[KAFKA-4478] Deadlock between heartbeat executor, group metadata manager and request handler

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Duplicate
• Affects Version/s: 0.10.1.0
• Labels: reliability
• Assignee: Unassigned
• Reporter: Sinóros-Szabó Péter
• Created: 02/Dec/16 13:00
• Updated: 03/Dec/16 00:12
• Resolved: 03/Dec/16 00:12

Description

We are running a 0.10.1.0 cluster with 3 brokers with ids 0, 1 and 2.
At about 2016-12-01 21:29 something happened with broker 1 and since then I see

java.io.IOException: Connection to 1 was disconnected before the response was read

errors in the logs of broker 0 and 2. Clients were unable to produce to broker 1's partitions and JMX counters indicates underreplicated partitions.

I took a stack trace on broker-1 and I see that there is a deadlock between the JVM threads:

Found one Java-level deadlock:
=============================
"executor-Heartbeat":
  waiting to lock monitor 0x00007ffa24029df8 (object 0x00000000cc52fe70, a kafka.coordinator.GroupMetadata),
  which is held by "group-metadata-manager-0"
"group-metadata-manager-0":
  waiting to lock monitor 0x00007ff9900a83a8 (object 0x00000000ca8b0820, a java.util.LinkedList),
  which is held by "kafka-request-handler-7"
"kafka-request-handler-7":
  waiting to lock monitor 0x00007ffa24029df8 (object 0x00000000cc52fe70, a kafka.coordinator.GroupMetadata),
  which is held by "group-metadata-manager-0"

Issue Links

• duplicates: KAFKA-3994 Deadlock between consumer heartbeat expiration and offset commit.