Alexey Serbin created KUDU-2320:
-----------------------------------

             Summary: Exponential back-off is not applied while retrying tablet 
replica deletion
                 Key: KUDU-2320
                 URL: https://issues.apache.org/jira/browse/KUDU-2320
             Project: Kudu
          Issue Type: Bug
          Components: master, tserver
    Affects Versions: 1.7.0
            Reporter: Alexey Serbin
         Attachments: raft_consensus_nonvoter-itest.txt.bz2

The {{RaftConsensusNonVoterITest::RestartClusterWithNonVoter}} scenario of 
\{{raft_consensus_nonvoter-itest}} creates a situation when one tablet server 
is shutdown and the catalog manager endlessly retries to delete a tablet from 
that server, outputting huge amount of messages like below:
{noformat}
W0218 11:09:50.375723 26207 catalog_manager.cc:2648] Async tablet task 
83456f56fde245c29db20ffa967275f1 Delete Tablet RPC for 
TS=293703ef5e1840e39f9b9a260ddae65b failed: Not found: failed to reset TS 
proxy: Could not find TS for UUID 293703ef5e1840e39f9b9a260ddae65b
I0218 11:09:50.404064 26207 catalog_manager.cc:2629] Scheduling retry of 
83456f56fde245c29db20ffa967275f1 Delete Tablet RPC for 
TS=293703ef5e1840e39f9b9a260ddae65b with a delay of 45 ms (attempt = 0)
W0218 11:09:50.404109 26207 catalog_manager.cc:2648] Async tablet task 
83456f56fde245c29db20ffa967275f1 Delete Tablet RPC for 
TS=293703ef5e1840e39f9b9a260ddae65b failed: Not found: failed to reset TS 
proxy: Could not find TS for UUID 293703ef5e1840e39f9b9a260ddae65b
I0218 11:09:50.449461 26207 catalog_manager.cc:2629] Scheduling retry of 
83456f56fde245c29db20ffa967275f1 Delete Tablet RPC for 
TS=293703ef5e1840e39f9b9a260ddae65b with a delay of 13 ms (attempt = 0)
W0218 11:09:50.449510 26207 catalog_manager.cc:2648] Async tablet task 
83456f56fde245c29db20ffa967275f1 Delete Tablet RPC for 
TS=293703ef5e1840e39f9b9a260ddae65b failed: Not found: failed to reset TS 
proxy: Could not find TS for UUID 293703ef5e1840e39f9b9a260ddae65b{noformat}
It's seems the issue is with wrong count of attempts, because for some reason 
the attempt counter stays 0 and the exponential back-off strategy is not 
applied.

The log from the test scenario run by the dist_test is attached.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

Reply via email to