HDDS-6685. Follower OM crashed when validating S3 auth info.

[ChenSammi]

https://issues.apache.org/jira/browse/HDDS-6685

[nandakumar131]

+1, the changes look good.

[adoroszlai]

Thanks @ChenSammi for working on this.

Can you please check TestOzoneManagerPrepare? Seems to be failing consistently (also locally).

[ChenSammi]

Thanks @ChenSammi for working on this.

Can you please check TestOzoneManagerPrepare? Seems to be failing consistently (also locally).

TestOzoneManagerPrepare is passed locally on my laptop. I just pushed a new patch. Let's see if TestOzoneManagerPrepare still fails in CI.

[ChenSammi]

Thanks @nandakumar131 and @adoroszlai for the code review.

[kaijchen]

Hi @ChenSammi, seems this PR breaks TestOzoneManagerPrepare#testPrepareDownedOM, could you please take a look?

• With this change: https://github.com/kaijchen/ozone/actions/runs/2273676445
• Without this change: https://github.com/kaijchen/ozone/actions/runs/2273677044

[adoroszlai]

@ChenSammiTestOzoneManagerPrepare was flaky before (HDDS-5990), but it is definitely much worse now than previously.

Seems like stop/start logic is not entirely fail-safe, or test needs additional/adjusted waitFor checks?

2022-05-05 01:41:52,188 [pool-2338-thread-1] ERROR om.OzoneManager (ExitUtils.java:terminate(133)) - Terminating with exit status 1: Failed to start RPC Server.
java.net.BindException: Problem binding to [localhost:41487] java.net.BindException: Address already in use
...
at org.apache.hadoop.ozone.om.OzoneManager.startRpcServer(OzoneManager.java:1084)
at org.apache.hadoop.ozone.om.OzoneManager.getRpcServer(OzoneManager.java:1054)
at org.apache.hadoop.ozone.om.OzoneManager.installCheckpoint(OzoneManager.java:3366)
at org.apache.hadoop.ozone.om.OzoneManager.installCheckpoint(OzoneManager.java:3254)
at org.apache.hadoop.ozone.om.OzoneManager.installSnapshotFromLeader(OzoneManager.java:3231)

https://github.com/adoroszlai/ozone-build-results/tree/master/2022/05/05/14826/it-flaky/hadoop-ozone/integration-test

[adoroszlai]

Idea for fixing the bind exception: adoroszlai@65e11fa

Running 10 iterations:
https://github.com/adoroszlai/hadoop-ozone/runs/6302199651

Let's see if it's enough.

[adoroszlai]

Yes, it seems bind exception is fixed. But testPrepareDownedOM still times out at assertClusterNotPrepared, because the restarted OM does not get the cancel prepare request. It is stuck after installing snapshot, thinking it is still in progress:

2022-05-05 10:01:07,994 [Listener at localhost/49178] INFO om.OzoneManager (OzoneManager.java:installCheckpoint(3396)) - Install Checkpoint is finished with Term: 2 and Index: 96. Spend 528 ms.
2022-05-05 10:01:07,994 [Listener at localhost/49178] INFO server.RaftServer$Division (RaftServerImpl.java:lambda$notifyStateMachineToInstallSnapshot$30(1595)) - omNode-2@group-523986131536: StateMachine successfully installed snapshot index 96. Reloading the StateMachine.
...
2022-05-05 10:01:08,682 [grpc-default-executor-2] INFO server.RaftServer$Division (RaftServerImpl.java:checkInconsistentAppendEntries(1296)) - omNode-2@group-523986131536: Failed appendEntries as snapshot (97) installation is in progress

[adoroszlai]

Upgrade to current Ratis 2.3.0-SNAPSHOT (current master) with Ratis Thirdparty 1.0.0 (pending release) seems to resolve the problem.

[kaijchen]

I have just run 100x testPrepareDownedOM with this PR reverted, and surprisingly got a clean run: https://github.com/kaijchen/ozone/actions/runs/2275278052

I guess there might be some issues around restarting the omRpcServer.

[adoroszlai]

HDDS-10177 seems to be caused by this change.