Red Hat Bugzilla – Bug 445585
multipath does not configure correctly when the ratio of paths / maps to system resources increases
Last modified: 2010-10-21 15:47:55 EDT
Description of problem:
This is a generic bz meant to track proper configuration of multipath device
maps. There are multiple problems with getting all maps to configure correctly.
Sometimes the maps will be created without all paths, sometimes the maps will
be created with a WWID name instead of the friendly name, etc. I believe at
least some of these problems are still present in upstream code.
Version-Release number of selected component (if applicable):
Some problems are hard to reproduce. Many are related to bursty, concurrent
executions of the multipath tool caused by many udev events happening in a short
period of time. I've been able to reproduce some of the problems with xen +
iscsi so specific hardware might not be required.
Steps to Reproduce:
1. Configure multipath for a large number of LUNs + paths
2. Start multipath
Not all paths / maps are created
All paths and maps created properly
Since this is a generic bugzilla for tracking issues hit when large no. of luns/paths (256 luns x 4 = 1024 paths) are mapped to the RHEL host, I am adding a few myself that we have observed for such scenarios:
1) OOM/Hang issue - https://bugzilla.redhat.com/show_bug.cgi?id=459969
2) Multipath creates wwid based names instead of mpath names despite enabling user_friendly_names - https://bugzilla.redhat.com/show_bug.cgi?id=445268
3) Multipath does not automatically remove maps corresponding to the deleted luns - https://bugzilla.redhat.com/show_bug.cgi?id=443380
4) Delay in scanning large no. of luns/paths - https://bugzilla.redhat.com/show_bug.cgi?id=460301
5) Multipathd daemon ends up dead with the message "multipathd dead but pid file exists". This is highly intermittent. - No bugzilla yet.
I'm closing this bug since all the specific bugs it referenced are now closed.