DBD rhobustness

[jnorell]

I left dbmail-postfix-policyd running idle over the weekend, connected to a postfix back end. When I set pointed postfix to it this morning, it immediately died with:

DBD::Pg::st execute failed: server closed the connection unexpectedly
 This probably means the server terminated abnormally
 before or while processing the request. at ./dbmail-postfix-policyd line 153.

Need to make the dbi pieces more rhobust: attempt reconnect to server, check for errors on function calls, etc.

[jnorell]

Reproduced that error again (now on line 183) by leaving dbmail-postfix-policyd running overnight without postfix forwarding mail to it. No errors in postgresql main.log. Another active dbmail-postfix-policyd ran overnight connected to the same database backend, without problem. Definitely something related to idle connections here.

[jnorell]

Still happening even after yesterday's commit which intended to fix this. (error on line 225 now)

[jnorell]

Still not fixed on the idle postgres connection, though it's improved. It no longer dies, but it does stop checking, with messages like:

dbmail_v1_quota_check: checking addr@domain (recurse_depth 0)
DBD::Pg::st execute failed: no connection to the server at ./dbmail-postfix-policyd line 262.
dbmail_v1_quota_check: alias_sth: DBI::st=HASH(0x86c1dd8)->errstr

A HUP will reconnect to the database and all starts working again.

[jnorell]

In more digging, this is probably an tcp keepalive issue - I'll see about setting SO_KEEPALIVE on the database socket.