Red Hat Bugzilla – Bug 456248
DNS outage can cause cluster to die
Last modified: 2010-10-22 23:02:25 EDT
Description of problem:
If a cluster member is the sole survivor in a 2-node cluster and a transient DNS
error occurs such that the cluster software can not do resolution of the IP
address, the cluster can "fail" to ping the tiebreaker.
Version-Release number of selected component (if applicable): 1.2.37-1.1
How reproducible: Rare
Steps to Reproduce:
1. Introduce transient network errors causing DNS to be unreachable
Actual results: cluster outage
Expected results: cluster stays up
Additional info: The cluster software MUST NEVER rely on DNS. This behavior is
unacceptable. The ping code does a gethostbyname() call when it should not.
It's a simple fix.
Created attachment 312347 [details]
Shortcuts DNS lookup if the IP tiebreaker is in dotted quad form
This patch does not introduce any regressions into the package.
This patch does not introduce rolling-upgrade incompatibilities.
This bug could potentially affect any user who is running with an IP tiebreaker.
The patch is low-impact and the work has been done. I recommend including it.
You can also short-circuit DNS by placing the IP tiebreaker in /etc/hosts - even
when specifying the dotted quad address instead of the hostname.
Because this workaround exists, I am not going to fix this bugzilla.