View Issue Details
| ID | Project | Category | View Status | Date Submitted | Last Update |
|---|---|---|---|---|---|
| 0000448 | Pgpool-II | Bug | public | 2018-11-28 00:02 | 2018-12-06 17:32 |
| Reporter | amar | Assigned To | t-ishii | ||
| Priority | high | Severity | major | Reproducibility | always |
| Status | resolved | Resolution | open | ||
| Platform | Linux | OS | Ubuntu | OS Version | 14.04 |
| Product Version | 3.7.5 | ||||
| Target Version | 3.7.8 | ||||
| Summary | 0000448: pgpool holds the connections as idle_in_transaction which are timed out due to idle_in_transaction setting on backend nodes | ||||
| Description | We have a PG cluster with streaming replication running on PG pool 3.7.5 and PG 9.6. idle_in_transaction_session_timeout is set on PG master to some 'x' ms. So, the master timeout the queries which are idle_in_transaction for more than 'x' ms. Even when the transaction on master is timed out. 1. ps -ef from pgpool still shows the idle_in_transactions whereas the pg_stat_activity on master has no traces since the transactions are already times out. postgres 46211 113280 0 Nov26 ? 00:00:00 pgpool: username db xxx.xx.xxx.xxx(56796) idle in transaction postgres 46225 113280 0 Nov26 ? 00:00:00 pgpool: username db xxx.xx.xxx.xxx(53464) idle in transaction postgres 48762 113280 0 Nov26 ? 00:00:00 pgpool: username db xxx.xx.xxx.xxx(50008) idle in transaction postgres 48768 113280 0 Nov26 ? 00:00:00 pgpool: username db xxx.xx.xxx.xxx(45138) idle in transaction postgres 48828 113280 0 Nov26 ? 00:00:00 pgpool: username db xxx.xx.xxx.xxx(45562) idle in transaction postgres 48912 113280 0 Nov26 ? 00:00:01 pgpool: username db xxx.xx.xxx.xxx(46498) idle in transaction 2. netstat -an | grep 5432 shows the connections in CLOSE_WAIT state tcp 1 0 xxx.xx.xxx.xxx:15625 xxx.xx.xxx.xxx:5432 CLOSE_WAIT tcp 1 0 xxx.xx.xxx.xxx:15868 xxx.xx.xxx.xxx:5432 CLOSE_WAIT tcp 1 0 xxx.xx.xxx.xxx:63032 xxx.xx.xxx.xxx:5432 CLOSE_WAIT tcp 1 0 xxx.xx.xxx.xxx:62956 xxx.xx.xxx.xxx:5432 CLOSE_WAIT Shouldn't the pgpool close these connections when the transactions are timeout? | ||||
| Steps To Reproduce | 1. Set the idle_in_transaction_session_timeout to some 'x' ms on pgmaster. 2. Initiate a connection to pg cluster via pg pool which ends up in idle_in_trasaction. 3. Wait till the transaction gets timed out on pgmaster. 4. Do a netstat and ps -ef to check the connection still exists even after the timeout. | ||||
| Tags | No tags attached. | ||||
|
|
Thanks for the report. I was able to reproduce the problem. To fix this, we need to disconnect session from Pgpool-II to client just like the case "connection was terminated due to conflict with recovery" error. Attached patch does that. |
|
|
Thanks for the quick response. Do you have a plan to patch it in 3.7.8? |
|
|
Yes, once you confirm the patch, basically we are going to apply it to all supported branches including 3.7. That means we are going to release 3.7.9 etc. in the future (the release schedule is not yet decided though). |
|
|
I have pushed the patch to all supported branches including 3.7 stable. So the patch will appear in the next minor releases. |
| Date Modified | Username | Field | Change |
|---|---|---|---|
| 2018-11-28 00:02 | amar | New Issue | |
| 2018-11-28 09:44 | administrator | Assigned To | => t-ishii |
| 2018-11-28 09:44 | administrator | Status | new => assigned |
| 2018-11-28 11:33 | t-ishii | File Added: idle_in_transaction.diff | |
| 2018-11-28 11:33 | t-ishii | Note Added: 0002294 | |
| 2018-11-28 11:34 | t-ishii | Status | assigned => feedback |
| 2018-11-28 11:34 | t-ishii | Target Version | => 3.7.8 |
| 2018-11-28 13:24 | amar | Note Added: 0002295 | |
| 2018-11-28 13:24 | amar | Status | feedback => assigned |
| 2018-11-28 13:36 | t-ishii | Note Added: 0002296 | |
| 2018-12-06 17:32 | t-ishii | Note Added: 0002306 | |
| 2018-12-06 17:32 | t-ishii | Status | assigned => resolved |
