[CASSANDRA-1463] Failed bootstrap can cause NPE in batch_mutate on every node, taking down the entire cluster

• Type: Bug
• Status: Resolved
• Priority: Normal
• Resolution: Fixed
• Fix Version/s: 0.6.6, 0.7 beta 2
• Severity: Normal
• Assignee: Jonathan Ellis
• Reporter: David King

Authors

Jonathan Ellis

Reviewers

Brandon Williams

• Created: 03/Sep/10 20:50
• Updated: 16/Apr/19 09:33
• Resolved: 04/Sep/10 20:34

Description

In adding a node to the cluster, the bootstrap failed (still investigating the cause). An hour later, the entire cluster failed, preventing any writes from being accepted. This exception started being printed to the logs:

INFO [Timer-0] 2010-09-03 12:23:33,282 Gossiper.java (line 402) FatClient /10.251.243.191 has been silent for 3600000ms, removing from gossip
ERROR [Timer-0] 2010-09-03 12:23:33,318 Gossiper.java (line 99) Gossip error
java.util.ConcurrentModificationException
at java.util.Hashtable$Enumerator.next(Hashtable.java:1048)
at org.apache.cassandra.gms.Gossiper.doStatusCheck(Gossiper.java:383)
at org.apache.cassandra.gms.Gossiper$GossipTimerTask.run(Gossiper.java:93)
at java.util.TimerThread.mainLoop(Timer.java:534)
at java.util.TimerThread.run(Timer.java:484)
ERROR [pool-1-thread-69153] 2010-09-03 12:23:33,857 Cassandra.java (line 1659) Internal error processing batch_mutate
java.lang.NullPointerException
at org.apache.cassandra.gms.FailureDetector.isAlive(FailureDetector.java:135)
at org.apache.cassandra.locator.AbstractReplicationStrategy.getHintedEndpoints(AbstractReplicationStrategy.java:85)
at org.apache.cassandra.service.StorageProxy.mutateBlocking(StorageProxy.java:204)
at org.apache.cassandra.thrift.CassandraServer.batch_mutate(CassandraServer.java:415)
at org.apache.cassandra.thrift.Cassandra$Processor$batch_mutate.process(Cassandra.java:1651)
at org.apache.cassandra.thrift.Cassandra$Processor.process(Cassandra.java:1166)
at org.apache.cassandra.thrift.CustomTThreadPoolServer$WorkerProcess.run(CustomTThreadPoolServer.java:167)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1110)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:603)
at java.lang.Thread.run(Thread.java:636)
ERROR [pool-1-thread-69154] 2010-09-03 12:23:33,869 Cassandra.java (line 1659) Internal error processing batch_mutate
java.lang.NullPointerException
at org.apache.cassandra.gms.FailureDetector.isAlive(FailureDetector.java:135)
at org.apache.cassandra.locator.AbstractReplicationStrategy.getHintedEndpoints(AbstractReplicationStrategy.java:85)
at org.apache.cassandra.service.StorageProxy.mutateBlocking(StorageProxy.java:204)
at org.apache.cassandra.thrift.CassandraServer.batch_mutate(CassandraServer.java:415)
at org.apache.cassandra.thrift.Cassandra$Processor$batch_mutate.process(Cassandra.java:1651)
at org.apache.cassandra.thrift.Cassandra$Processor.process(Cassandra.java:1166)
at org.apache.cassandra.thrift.CustomTThreadPoolServer$WorkerProcess.run(CustomTThreadPoolServer.java:167)
at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1110)
at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:603)
at java.lang.Thread.run(Thread.java:636)

After a large number of iterations of that (at least thousands), the printed exception was shortened (this shortening is what made me mistakenly file #1462) to

ERROR [pool-1-thread-68869] 2010-09-03 12:39:22,857 Cassandra.java (line 1659) Internal error processing batch_mutate
java.lang.NullPointerException
ERROR [pool-1-thread-68869] 2010-09-03 12:39:22,883 Cassandra.java (line 1659) Internal error processing batch_mutate
java.lang.NullPointerException
ERROR [pool-1-thread-68869] 2010-09-03 12:39:22,894 Cassandra.java (line 1659) Internal error processing batch_mutate
java.lang.NullPointerException
ERROR [pool-1-thread-68970] 2010-09-03 12:39:22,985 Cassandra.java (line 1659) Internal error processing batch_mutate
java.lang.NullPointerException
ERROR [pool-1-thread-68970] 2010-09-03 12:39:23,084 Cassandra.java (line 1659) Internal error processing batch_mutate
java.lang.NullPointerException

Rolling a restart over the cluster fixed it, but every node had to be restarted before it started accepting writes again.