[CASSANDRA-16673] Avoid race in AbstractReplicationStrategy endpoint caching

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 3.0.26, 3.11.12, 4.0.2, 4.1-alpha1, 4.1
• Component/s: Cluster/Membership
• Bug Category: Availability - Unavailable
• Severity: Low
• Complexity: Normal
• Discovered By: Fuzz Test
• Platform: All
• Since Version: 3.0 alpha 1

Source Control Link

https://github.com/apache/cassandra/commit/b1a8a56c563b85ab9a34d3bbf9c16278dd441157

Test and Documentation Plan

new test, cci run

• Assignee: Marcus Eriksson Assign to me
• Reporter: Marcus Eriksson

Authors

Marcus Eriksson

Reviewers

Alex Petrov, Jon Meredith

• Created: 18/May/21 11:45
• Updated: 27/May/22 19:25
• Resolved: 17/Jan/22 13:20

Description

We should make sure we track which ringVersion we are caching in AbstractReplicationStrategy to avoid a race where we might return the wrong EndpointsForRange.

Caused by: java.lang.IllegalArgumentException: 9010454139840013625 is not contained within (9223372036854775801,-4611686018427387905]
	at org.apache.cassandra.locator.EndpointsForRange.forToken(EndpointsForRange.java:59)
	at org.apache.cassandra.locator.AbstractReplicationStrategy.getNaturalReplicasForToken(AbstractReplicationStrategy.java:104)
	at org.apache.cassandra.locator.ReplicaLayout.forTokenReadLiveSorted(ReplicaLayout.java:330)
	at org.apache.cassandra.locator.ReplicaPlans.forRead(ReplicaPlans.java:594)