[jira] [Commented] (IMPALA-9197) Hash table lookup should be read-only

["ASF subversion and git services (Jira)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/IMPALA-9197?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17018359#comment-17018359 ] 

ASF subversion and git services commented on IMPALA-9197:
---------------------------------------------------------

Commit df13de70fe5a286d7d3596bf2f1ecd71945e8f27 in impala's branch refs/heads/master from Tim Armstrong
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=df13de7 ]

IMPALA-9197: make HashTable lookups thread-safe

This makes it possible to do HashTable lookups from
multiple threads without any data races. This
requires moving statistics that are updated during
probing to the HashTableCtx object.

There are some small changes to the hash table stat
logging behaviour as a result of the stats being moved
to the context. I don't believe these logs are used much,
if at all.

Testing:
Ran exhaustive tests.

Manually inspected some aggregation and join query profiles
to check that hash table stats looked reasonable.

Perf:
Ran TPC-H scale factor 30 on a single node. No significant
change in perf.

Change-Id: I92fbfa8cc000477b8e01975a102d818f9fa27c61
Reviewed-on: http://gerrit.cloudera.org:8080/14917
Reviewed-by: Impala Public Jenkins <im...@cloudera.com>
Tested-by: Impala Public Jenkins <im...@cloudera.com>


> Hash table lookup should be read-only
> -------------------------------------
>
>                 Key: IMPALA-9197
>                 URL: https://issues.apache.org/jira/browse/IMPALA-9197
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Backend
>            Reporter: Tim Armstrong
>            Assignee: Tim Armstrong
>            Priority: Major
>              Labels: multithreading
>
> For IMPALA-9156, we need concurrent lookups to the hash table to be thread safe. We are pretty close to that, except a few stats are maintained in HashTable and would be modified concurrently from multiple threads.
> We should modify those  places to update the stats in HashTableCtx instead.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org