Remote TLS queries to the CIB leaks filedescriptors

[Dag Stenstad]

Binary package hint: pacemaker

I have some scripts that regularly poll cluster status remotely, and have
noticed that they stop working after a while. Investigations seems to point to
cib leaking filedescriptors:

cib 5341 hacluster 1016u IPv4 197563 0t0 TCP
node1:5225->adminserver:38057 (CLOSE_WAIT)
cib 5341 hacluster 1017u IPv4 197580 0t0 TCP
node1:5225->adminserver:38064 (CLOSE_WAIT)
cib 5341 hacluster 1018u IPv4 197840 0t0 TCP
node1:5225->adminserver:51987 (CLOSE_WAIT)

.. times 887, give or take. cib runs out of filedescriptors.

Currently running 1.0.8+hg15494-2ubuntu2 on Ubuntu Lucid.

Reported upstream:

http://developerbugs.linux-foundation.org/show_bug.cgi?id=2469

Please note that this basically breaks cib if you poll it remotely.

[Ante Karamatić (ivoks)]

Did you try patch from the upstream?

[Dag Stenstad (dag-stenstad)]

Sorry, I've been on leave for a while, and I can't touch the cluster
right now.

If you have a cluster to test on, all it takes is using any of the
tools from remote to trigger the leak. You can see how the number of
open filedescriptors increase in /proc/<pid>/fd/ for the "cib" process.

[Serge Hallyn (serge-hallyn)]

Is this bug still valid?

[Dag Stenstad (dag-stenstad)]

No idea, I haven't touched it in ages.

Let's just close it.

[Launchpad Janitor (janitor)]

[Expired for pacemaker (Ubuntu) because there has been no activity for 60 days.]