[GitHub] [incubator-doris] wangbo opened a new issue #2822: RollupJobV2 Bug Report When Replay EditLog

[GitBox <gi...@apache.org>]

wangbo opened a new issue #2822: RollupJobV2 Bug Report When Replay EditLog
URL: https://github.com/apache/incubator-doris/issues/2822
 
 
   #### Bug 1 
   ##### Description
    Job State transfer to error state when call 'RollupJobV2.replayPending' method
   In RollupJobV2.replayPending,job's state is transfered from PENDING pending to WAITING_TXN;
   If Fe stops when job in pending state but didn't send create tablet request to be,when fe start and replay pending job,the job's state is WAITING_TXN,and then the job running in WAITING_TXN,this will cause the rollup job failed,because be doesn't have new tablet info.
   
   ##### To Reproduce
   restart fe when job in pending  state.
   
   ##### Solution
   when replay pending job,set job'state to pending.
   Send one create tablet request to be twice is ok.
   
   #### Bug 2
   ##### Description
   When the job create tablet success and transfer to waitingTXN state, restart the server to trigger checkpoint ,the pending job is in checkpoint and will never be replayed;
   When the job is transfered to running,restart the server again,the rollup job will be failed;
   
   ##### analysis
   1. when job is finish pending work and transfered to waitingTXN,the index meta is written to olaptable in memory.
   2. restart fe
   2.1 by replayPending, index meta is written to `TabletInvertedIndex`
   2.2 index meta in olaptable is lost,so the olaptable info checkpoint image lost index meta.
   2.3 checkpoint happens,replayPending won't being called forever.
   3. when job running in running restart fe
   3.1 load image
   3.1.1 loadDb,actually load olaptable,but the olaptable doesn't have index meta
   3.1.2  recreateTabletInvertIndex,  index meta can't be written to `TabletInvertedIndex`,because of `3.1.1`
   3.1.3 loadAlterJob,index meta is written to olaptable in memory
   4. Finally, `TabletInvertedIndex` lost index meta
   5. When BE report tablets to FE,FE can't find tablet in `TabletInvertedIndex`,so Fe send Drop Tablet Request
   6. Job Failed
   
   3.1 by replayWaitingTxn,index meta is written to olaptable in memory again.
   3.2 index meta can't be written to `TabletInvertedIndex`
   
   ##### To Reproduce
   Fix bug 1,then do the same operator;
   
   ##### Solution
   when replayPending,set index meta to olaptable if possible
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@doris.apache.org
For additional commands, e-mail: commits-help@doris.apache.org

[GitHub] [incubator-doris] wangbo closed issue #2822: RollupJobV2 Bug Report When Replay EditLog

[GitBox <gi...@apache.org>]

wangbo closed issue #2822: RollupJobV2 Bug Report When Replay EditLog
URL: https://github.com/apache/incubator-doris/issues/2822
 
 
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services

---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscribe@doris.apache.org
For additional commands, e-mail: commits-help@doris.apache.org