Peter Vary commented on HIVE-23764:

[~rajesh.balamohan]: I see that in HIVE-23597 we have issues with some tests.
Also caching the OrcTail might be better placed in LLAP IO, and [~szita] is 
working on a possible solution.
What do you think about pushing this change, and if we hit some road-block with 
the LLAP IO solution then we might pick up HIVE-23597 again?


> Remove unnecessary getLastFlushLength when checking delete delta files
> ----------------------------------------------------------------------
>                 Key: HIVE-23764
>                 URL: https://issues.apache.org/jira/browse/HIVE-23764
>             Project: Hive
>          Issue Type: Improvement
>          Components: Transactions
>            Reporter: Peter Vary
>            Assignee: Peter Vary
>            Priority: Major
>              Labels: pull-request-available
>          Time Spent: 10m
>  Remaining Estimate: 0h
> VectorizedOrcAcidRowBatchReader$ColumnizedDeleteEventRegistry calls 
> OrcAcidUtils.getLastFlushLength for every delete delta file.
> Even the comment says:
> {code}
>               // NOTE: Calling last flush length below is more for 
> future-proofing when we have
>               // streaming deletes. But currently we don't support streaming 
> deletes, and this can
>               // be removed if this becomes a performance issue.
> {code}
> If we have a table with 5 updates (1 base + 5 delta + 5 delete_delta), then 
> for every base + delta dir we will check all of the delete_delta directories, 
> and check the getLastFlushLength method which will result in 6*5=30 
> unnecessary NN/S3 calls.
> We should remove the check as already proposed in the comment.

This message was sent by Atlassian Jira

Reply via email to