[pgpool-general: 4265] Pgpool - connection hangs in DISCARD ALL

[Gerhard Wiesinger]

Hello,

Looks like the issue with hanging connections has not been solved in all 
cases, backend hangind in DISCARD ALL, TCP connection is still alive, 
but pgpool doesn't know the connection anymore.

Version: pgpool: 3.4.3
PostgreSQL: 9.4.5

Relevant pgpool config:
listen_addresses = '*'
child_life_time = 10
child_max_connections = 10
connection_life_time = 2

# One connection to the clients, but backend connection still alive
netstat -pan | grep pgpool | grep -i ESTABLISHED
tcp        0      0 127.0.0.1:50176         127.0.0.1:5432 ESTABLISHED 
32119/pgpool: hei
tcp        0      0 127.0.0.1:50180         127.0.0.1:5432 ESTABLISHED 
32119/pgpool: hei
tcp        0      0 127.0.0.1:52268         127.0.0.1:5432 ESTABLISHED 
9967/pgpool: wait f
tcp        0      0 192.168.20.104:9999 192.168.20.113:42760     
ESTABLISHED 32119/pgpool: hei

pcp_proc_info for all returned:
Database     : ab3
Username     : nag
Start time   : 2015-12-27 15:48:38
Creation time: 2015-12-27 15:49:16
Major        : 3
Minor        : 0
Counter      : 1
Backend PID  : 32141
Connected    : 0
Database     : hei
Username     : hei
Start time   : 2015-12-27 15:48:38
Creation time: 2015-12-27 15:49:16
Major        : 3
Minor        : 0
Counter      : 1
Backend PID  : 32143
Connected    : 1

# Tuned tcp stack:
net.ipv4.tcp_keepalive_intvl=10
net.ipv4.tcp_keepalive_probes=3
net.ipv4.tcp_retries2=3

Looks like that this bug in release 3.3.3 should have been fixed in 3.4. 
But I guess there are some corner cases still available.
http://www.pgpool.net/docs/pgpool-II-3.4.0/NEWS.txt

Had it only once yet, connection was terminated after some minutes 
automatically.

Any ideas?

Thnx.

Ciao,
Gerhard