[Pgpool-general] pgpool losing queries

[Daniel Codina]

Sorry,.... mistake,.. maybe it can help somebody:

It was not the value in:
/proc/sys/net/ipv4/tcp_keepalive_probes

But the one in:
/proc/sys/net/ipv4/tcp_syn_retries



2010/2/26 Daniel Codina <dcodina at laigu.net>

> Hi there,
>
> I discovered a part of my problem. The loosing queries thing. It was a
> permisions problems in the wal arxive directory, and now I am not loosing
> those many queries anymore, but still loosing one every time I stop any
> node.
>
> About the "unpluged wire".
> Am I right tinking that to have health_check actives it is only necessari
> something like that?
> health_check_timeout = 1
> health_check_period = 1
> health_check_user = 'postgres'
>
> I also think that if health_check_timeout is 0 is the tcp values that
> counts.
>
> Anyway, I tryed it with 0 and 1, and yet have the same problem. I don't
> know if pgpool is supposed to notice that, one of the nodes is not online
> anymore. I noticed that every time, pgpool rested without listen 48 seconds.
> I have been debugging and I discovered that it is waiting because of a SYN
> sent.
>
> I can reduce this time by changeing the value on
> /proc/sys/net/ipv4/tcp_keepalive_probes to something smaller than 5 (5 will
> be 48 secs,... starting with 3 ecery retry will be double, 3, 6, 12, 24, 48
> secs).
> I don't think it is a good solution, and, I suppose pgpool would have to
> notice it, so I suppose I am doing something wrong.
>
> Is it working for everybody else? I mean, if you unglug one node, does
> pgpool continue working without a long wait? If is working for everybody,..
> jeje,. then it's me,... if not,... maybe the SYN think can help somebody!
>
> Thanks!
>
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://pgfoundry.org/pipermail/pgpool-general/attachments/20100226/de8c5497/attachment.html>