[rhq-project/rhq] 3bf310: [1053888] Deadlock in update RHQ_RESOURCE_AVAIL
Branch: refs/heads/RHQ_4_9_0_JON323GA/bug/1367491
Home: https://github.com/rhq-project/rhq
Commit: 3bf310242a26d62bda31578c73c77432420f16b5
https://github.com/rhq-project/rhq/commit/3bf310242a26d62bda31578c73c7743...
Author: Jay Shaughnessy <jshaughn(a)redhat.com>
Date: 2016-08-23 (Tue, 23 Aug 2016)
Changed paths:
M
modules/enterprise/server/jar/src/main/java/org/rhq/enterprise/server/discovery/DiscoveryBossBean.java
M
modules/enterprise/server/jar/src/main/java/org/rhq/enterprise/server/measurement/AvailabilityManagerBean.java
M
modules/enterprise/server/jar/src/main/java/org/rhq/enterprise/server/measurement/AvailabilityManagerLocal.java
Log Message:
-----------
[1053888] Deadlock in update RHQ_RESOURCE_AVAIL
I didn't try to reproduce this but after scouring the code found one
place we looked vulnerable. In general we protect against locking (and
provide ordering) by serializing avail report processing for each agent.
We were not applying this locking when ignoring resources; subsequently
setting the resources UNKNOWN, cross-agent. This may explain the behavior
although it's also possible that non-row DB-level locking could have caused
this, perhaps. Anyway, the applied solution is to just not update the avails
for the ignored resources. This is likely fine but if it confuses anyone
(at uningnore time) I've updated DiscoveryBossBean.setResourceAvailabilities()
ina fashion that would make it safe to be called.
(cherry picked from commit f19d97ba62416ab9d333c3cf4a5e1b28bf0a6d48)
Also added missing imports from unrelated commits.