[
https://issues.apache.org/jira/browse/PHOENIX-5193?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16807075#comment-16807075
]
Vincent Poon commented on PHOENIX-5193:
---------------------------------------
[~mihir6692] As discussed, we should write a patch that combines your patches
in PHOENIX-5193 and PHOENIX-5194
Just spoke with [~abhishek.chouhan], who pointed out that we should put the fix
in MutationState's mutationCommand#doMutation(), since if we put it in
PhoenixIndexFailurePolicy#doBatchWithRetries, that would affect server-side
DELETE and UPSERT_SELECT as well, and in those cases we don't need this logic
because in theory we shouldn't be missing the index metadata (since it's sent
as part of the scan). And even if we did need to set the metadata cache, we
wouldn't want it to get set by every server, but instead have the client do it.
> Handle SQLException:ERROR 2008 (INT10) in
> PhoenixIndexFailurePolicy#doBatchWithRetries
> ---------------------------------------------------------------------------------------
>
> Key: PHOENIX-5193
> URL: https://issues.apache.org/jira/browse/PHOENIX-5193
> Project: Phoenix
> Issue Type: Sub-task
> Affects Versions: 4.14.0, 5.0.0, 4.15.0, 4.14.1
> Reporter: Monani Mihir
> Assignee: Monani Mihir
> Priority: Critical
> Attachments: PHOENIX-5193-4.x-HBase-1.3.01.patch
>
>
> In MutationState#send(Iterator<TableRef>)#doMutation , we use htable.batch()
> which will throw SQLException wrapped in
> RetriesExhaustedWithDetailsException.
> These wrapped SQLException are of two types,
> 1) INDEX_WRITE_FAILURE(1121, "XCL21", "Write to the index failed.")
> 2) INDEX_METADATA_NOT_FOUND(2008, "INT10", "Unable to find cached index
> metadata. ")
> Out of this we only handle 1st (INDEX_WRITE_FAILURE) and we should handle 2nd
> one too (INDEX_METADATA_NOT_FOUND)
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
