.2 This version fixes various bugs in 2.2.1 or before. Please note that an importan fix is made to avoid data incositency risk, which could happen when client does not exit gracely(without sending "X" packet) while pgpool is trying to send data to it. This could happen with all version of pgpool-II. __________________________________________________________________ * Bug fixes - Ignore write error on frontend connection. This is needed to continue processing with backend, otherwise we risk data incositency(Tatsuo Ishii) - Fix bug introduced in 2.2.1 (In master slave mode, sometimes DEALLOCATE fails). If prepared statement reused, pgpool hangs(Toshihiro Kitagawa) - Fix pgpool crash when SQL command PREPARE and protocol level EXECUTE are mixed. The bug was introduced in 2.2(Tatsuo Ishii) - Avoid "unexpected EOF on client connection" error in PostgreSQL when reset query fails(Tatsuo Ishii) =============================================================================== 2.2.1 (urukiboshi) 2009/4/25 * Version 2.2.1 This version fixes various bugs in 2.2. __________________________________________________________________ * Bug fixes - In master slave mode, sometimes DEALLOCATE fails. This is caused by that the first PREPARE was not executed on the slave(Toshihiro Kitagawa) - Update pgpool.spec along with related files(Devrim) - Fix inser_lock so that it is ignored when protocol version is 2(Tatsuo Ishii) - Remove excessive log messages regarding parameter change notice(Tatsuo Ishii) - Add missing files to doc(Tatsuo Ishii) =============================================================================== 2.2 (urukiboshi) 2009/2/28 * Version 2.2 This version enhances SERIAL data type handling and on line recovery. Also an important bug, serializable transactions could cause data inconsistency among DB nodes, is fixed. Query cancelation, which never worked since pgpool-II was born, is finally fixed. __________________________________________________________________ * New features/enhancements - With insert_lock, now a table is locked only if it has SERIAL data type and now the default value for insert_lock is true(Tatsuo Ishii) - Start internal transaction other than INSERT, UPDATE, DELETE and SELECT to keep node consistency(Tatsuo Ishii) - Add client_idle_limit_in_recovery directive. This will prevent 2nd stage of on line recovery from not going forward by idle clients sitting forever(Tatsuo Ishii) - Add pid_file_name directive which specifies a path to the file containing pgpool process id. "logdir" is no more used(Tatsuo Ishii) - Allow to load balance DECLARE, FETCH and CLOSE(Tatsuo Ishii) - Add -d option to pcp commands(Jun Kuriyama) - Enahnce kind mismatch error message to include originarl query string(Tatsuo Ishii) __________________________________________________________________ * Bug fixes - Close all file descriptors when running in daemon mode. Otherwise we inherit sockets from apache when it's started by pgpoolAdmin. This results in that port 80 is occupied for example. Patch provided by Akio Ishida. Also add chdir("/"). This is always good for daemon programs(Tatsuo Ishii) - Allow MD5 authentication in raw mode as stated in docs(Tatsuo Ishii) - Check transaction serialization failure error in serializable mode and abort all nodes if so. Otherwise we allow data inconsistency among DB nodes(Tatsuo Ishii). See following scenario: (M:master, S:slave) M:S1:BEGIN; M:S2:BEGIN; S:S1:BEGIN; S:S2:BEGIN; M:S1:SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; M:S2:SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; S:S1:SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; S:S2:SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; M:S1:UPDATE t1 SET i = i + 1; S:S1:UPDATE t1 SET i = i + 1; M:S2:UPDATE t1 SET i = i + 1; <-- blocked S:S1:COMMIT; M:S1:COMMIT; M:S2:ERROR: could not serialize access due to concurrent update S:S2:UPDATE t1 SET i = i + 1; <-- success in UPDATE and data becomes inconsistent! - avoid kind mismatch error caused by "SET TRANSACTION ISOLATION LEVEL must be called before any query"(Tatsuo Ishii). This could happen in following scenario: M:S1:BEGIN; S:S1:BEGIN; M:S1:SELECT 1; <-- only sent to MASTER M:S1:SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; S:S1:SET TRANSACTION ISOLATION LEVEL SERIALIZABLE; M: <-- error S: <-- ok since no previous SELECT is sent. kind mismatch error occurs! - Process status display has extra space on FreeeBSD(Jun Kuriyama) - Fix incorrect kind mismatch detection case. e.g: BEGIN; BEGIN; (Tatsuo Ishii) - If PostgreSQL sends lots of DEBUG message, sometimes pgpool complains: 2008-11-08 22:41:53 ERROR: pid 23744: do_command: backend does not return ReadyForQuery. This due to a wrong assumption for the client/server protocol(Tatsuo Ishii) - Fix the case when sending an erronous query to abort transaction. It assumed that after sending an error query, always ReadyForQuery came right after that. If some debugging or logging verboseness is set, PostgreSQL might sends NOTICE before ReadyForQuery(Tatsuo Ishii) - Query cancelation now works. It never worked since pgpool-II was born(Tatsuo Ishii) - Fix online recovery to wait for failback done before allowing to accept connections from clients. It was supposed to work like this but actually was not since the day 0 when online recovery was born. Without the fix there could be potential data inconsistency among DB nodes(Tatsuo Ishii) - Fix pgpool-II crash after on line recovery. This happens after the failback process adds a recovered node which has no connection to the node(Tatsuo Ishii) - Fix pgpool-II errors when postgresql.conf is reloaded. This was caused by parameter status packet sent asynchronously from backend, which indicates the internal setting of backend has been changed(Tatsuo Ishii) __________________________________________________________________ * Incompatible changes - Always fail over and restart all children. Before we do restart only if master has not been changed. This is wrong. If we have trouble with network cable or something, TCP/IP stack keeps on retrying for long time and the only way to prevent it is restarting process(Tatsuo Ishii) - "logdir" is no more used. Instead use "pid_file_name"(Tatsuo Ishii) - Default value for insert_lock is now true(Tatsuo Ishii) =============================================================================== 2.1 Series (2008/07/25 - 2008/07/25) =============================================================================== 2.1 (inamiboshi) 2008/7/25 * Version 2.1 __________________________________________________________________ * New feature - Add '%m' format to failover_command and failback_command to obtain new master node ID. (Yoshiyuki Asaba) - Add '%m' format to failover_command and failback_command to obtain old master node ID. (Yoshiyuki Asaba) - Add new directive "recovery_timeout" to specify recovery timeout in second. (Taiki Yamaguchi) - Add optino '-v' to print pgpool version. (Yoshiyuki Asaba) __________________________________________________________________ * Incompatibility - Restrict pgpool_recovery() and pgpool_remote_start() functions to superusers. (Yoshiyuki Asaba) - Do not create a connection pool to standby node in raw mode. (Yoshiyuki Asaba) - Remove "replication_timeout" parameter. (Yoshiyuki Asaba) - This enabled if replication_strict was false. However, replication_strict was already removed. - Ignore timeout argument of pcp commands. (Taiki Yamaguchi) - Do not replicate "COPY TO STDOUT" when replicate_select is false. (Yoshiyuki Asaba) __________________________________________________________________ * Bug fix ** General - Fix crash when CloseComplete message was received. (Yoshiyuki Asaba) - Improve network I/O routine. (Yoshiyuki Asaba) - Fix compile errors on Solaris 10. (Yoshiyuki Asaba) - Improve log messages of health check and recovery. (Tatsuo Ishii) - Change error level of the "failed to read kind from frontend" message from ERROR to LOG. (Yoshiyuki Asaba) - Fix failover failure in raw mode. (Taiki Yamaguchi) - Fix zombie process bug. (Yoshiyuki Asaba) - Fix health_check_timeout to work correctly. (Kenichi Sawada) - Support ps status on FreeBSD. (ISHIDA Akio) - Improve bind(2) failure report. (Jun Kuriyama) - Improve error message when client authentication failed. (Tatsuo Ishii) ** Replication - Fix replicate_select to work correctly. (Tatsuo Ishii) - Fix a wrong rollback bug with extended query. (Yoshiyuki Asaba) - Fix a bug with asynchronous query. (Yoshiyuki Asaba) - Fix hint clause handling like /*REPLICATION*/ with extended query. (Yoshiyuki Asaba) - Fix crash of "DEALLOCATE ALL". (Yoshiyuki Asaba) - Fix hang up when a backend node does immediate shutdown. (Yoshiyuki Asaba) - Fix hang up online recovery in high load. (Yoshiyuki Asaba) - Fix hang up with extended query protocol when SELECT is failed inside a transaction block. (Yoshiyuki Asaba) ** Master Slave - Fix load balancing to work correctly. (Yoshiyuki Asaba) - Fix crash if SET, PREPARE or DEALLOCATE is executed inside a transaction block. (Yoshiyuki Asaba) ** Parallel query - Fix INSERT failure. (Yoshiharu Mori) - Fix syntax error when a query contains "AS" in FROM clause. (sho) - Fix Hung up when two or more statment was executed in parallel mode (Yoshiharu Mori) - Fix Query rewriting of Join Expression and DISTINCT ON (Yoshiharu Mori) =============================================================================== 2.0 Series (2007/11/16 - 2007/11/21) =============================================================================== 2.0.1 (hikitsuboshi) 2007/11/21 * Version 2.0.1 * Fix process down with UPDATE or DELETE query.(Yoshiyuki Asaba) * Send a syntax query only to a master node if master_slave is true.(Yoshiyuki Asaba) =============================================================================== 2.0 (hikitsuboshi) 2007/11/16 * Version 2.0 __________________________________________________________________ * Incompatibility since pgpool-II 1.x - the default value for ignore_leading_white_space is now true(Yoshiyuki Asaba) - replicate_strict is removed. The value is always true(Yoshiyuki Asaba) __________________________________________________________________ * General - Allow to reload pgpool.conf(Yoshiyuki Asaba) - The paraser is now compatible with PostgreSQL 8.3(Yoshiyuki Asaba) - Add new directive "failover_command" to specify command when a node is detached(Yoshiyuki Asaba) - Add new directive "client_idle_limit" to specify the time out since the last command is arrived from a client(Tatsuo Ishii) __________________________________________________________________ * Replication - Always start a new transaction even if the query is not in an explicit transaction to enhance the reliabilty of replication(Yoshiyuki Asaba) - Enhance the performance of replication for write queries. Now the worst case is 1/2 compared with single DB node regardless the number of DB nodes. Previous release tends to degrade according to the numer of DB nodes(Yoshiyuki Asaba) - Add "online recovery" which allows to add a DB node and sync with other DB nodes without stopping the pgpool server(Yoshiyuki Asaba) - Abort a transaction if INSERT, UPDATE and DELETE reports different number of result rows(Yoshiyuki Asaba) x=# update t set a = a + 1; ERROR: pgpool detected difference of the number of update tuples HINT: check data consistency between master and other db node - If the results from DB nodes do not match, select the possible correct result by "decide by majority". Previous release always trust the result of the master DB node(Yoshiyuki Asaba) - Allow load balance in V2 frontend/backend protocol(Yoshiyuki Asaba) __________________________________________________________________ * Parallel query - Allow "partial replication" to enhance the performance of the parallel query(Yoshiharu Mori) =============================================================================== 1.3 Series (2007/10/23 - 2007/10/23) =============================================================================== 1.3 (sohiboshi) 2007/10/23 * Version 1.3 * Add new "authentication_timeout" directive, being the default value is 60. (Yoshiyuki Asaba) - Maximum time in seconds to complete client authentication. * Reject a connection when startup packet length is greater than 10,000 byte. (Yoshiyuki Asaba) * Fix invalid memory access when pgpool processed DEALLOCATE statement. (Yoshiyuki Asaba) * Fix hang up in load balance mode. (Yoshiyuki Asaba) - This was introduced in V1.2. * Fix segmentation fault in 64-bit environment when query cache is enable. (Yoshiyuki Asaba) =============================================================================== 1.2 Series (2007/08/01 - 2007/09/28) =============================================================================== 1.2.1 (tomoboshi) 2007/09/28 * Version 1.2.1 * Fix deadlock while processing Parse message. (Yoshiyuki Asaba) * Fix memory leak in reset_prepared_list(). (Yoshiyuki Asaba) * Fix compile error on FreeBSD 4.11. (Yoshiyuki Asaba) * SET, PREPARE and DEALLOCATE statements are replicated in master/slave mode. (Yoshiyuki Asaba) =============================================================================== 1.2 (tomoboshi) 2007/08/01 * Version 1.2 * Add new "replicate_select" directive, being the default value is false. (Yoshiyuki Asaba) - If it is true, SELECT query is replicated. This behavior is same as V3.2 or earlier. * Improve signal handling. (Yoshiyuki Asaba) - Occasionaly, zombie processes were remained. Or processes were unstable. * Fix hang up when SELECT was error inside a transaction block. The bug was introduced in V3.3. (Yoshiyuki Asaba) * Fix PREPARE/EXECUTE handling in master slave mode. (Yoshiyuki Asaba) * Fix "kind mismatch error" when deadlock error * Fix hang up and SEGV in extended query protocol when a warning SQL like "SELECT '\'';" executed. (Yoshiyuki Asaba) * Fix hang up when postmaster did fast or immediate shutdown. (Yoshiyuki Asaba) * Fix memory leak when connection cache was full. (Yoshiyuki Asaba) * Load balancing node is selected when a session starts. (Yoshiyuki Asaba) * Fix buffer overrun if connection_life_time was set. (Yoshiyuki Asaba) =============================================================================== 1.1 Series (2007/5/25 - 2007/6/15) =============================================================================== 1.1.1 (amiboshi) 2007/6/15 * Version 1.1.1 * Fix "kind mismatch" bug when load_balance_mode is true introduced in 1.1 (Yoshiyuki Asaba) * Fix deadlock with extended query protocol(Yoshiyuki Asaba) * Fix numerous bugs with protocol V2(Yoshiyuki Asaba) =============================================================================== 1.1 (amiboshi) 2007/5/25 * Version 1.1 * Support HBA authentication(Taiki Yamaguchi) * Support log_connections(Taiki Yamaguchi) * Support log_hostname(Taiki Yamaguchi) * Show pgpool status in ps command(Taiki Yamaguchi) * Fix compile error on MacOS X(Yoshiyuki Asaba) * Allow load balancing with extended protocol(Yoshiyuki Asaba) * Improve replication. SELECT nextval() and SELECT setval() are now replicated. (Yoshiyuki Asaba) * Change SELECT query is only sent to the master node. (Yoshiyuki Asaba) - Use /*REPLICATION*/ comment to repliate a SELECT query. * Fix unexpected failover error due to receiving an interrupt signal while connecting to the backend. (Yoshiyuki Asaba) * Add "pgpool.pam" file, for PAM configuration file, to be installed under "$PREFIX/share/pgpool-II/". (Taiki Yamaguchi) * Fix core dump when executing large SQL. (Yoshiyuki Asaba) =============================================================================== 1.0 Series (2006/09/08 - 2007/02/12) =============================================================================== 1.0.2 (suboshi) 2007/02/12 * Version 1.0.2 * Fix bug when executing large SQL to prevent pgpool goes into infinite loop(Yoshiyuki Asaba) * Fix bug with extended protocol handling(Yoshiyuki Asaba) * Enhance log for failover and failback(Tatsuo Ishii) * Add backend status info to show pool_status(Tatsuo Ishii) * Fix UPDATE/DELETE returns wrong number of rows(Tatsuo Ishii) * Fix configure fails to link libpq when used with older version of gcc(Yoshiyuki Asaba) * Fix DEALLOCATE treatment when used with PHP:PDO DBD-Pg(Yoshiyuki Asaba) * Do not load balance SELECT FOR UPDATE, SELECT INTO and SELECT with comments. This behavior is compatible with pgpool-I(Yoshiyuki Asaba) * Obtain path to libpq using pg_config. --with-pgsql will be removed in next version(Yoshiyuki Asaba) * When reusing connection pool, reconnect to backend if the socket is broken(Yoshiyuki Asaba) * Fix error with configure when used with PostgreSQL 7.4(Yoshiyuki Asaba) =============================================================================== 1.0.1 (suboshi) 2006/09/22 * Version 1.0.1 * This version fixes bugs including master/slave not being working, deadlock problem in COPY FROM STDIN. Also documents are improved. =============================================================================== 1.0 (suboshi) 2006/09/08 * Initial release Local Variables: mode: outline End: