Description of problem:
(a) Cluster locking leaves hundreds of sockets open in the TIME_WAIT state on
clusters with many services. Switching to UNIX domain sockets would not only
improve performance, but also significantly reduce this number
(b) Cluster lock is not correctly released from cludb in all cases.
(c) Clulockd panics during unlock if the connection times out in this case.
Version-Release number of selected component (if applicable): 1.2.28
Steps to Reproduce:
(a) Create 10+ services and watch the number of sockets in TIME_WAIT as output
by netstat -an .
(b) Indeterminate / difficult to reproduce. This is a timing issue.
(c) Difficult, but possible to reproduce. Edit the clulockd code to sleep(15)
or so after replying to a LOCK_LOCK request (with a LOCK_ACK...). The
connecting node will fail to release the lock and raise:
<alert> Unhandled Exception at clulock.c:408 in _clu_process_unlock
Actual results: Stated.
Created attachment 120909 [details]
Fixes all three conditions
Setting to MODIFIED
A new *TEST* package build with several bugfixes (fixes bugzillas: 171637 172735
172893 172894 ) is available. Gulm-bridge support has been disabled in this
release to prevent having to install with the "--nodeps" option:
Let us know if this works for you.
An advisory has been issued which should help the problem
described in this bug report. This report is therefore being
closed with a resolution of ERRATA. For more information
on the solution and/or where to find the updated files,
please follow the link below. You may reopen this bug report
if the solution does not work for you.