You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@phoenix.apache.org by "Jesse Yates (JIRA)" <ji...@apache.org> on 2014/10/01 03:06:33 UTC
[jira] [Updated] (PHOENIX-1289) Drop index during upsert may abort
RS
[ https://issues.apache.org/jira/browse/PHOENIX-1289?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Jesse Yates updated PHOENIX-1289:
---------------------------------
Attachment: phoenix-1289-v1.patch
How about this version? I moved around the logic in the recover method a little bit to make it a little more readable and save on writes to the metadata table. Unfortunately, means a fair amount of line changes (though most of the original logic is preserved).
Only notable change from the previous is now we just keep a map of tableName -> min timestamp and then do the update all at once, rather than for each index table that failed.
> Drop index during upsert may abort RS
> -------------------------------------
>
> Key: PHOENIX-1289
> URL: https://issues.apache.org/jira/browse/PHOENIX-1289
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 3.0.0
> Reporter: daniel meng
> Attachments: DropIndexDuringUpsertIT.java, PHOENIX-1289.PATCH, phoenix-1289-v1.patch
>
>
> below execute path will abort RS:
> 1. client A write to table T with mutation m。 and T has an index of name IDX
> 2. m arrival at RS, but not start processing yet
> 3. client B drop index IDX
> 4. RS try to process m, and we get m' for IDX
> 5. RS try to write m' but fail as HBase Table IDX not exist
> 6. RS try to disable IDX but fail as Metadata has been deleted
> 7. KillServerOnFailurePolicy is triggered, server abort
> 8. recovery will fail with the same reason.
> an IT is attached
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)