[pgpool-general: 5220] Re: ERROR: unable to read data from frontend

[Tatsuo Ishii]

> Hi,
> 
> we are encountering an issue in our labs in which some pgpool connections
> are stuck until reboot to the machine or the pgpool service.
> 
> this issue has occurred at least 4 times so far in our labs in 3 different
> setups.
> 
> this is from the logs (the same lines several times):
> ..
> 
> *2017-01-05 22:35:18: pid 1729: ERROR:  unable to read data from
> frontend2017-01-05 22:35:18: pid 1729: DETAIL:  EOF encountered with
> frontend*
> ...
> 
> another occurrence (the same lines several times):
> ...
> 
> 
> *2017-01-05 22:18:47: pid 11860: DETAIL:  EOF encountered with frontend*
> 
> 
> *2017-01-05 22:18:47: pid 21006: ERROR:  unable to read data from frontend*
> 
> *2017-01-05 22:18:47: pid 21006: DETAIL:  EOF encountered with frontend*

Probably these are not related to the stuck problem.

> *2017-01-05 22:18:47: pid 14895: LOG:  failback event detected*
> * 2017-01-05 22:18:47: pid 14895: DETAIL:  restarting myself*
> ...
>
> In our labs pgpool and postgres is running on the same machine.
> 
> after this issue occurred, although we brought down the application, we
> could see processes in the following state: "pgpool: idle in transaction"
> (until we restarted pgpool)

Have you enabled client_idle_limit in pgpool.conf?

I suspect "pgpool: idle in transaction" is caused by lazy clients
which do not close transactions or connections. By setting
client_idle_limit to non 0, Pgpool-II could kick out such that clients
and keep child process can accept new connections from clients.

> pgpool version: 3.6.1
> PostgreSQL version: 9.5.4
> OS: CentOS 6.5
> 
> Thanks in advance,
> Rolan.

Best regards,
--
Tatsuo Ishii
SRA OSS, Inc. Japan
English: http://www.sraoss.co.jp/index_en.php
Japanese:http://www.sraoss.co.jp