[Pgpool-general] pgpool-II-2.2.1 dropping secondary back-end

[Tatsuo Ishii]

> > 2009-07-13 11:54:03 DEBUG: pid 785: health_check: 0 th DB node status: 2
> > 2009-07-13 11:54:03 DEBUG: pid 785: health_check: 1 th DB node status: 2
> 
> Sounds good so far...
> 
> > 2009-07-13 11:54:55 ERROR: pid 5587: pool_read: read failed (Connection
> > reset by peer)
> 
> Looks like one of your PostgreSQL servers is playing. I would check
> the configuration of that node (everything in it), because something
> is wrong. I cannot really provide any more information with the log
> you provide, but I've had this error before, too. I would also check
> that the clients connecting to the cluster are doing it properly
> (including closing connections).

From the log what I can read is:

pgpool child(pid 5587) tried to read some data from backend and failed
then triggered failover. Note that pgpool does not triggers failover
if it fails to read/write from pgpool client.

So next question is why the second backend
failed. ("notice_backend_error: 1" indicates this was the second
backend. If it were the first backend, then the log should have been
"notice_backend_error: 0")

This is not clear from the pgpool log. The possible reasons are:

1) the second backend failed for some reason. You will find something
   in the PostgreSQL log in this case.

2) the network connection between pgpool and the second backend isn't
   realiable.

> > 2009-07-13 11:54:55 LOG:   pid 5587: notice_backend_error: 1 fail over
> > request from pid 5587
> > 2009-07-13 11:54:55 ERROR: pid 5587: pool_flush_it: write failed (Broken
> > pipe) offset: 0 wlen: 5
> 
> First time I see this error, so I cannot say anything beyond what's
> obvious (what the message says).
> 
> > 2009-07-13 11:54:55 DEBUG: pid 785: failover_handler called
> [..]
> > 2009-07-13 11:54:55 DEBUG: pid 788: child receives shutdown request signal 3
> 
> Normal degeneration of a node...

Yup.
--
Tatsuo Ishii
SRA OSS, Inc. Japan