[
https://issues.apache.org/jira/browse/HBASE-18770?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16210595#comment-16210595
]
stack commented on HBASE-18770:
-------------------------------
Summarizing the above and what came of the discussion up on the dev list (See
link cited above in this JIRAs description):
* Only a select few preXXX ops can be by-passed; all others cannot opt to
by-pass the core call.
** Which few is TBD: candidates include flush, compaction, and RPC preXXX
calls for query and mutation, checkAndPutt, etc.
** We will make it so you CANNOT bypass loading of an hfile in
RegionObserver.preBulkLoadHFile (see comment above).
* The bypass will be flagged by a boolean returned out of the preXXX call;
true to by-pass. The current bypass flag in CP will be removed.
* The CP can say 'bypass' and provide its own answer (e.g. the Timeline Server
says 'bypass' and returns its own Get result out of the preGet to return to the
caller -- this would still work).
* Even if we bypass, we must update pertinent core metrics (in old days, on
by-pass, a CP might want to increment certain metrics anyways for whatever
reason); we'd do this so we do NOT have to expose internal metrics systems to
CPs (It has its own metrics to which it can gain access via CPEnv).
The thread up on dev has good examples of CPs using bypass; e.g. Tephra and
CDAP. Make sure these cases still work.
> Remove bypass method in ObserverContext and implement the 'bypass' logic case
> by case
> -------------------------------------------------------------------------------------
>
> Key: HBASE-18770
> URL: https://issues.apache.org/jira/browse/HBASE-18770
> Project: HBase
> Issue Type: Sub-task
> Components: Coprocessors
> Reporter: Duo Zhang
> Assignee: stack
> Fix For: 2.0.0-alpha-4
>
>
> http://search-hadoop.com/m/HBase/YGbbXd0RDCIHSC1
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
