Trac leaks db connections with PgSQL and mod_python

[jaa-trac@…]

I installed Trac from trunk @ r1757, and when I used it with mod_python, I found that Trac is leaking 1 db connections per any action on Trac site (klick roadmap ⇒ +1 dbconn, klick timeline ⇒ +1 dbconn etc.)

[Christopher Lenz]

We're using a connection pool that will allocate up to 5 connections. Are you seeing the number of connections increase beyond that maximum?

[jaa-trac@…]

yes, I thought that, 18 was when I stopped testing, because it was a kind-of-production server. =)

I think that roughly half of those were 'idle in transaction', and rest were just 'idle'. I was the only person/client who used trac at the moment.

[Christopher Lenz]

Can you please check whether this problem still exists after [1804]? There have been a couple of fixes to the connection pooling since the time this ticket was created.

[tjb@…]

The problem still exists as at [2006]. I ran up our mod_python interface, accessed trac, and watched the postgres connections - at 19 connections I stopped. There was a definite pattern of 1 connection left open for each transaction.

[jaa]

Added me to CC: field. Why don't I get updates of this ticket, even when I am the reporter?

[jaa-trac@…]

I finally had time to upgrade our system to 0.9.x and I don't see this anymore. The system uses at the moment: Trac 0.9-stable@2529, pygpsql 2.4, pgsql 8.0.4, python 2.4.2, mod_python 3.1.4.

The upgrade was really smoosh, only hiccups was db16.py upgrade, and it was an easy one to fix (I had primary keys already in the table).

Very nice work, thanks!