IDP V3: LockTimeoutException with remote PG DB

[Marvin Addison]

>
> The following observation appears only when IDP is asking PG DB server
> remotely. I tried different (in global.xml) maxActive and maxIdle jdbc
> connectors (up to 32) without solving the problem. The problem does not
> appear when IDP is asking PG DB locally.
>

I took some time to attempt to reproduce the error on a test platform, but
was unable. It's a pretty synthetic test using a VM with everything (IdP,
OpenLDAP, and PG) running on the same box, but I hit it with enough
concurrent load that if there were a fundamental problem with the locking
code, it should have showed up.

In particular, I was interested in the interaction of the cleaner process
with front-end threads. I turned the session timeout and authn result
timeouts to 1m, turned the cleaner poll process down to 30s, and ran the
test for 5m, which ought to have been long enough to mix in several cleaner
processes with request threads. I simply couldn't generate failures.

I should note that I used HikariCP for the pooling library with default
settings (10 connections max).

Your comment about the problem not showing up when the DB is local is
puzzling. Perhaps there's some subtle but very bad behavior when
application locks are held marginally longer, which I would expect would be
the case for connections to a remote host. I don't have any more time to
spend on further investigation at the moment, but I believe we'll need more
data to investigate further.

M <users-unsubscribe at shibboleth.net>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://shibboleth.net/pipermail/users/attachments/20151008/480e590b/attachment.html>