[CASSANDRA-18725] IsolatedJMX should not release all TCPEndpoints on instance shutdown - ASF JIRA

XML

Word

Printable

JSON

Details

Type: Improvement
Status: Resolved
Priority: Normal
Resolution: Fixed
Fix Version/s: 3.11.17, 4.0.12, 4.1.4, 5.0-alpha1, 5.x
Component/s: Test/dtest/java
Labels:
None

Change Category:
Operability
Complexity:
Challenging
Platform:

All
Impacts:

None
Source Control Link:

https://github.com/apache/cassandra/commit/c6d7d070c59d81db8949683d3e5670b909efb48c
Test and Documentation Plan:

Hide

Tests are updated in the dtest JMXFeatureTest and JMXGetterCheckTest for each branch.

Show
Tests are updated in the dtest JMXFeatureTest and JMXGetterCheckTest for each branch.

Description

In the original implementation of the JMX feature, we fixed some memory leaks by clearing some internal state in Java’s TCPEndpoint. However, that implementation was overly aggressive and cleared the whole map, vs. just removing the endpoints created by the individual instances. This causes issues when you remove a node from the cluster (as all of the endpoints are cleared, not just the ones in use by that instance).

In stead, we should check if the endpoint was created by the instance in question and only remove it if it was.