From: | Vladimir Sitnikov <sitnikov(dot)vladimir(at)gmail(dot)com> |
---|---|
To: | Vitalii Tymchyshyn <vit(at)tym(dot)im>, Dave Cramer <pg(at)fastcrypt(dot)com>, "jingzhi(dot)zhang(at)outlook(dot)com" <jingzhi(dot)zhang(at)outlook(dot)com> |
Cc: | List <pgsql-jdbc(at)postgresql(dot)org> |
Subject: | Re: PrepareStatement.execute() blocked because of long time 'create index' operation, connection leak |
Date: | 2016-06-07 12:44:45 |
Message-ID: | CAB=Je-Hop9x03hvwLOoeONGtCCRzybw=GDLfGC9bZz0v0tfGSg@mail.gmail.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-jdbc |
Vitalii>But do we already use SO_KEEPALIVE on the socket layer?
Vitalii>Also I am not really sure socket was closed on the server end.
Vitalii> TS said he could see server backend in the idle state. Also backed
server did not fail, so unless there were some network glitch we should
have received tcp reset.
tcpKeepAlive connection setting is false by default. I think it's better to
make "true" the default value.
Sehrope>I don't think it's a good idea to flood the server with dummy
messages for the purposes of keeping the connection alive.
Sehrope>Besides the usual problems of filling up the send buffer, you'd
also have to have it done in a separate thread
Sehrope>as the usual sender is already blocked waiting for the command
response.
Sending a sync message once every 10m will hardly fill up the buffers.
On the other hand, it could make the connection more resilient.
Having several threads for (e.g. one for read and one for write) would
resolve "deadlock out of buffers got full" issue, so it is good no matter
if we add "ping" there.
Vladimir
From | Date | Subject | |
---|---|---|---|
Next Message | jingzhi.zhang@outlook.com | 2016-06-07 13:05:21 | Re: PrepareStatement.execute() blocked because of long time 'create index' operation, connection leak |
Previous Message | Vitalii Tymchyshyn | 2016-06-07 12:34:33 | Re: PrepareStatement.execute() blocked because of long time 'create index' operation, connection leak |