[CASSANDRA-19934] NPE in AffectedRangesImpl#intersects when other has different keyspaces than local

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 5.x
• Component/s: Transactional Cluster Metadata
• Bug Category: Correctness - Recoverable Corruption / Loss
• Severity: Critical
• Complexity: Low Hanging Fruit
• Discovered By: Fuzz Test
• Platform: All
• Assignee: Unassigned
• Reporter: David Capwell
• Created: 18/Sep/24 21:27
• Updated: 20/Sep/24 16:24
• Resolved: 20/Sep/24 16:24

Description

This error was found on Accord branch.

 org.apache.cassandra.simulator.SimulationException: Failed on seed 0xfa3d51da237d56e5; Failure creating the simulation
Caused by: java.lang.AssertionError: Errors detected during simulation
	Suppressed: java.lang.NullPointerException
		at org.apache.cassandra.tcm.sequences.LockedRanges$AffectedRangesImpl.intersects(LockedRanges.java:337)
		at org.apache.cassandra.tcm.sequences.LockedRanges.intersects(LockedRanges.java:97)
		at org.apache.cassandra.tcm.transformations.PrepareLeave.execute(PrepareLeave.java:108)
		at org.apache.cassandra.tcm.AbstractLocalProcessor.executeStrictly(AbstractLocalProcessor.java:167)
		at org.apache.cassandra.tcm.AbstractLocalProcessor.commit(AbstractLocalProcessor.java:77)
		at org.apache.cassandra.distributed.test.log.TestProcessor.commit(TestProcessor.java:62)
		at org.apache.cassandra.tcm.Commit$Handler.doVerb(Commit.java:368)
		at org.apache.cassandra.net.InboundSink.lambda$new$0(InboundSink.java:102)
		at org.apache.cassandra.net.InboundSink$Filtered.accept(InboundSink.java:70)
		at org.apache.cassandra.net.InboundSink$Filtered.accept(InboundSink.java:56)
		at org.apache.cassandra.net.InboundSink.accept(InboundSink.java:122)

In debugger I see the following

other = {LockedRanges$AffectedRangesImpl@92088} "AffectedRangesImpl{map={ReplicationParams{class=org.apache.cassandra.locator.MetaStrategy, dc0=2, dc2=2, dc1=2}=[(-9223372036854775808,-9223372036854775808]]}}"
 map = {HashMap@93801}  size = 1
  {ReplicationParams@92097} "ReplicationParams{class=org.apache.cassandra.locator.MetaStrategy, dc0=2, dc2=2, dc1=2}" -> {HashSet@92098}  size = 1

map = {HashMap@92092}  size = 3
 {ReplicationParams@93810} "ReplicationParams{class=org.apache.cassandra.locator.SimpleStrategy, replication_factor=2}" -> {HashSet@93811}  size = 3
 {ReplicationParams@93812} "ReplicationParams{class=org.apache.cassandra.locator.SimpleStrategy, replication_factor=3}" -> {HashSet@93813}  size = 4
 {ReplicationParams@93814} "ReplicationParams{class=org.apache.cassandra.locator.NetworkTopologyStrategy, dc0=3, dc2=3, dc1=3}" -> {HashSet@93815}  size = 7

Locally we are missing the meta keyspace so can’t find it from the map and NPE.

The reason for this is that the current node is starting up

StorageService.instance.operationMode == STARTING

Issue Links

• is fixed by: CASSANDRA-19904 Deprecate use of gossip state for paxos electorate verification