Red Hat Bugzilla – Bug 1030111
Recovery alert cache refresh needs to happen prior to alert notification processing
Last modified: 2014-01-02 15:35:02 EST
From a user's environment we can see a 3-4 second delay between when the alert is fired and the cache reload request is made due to the time it takes to execute the alert notifications. Therefore, to ensure a higher level of alert stability, this issue needs to be fixed in the JBoss ON as well. +++ This bug was initially created as a clone of Bug #1028487 +++ Currently alert notifications are processed in the same transaction as the alert creation, recovery alert activation, etc. There are a few issues related to the notif processing taking place in this transaction: 1) It simply extends the length of an already complicated transaction, potentially holding locks, delaying the alert commit, delaying the global cache refresh flags being committed (such that HA servers can pick up on it), etc... 2) Notifications can actually initiate recovery actions such as executing resource operations, invoking CLI scripts, etc. This should not happen prior to our ability to update the global alert condition cache, which must happen to begin condition matching on activated recovery alert definitions. Otherwise we risk an actual recovery happening prior to the recovery alert being ready. The alert notification processing should happen outside of the alert creation transaction and after cache refresh. --- Additional comment from Jay Shaughnessy on 2013-11-08 10:57:06 EST --- master commit 28becd282f8cd3ed4327a56ea0c08f8431845dba Author: Jay Shaughnessy <jshaughn@redhat.com> Date: Fri Nov 8 10:42:46 2013 -0500 Restructure SLSB methods (locals only, no remote changes) to process alert notifications later in the workflow, after the alert is committed and after we have a chance to update the condition caches (for more reliable recovery alerting). Needed to be able to pass back the new alert through the call chain.
Test Case: The test case for Bug 1030108 is also effective for testing the code paths for this bug. Additionally, ensure every type of notification is successfully executed. release/jon3.2.x commit e6320059f3bc44dcc9b5f4b3ca348460a1556e2f Author: Jay Shaughnessy <jshaughn@redhat.com> Date: Thu Nov 14 11:11:55 2013 -0500 Restructure SLSB methods (locals only, no remote changes) to process alert notifications later in the workflow, after the alert is committed and after we have a chance to update the condition caches (for more reliable recovery alerting). Needed to be able to pass back the new alert through the call chain. Cherry-Pick Master: 28becd282f8cd3ed4327a56ea0c08f8431845dba
Moving to ON_QA as available for testing with new brew build.
Mass moving all of these from ER6 to target milestone ER07 since the ER6 build was bad and QE was halted for the same reason.
Created attachment 833674 [details] alert.png
Created attachment 833675 [details] server.log.png
verified in 3.2 GA - cache reload is called after alert fired and before the recovery alert