[Pgpool-general] Sometimes "connection_life_time" does not work as expected.

[Pablo Milano]

I am using pgpool 2.5.2 (Debian Stable version) in a real production
environment.
In order to purge unused connections from pgpool to the postgres backends, I
changed the igured the "connection_life_time" parameter from 0 (default) to
90 seconds. At first, everithing worked fine, the unused connections were
released in 90 seconds as expected, but after a while, pgpool ended ALL the
connections with one of the backends, leading to degeneration mode, even
though that backend works properly. The following log entries are present in
"postgresql.log"

2005-12-18 16:02:53 [30104] LOG:  could not receive data from client:
Connection reset by peer
2005-12-18 16:02:53 [30104] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [30105] LOG:  could not receive data from client:
Connection reset by peer
2005-12-18 16:02:53 [30105] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [29728] LOG:  could not receive data from client:
Connection reset by peer
2005-12-18 16:02:53 [29728] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [29125] LOG:  could not receive data from client:
Connection reset by peer
2005-12-18 16:02:53 [29125] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [28183] LOG:  could not receive data from client:
Connection reset by peer
2005-12-18 16:02:53 [28183] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [24016] LOG:  could not receive data from client:
Connection reset by peer
2005-12-18 16:02:53 [24016] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [29435] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [29331] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [30212] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [29121] LOG:  unexpected EOF on client connection
2005-12-18 16:02:53 [30103] LOG:  could not receive data from client:
Connection reset by peer
2005-12-18 16:02:53 [30103] LOG:  unexpected EOF on client connection


All the conn identifiers are related to the same backend. I can't isolate
the cause of this, but when this behaviour occurs, pgpool enters
degeneration mode, even though all the backends are fully working. These are
the pgpool log entries at that time


Dec 18 16:02:00 asimov pgpool: 2005-12-18 16:02:00 LOG:   pid 19605:
starting health checking
Dec 18 16:02:30 asimov pgpool: 2005-12-18 16:02:30 LOG:   pid 19605:
starting health checking
Dec 18 16:02:53 asimov pgpool: 2005-12-18 16:02:53 ERROR: pid 26764:
connect_inet_domain_socket: connect() failed: nterrupted system call
Dec 18 16:02:53 asimov pgpool: 2005-12-18 16:02:53 LOG:   pid 26764:
notice_backend_error: master: 0 fail over request from pid 26764
Dec 18 16:02:53 asimov pgpool: 2005-12-18 16:02:53 LOG:   pid 19605:
starting degeneration. shutdown secondary host bradbury(5432)


May it be a wrong behaviour in pgpool when closing "idle" connections?
Thanks in advance.