[pgpool-general: 7101] Strange behavior after failover

Radosław Szczygieł radoslaw.szczygiel at interia.pl
Mon Jun 22 20:16:42 JST 2020


Hi PGPOOL TEAM,I have strange behavior of pgpool after failover.My pgpool-pg cluster is:2xPgpool 4.1, 2xPG 12.1 (streaming replication)PGpool conf:# - Concurrent session and pool size -num_init_children = 600                                   # Number of concurrent sessions allowed                                   # (change requires restart)max_pool = 5                                   # Number of connection pool caches per connection                                   # (change requires restart)# - Life time -child_life_time = 1080                                 # Pool exits after being idle for this many secondschild_max_connections = 2900                                   # Pool exits after receiving that many connections                                   # 0 means no exitconnection_life_time = 840                                   # Connection to backend closes after being idle for this many seconds                                   # 0 means no closeclient_idle_limit = 600After failover from master ip .. 72 (node 0) to new master ip ..73 (node 1) and rebuild old master to new slave there were strange behavior of connections and transactions.  Before failover there were about 1500 connections per backend. After failover and rebuild slave there were almost 3000 connections per backend and many of connections to databases were on max level of Pgpool num_init_children parameter.Also on new slave I've got many warnings from transations "WARNING:  there is no transaction in progress" like PGPOOL send sql COMMIT; to slave.I had to make failover again to state before last failover because I was afraid of this situation. Logs looks normal but there was something wrong with closing connections to backends and this abnormal behavior of transations.Please look on screenshot from connections graph. Its shot from last failover where was bad state to normal state (master ip ..73 node 1 to new master ip ..72 node 0).Do You know something about problems with connections after failover?pgpool1.png is graph all of connections to backend moment of last failover to normal statepgpool3.png is graph from one of databases moment of last failover to normal state --Pozdrawiam
Radosław Szczygieł
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20200622/9797b567/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: image/jpeg
Size: 17571 bytes
Desc: not available
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20200622/9797b567/attachment-0001.jpe>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgpool1.PNG
Type: image/png
Size: 37447 bytes
Desc: not available
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20200622/9797b567/attachment-0002.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgpool3.PNG
Type: image/png
Size: 28398 bytes
Desc: not available
URL: <http://www.sraoss.jp/pipermail/pgpool-general/attachments/20200622/9797b567/attachment-0003.png>


More information about the pgpool-general mailing list