Joe Witt created NIFI-11575:
-------------------------------
Summary: When you terminate a component - it increases the thread
pool size and does not restore it until restart
Key: NIFI-11575
URL: https://issues.apache.org/jira/browse/NIFI-11575
Project: Apache NiFi
Issue Type: Improvement
Components: Core Framework
Affects Versions: 1.21.0, 2.0.0
Reporter: Joe Witt
Assignee: Mark Payne
Currently when you stop a component and then choose to terminate active threads
we do this by expanding the core pool size by the number of currently running
threads for that component. So if you have a pool size of 10 as configured in
the flow controller and you choose to terminate a processor that currently has
5 running tasks the flow controller will now be updated to allow 15 threads.
This happens only at runtime it doesn't change the config so the issue will
resolve itself on restart.
Why we need to do this makes sense. The stuck threads you wish to terminate
may in fact never stop so these new threads replace those lost threads and work
can continue. And indeed termination in general is an 'in case of emergency
break glass' operation. But assuming termination works out we should return to
the desired pool size.
This becomes even more important on a system that is already resource saturated
where termination might be a necessary/helpful tool in reworking the live flow
and as of now doing so actually increase the thread pool size which increases
the amount of work that can be done in parallel thus making the situation
harder. Fixing this logic will help termination be useful without expanding
the resource burden.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
