[
https://issues.apache.org/jira/browse/IMPALA-12155?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17745486#comment-17745486
]
ASF subversion and git services commented on IMPALA-12155:
----------------------------------------------------------
Commit ae95c43eda016e3380f395a235b373cec1337522 in impala's branch
refs/heads/master from wzhou-code
[ https://gitbox.apache.org/repos/asf?p=impala.git;h=ae95c43ed ]
IMPALA-12286: Make CatalogD HA robust
IMPALA-12155 added support for CatalogD HA. The statestore assigns the
roles for the catalogd in the HA pair. When an active catalogd is
elected, statestore sends RPCs to coordinators and catalogds to notify
the new active catalogd. But the RPCs could fail due to network
conditions. To avoid missing the notification of new active catalogd,
this patch makes statestore to resend RPCs to subscribers if there are
RPC failures.
This patch also rename the metric "catalog-server.ha-active-status" to
"catalog-server.active-status" since this metric is also set for
catalogd when CatalogD HA is not enabled.
As discussed in IMPALA-12267, catalogd need to re-generate its Catalog
Service ID when it becomes active.
Testing:
- Added unit-test cases for CatalogD HA with simulated RPC failures.
- Passed core tests.
Change-Id: Ibdfea022031c3cc1cbaf4ad52e947720a5d5630f
Reviewed-on: http://gerrit.cloudera.org:8080/20220
Reviewed-by: Andrew Sherman <[email protected]>
Reviewed-by: Abhishek Rawat <[email protected]>
Tested-by: Impala Public Jenkins <[email protected]>
Reviewed-by: Wenzhe Zhou <[email protected]>
> Support Impala CatalogD HA
> --------------------------
>
> Key: IMPALA-12155
> URL: https://issues.apache.org/jira/browse/IMPALA-12155
> Project: IMPALA
> Issue Type: Improvement
> Components: Backend
> Reporter: Wenzhe Zhou
> Assignee: Wenzhe Zhou
> Priority: Critical
> Fix For: Impala 4.3.0
>
>
> To support catalog HA, we need to have two CatalogD in HA pair, and add the
> preemptive behavior for Catalogd. When CatalogD HA is enabled, the preemptive
> behavior allows the CatalogD with the higher priority to become active and
> the paired CatalogD becomes standby. The active CatalogD acts as the source
> of metadata and provides catalog service for the Impala cluster. When active
> CatalogD is not healthy, the standby CatalogD will take over catalog service.
> Preemption is disabled on the CatalogD when it is not running as HA pair.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
