You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues-all@impala.apache.org by "ASF subversion and git services (Jira)" <ji...@apache.org> on 2021/06/03 07:43:01 UTC

[jira] [Commented] (IMPALA-10197) Add KUDU_REPLICA_SELECTION config and query option

    [ https://issues.apache.org/jira/browse/IMPALA-10197?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17356260#comment-17356260 ] 

ASF subversion and git services commented on IMPALA-10197:
----------------------------------------------------------

Commit 4c07aff664beddf0d0f8e93a45e936176700a31b in impala's branch refs/heads/dependabot/pip/infra/python/deps/py-1.10.0 from wzhou-code
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=4c07aff ]

IMPALA-10197 (Part 2): Add KUDU_REPLICA_SELECTION query option

The previous patch added a new test case in PlannerTest for Kudu,
which check for specific number of hosts and instances for distributed
plan with query option KUDU_REPLICA_SELECTION set as LEADER_ONLY.
However, the leadership isn't deterministic since each Kudu partition
has its own Raft group and the leaders of these groups are not exactly
balanced across hosts in a cluster. There's no guarantee that we'll get
a certain number of leader hosts for a query which access multiple KUDU
partitions. This makes the unit-test flaky.

This patch removed the distributed plans to avoid nondeterminism.

Testing:
 - Reran the Planner test and verified the issue did not happen.

Change-Id: I3e23667c06c273a261e03de3d81fc7ee1f6b0682
Reviewed-on: http://gerrit.cloudera.org:8080/17502
Reviewed-by: Quanlong Huang <hu...@gmail.com>
Tested-by: Impala Public Jenkins <im...@cloudera.com>


> Add KUDU_REPLICA_SELECTION config and query option
> --------------------------------------------------
>
>                 Key: IMPALA-10197
>                 URL: https://issues.apache.org/jira/browse/IMPALA-10197
>             Project: IMPALA
>          Issue Type: Improvement
>          Components: Frontend
>    Affects Versions: Impala 4.0
>            Reporter: Grant Henke
>            Assignee: Wenzhe Zhou
>            Priority: Trivial
>              Labels: kudu, newbie, ramp-up
>             Fix For: Impala 4.1
>
>
> Sometimes it is useful to target queries at the LEADER_ONLY instead of the default CLOSEST_REPLICA. We should expose this option in Impala similar to the way KUDU_READ_MODE is exposed:
> https://impala.apache.org/docs/build/html/topics/impala_kudu_read_mode.html
> Currently there is a test only flag, FLAGS_pick_only_leaders_for_tests, that somewhat does this:
> https://github.com/apache/impala/blob/da2999afd9ddc45d35141649d17db507e03ee9bf/be/src/exec/kudu-scanner.cc#L195-L198



--
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