[FLINK-34496] Classloading deadlock between ExecNodeMetadataUtil and JsonSerdeUtil

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.18.1
• Fix Version/s: 1.19.0, 1.18.2
• Component/s: Table SQL / Planner
• Labels: pull-request-available
• Assignee: Chesnay Schepler
• Reporter: Chesnay Schepler
• Created: 22/Feb/24 09:43
• Updated: 23/Feb/24 09:32
• Resolved: 23/Feb/24 09:32

Description

This is a fun one!

ExecNodeMetadataUtil and JsonSerdeUtil have a circular dependency in their static initialization, which can cause a classloading lockup when 2 threads are running the class initialization of each class at the same time because during class initialization they hold a lock.

https://ternarysearch.blogspot.com/2013/07/static-initialization-deadlock.html

JsonSerdeUtils#createFlinkTableJacksonModule calls into the ExecNodeMetadataUtil, while ExecNodeMetadataUtil#addToLookupMap calls into the JsonSerdeUtils.

 "ForkJoinPool-3-worker-11" #25 daemon prio=5 os_prio=0 cpu=219.87ms elapsed=995.99s tid=0x00007ff11c50e000 nid=0xf0fc in Object.wait()  [0x00007ff12a4f3000]
    java.lang.Thread.State: RUNNABLE
 	at o.a.f.t.p.plan.nodes.exec.serde.JsonSerdeUtil.createFlinkTableJacksonModule(JsonSerdeUtil.java:133)
 	at o.a.f.t.p.plan.nodes.exec.serde.JsonSerdeUtil.<clinit>(JsonSerdeUtil.java:111)

"ForkJoinPool-3-worker-7" #23 daemon prio=5 os_prio=0 cpu=54.83ms elapsed=996.00s tid=0x00007ff11c50c000 nid=0xf0fb in Object.wait()  [0x00007ff12a5f4000]
   java.lang.Thread.State: RUNNABLE
	at o.a.f.t.p.plan.utils.ExecNodeMetadataUtil.addToLookupMap(ExecNodeMetadataUtil.java:235)
	at o.a.f.t.p.plan.utils.ExecNodeMetadataUtil.<clinit>(ExecNodeMetadataUtil.java:156)

Issue Links

• links to: GitHub Pull Request #24371