[
https://issues.apache.org/jira/browse/KUDU-3248?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17318043#comment-17318043
]
ASF subversion and git services commented on KUDU-3248:
-------------------------------------------------------
Commit ccbbfb3006314f2c37f3a40bfec355db9fc90e02 in kudu's branch
refs/heads/master from Grant Henke
[ https://gitbox.apache.org/repos/asf?p=kudu.git;h=ccbbfb3 ]
KUDU-3248: Match C++ replica selection behavior of Java client
The C++ client currently uses a new random value every time
a replica is selected. Alternatively, the Java client uses a static
value that ensures the selection for a single process or application
remains deterministic. This patch adjusts the C++ implementation
to match the Java client behavior.
This is expected to be a good balance of efficient use of cache
memory and distribution of load across the replicas given a
single process will always get the same choice resulting in
cache hits for follow up scans, but separate processes
will get a potentially different random selection.
The reviews on the Java patch here have some good context
and discussion: https://gerrit.cloudera.org/#/c/12158/
Change-Id: Iaa55e88b4a222fabfaa7fa521c24482cc6816b04
Reviewed-on: http://gerrit.cloudera.org:8080/17129
Tested-by: Kudu Jenkins
Reviewed-by: Attila Bukor <[email protected]>
> Expose replica selection option that provides better affinity for remote scans
> ------------------------------------------------------------------------------
>
> Key: KUDU-3248
> URL: https://issues.apache.org/jira/browse/KUDU-3248
> Project: Kudu
> Issue Type: Improvement
> Components: client
> Reporter: David Rorke
> Assignee: Grant Henke
> Priority: Major
>
> Kudu clients like Impala that use the CLOSEST_REPLICA selection option can
> suffer from bad tserver buffer cache warmup/locality during remote scans due
> to lack of tserver affinity as described in IMPALA-10481.
> The Kudu client should expose a replica selection option that provides better
> affinity when doing remote scans. The comments in IMPALA-10481 suggest a
> new LOCAL_OR_LEADER option that would select a local replica if one exists,
> otherwise fallback to the leader.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
