Problem

Jira has an issue (covered by JRASERVER-64908) which is the following: when loading/reloading a plugin/module Jira flushes a number of caches.

In Jira DC this problem was even intensified as any flush on a remote cache happening on every node is being replicated to all other nodes. This issue is only covering this problem which is specific to Jira DC.

Fix

Any operation which is performed on a remote cache when doing a plugin/module load/reload is not replicated to other nodes. Note that the plugin/module reload event is still send to all nodes so the cache clear will still happen on all nodes (JRASERVER-64908).

Logging

When setting DEBUG logging on the com.atlassian.jira.plugin.JiraCacheResetter the following logs will appear when loading/reloading a plugin/module:

[thread1] Start resetting caches triggered by: {}, jiraIsShuttingDown: {}
...
[thread1] Done resetting caches triggered by: {}, timeMillis: {}, jiraIsShuttingDown: {} 

All remote cache operations which are happening between those 2 log lines should not and are not replicated (with this fix these are local only operations).

Hey James, jhunt

Thanks for your comment.
You mentioned: with only one node active. That effectively makes your Jira setup during that period of time non-DC, so better description for the problem should be another ticket JRASERVER-64908, which is focusing on UPM action itself.

Still, there is something got my attention: triggered the JiraCacheResetter storm in the logs.
You are taking about those events: com.atlassian.jira.plugin.JiraCacheResetter$Delegate.onPluginModule<Action> ?
Can you please share small snippet from the logs so we can verify that? (if you prefer not to do this in public ticket, please let me know we can raise separate support request)

Thanks.
Cheers

Encountered today after upgrading from Jira Software 7.2.12 Data Center to Jira Software 7.6.4 Data Center. Updated Jira Service Desk via OBR (with only one node active and only one user logged in) and triggered the JiraCacheResetter storm in the logs.

Encountered this with 7.4.3. Luckily the updates were happening during evening time. Problem was resolved by rebooting a node.

I remember in 7.2.8 I think where updating a plugin made our Jira DC insfance unavailable for 30 mins.