[
https://issues.apache.org/jira/browse/PHOENIX-3326?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15519226#comment-15519226
]
James Taylor commented on PHOENIX-3326:
---------------------------------------
The problem with putting the lock on SYSTEM.CATALOG cell is that then our mutex
is in the snapshot, so if it's restored it breaks things. Two clients upgrading
at the same time is a recipe for disaster.
I think the first thing we should do in the upgrade code is create a new HBase
table (and ignore exception if it already exists) for the mutex called
SYSTEM.MUTEX (not a Phoenix table, so we have no dependencies on
SYSTEM.CATALOG). Just use the right logic to get the name correct wrt system
namespaces being on (we've got a util for that, I believe).
> Restoring SYSTEM.CATALOG from snapshot causes clients to run into
> UpgradeInProgressException
> --------------------------------------------------------------------------------------------
>
> Key: PHOENIX-3326
> URL: https://issues.apache.org/jira/browse/PHOENIX-3326
> Project: Phoenix
> Issue Type: Bug
> Reporter: Samarth Jain
> Assignee: Samarth Jain
> Attachments: PHOENIX-3326_4.8-HBase-0.98.patch,
> PHOENIX-3326_4.8-HBase-0.98_v2.patch, PHOENIX-3326_wip.patch
>
>
> We create a snapshot of the SYSTEM.CATALOG table only after the client is
> able to successfully acquire a distributed mutex of sorts. This means the
> snapshot also ends up containing the row that serves as the mutex. Now when
> restoring the table from snapshot, this rows is still present which causes
> clients to throw UpgradeInProgress exception.
> I can think of a couple of ways to fix this:
> 1) Do the checkAndPut for the UPGRADE_MUTEX after creating the snapshot. I am
> not too sure though how about HBase handles concurrent snapshot requests. Do
> clients get an exception? Also we potentially could end up creating more
> snapshots than we really need to.
> 2) Do the checkAndPut for the UPGRADE_MUTEX in a different table (possibly
> SYSTEM.SEQUENCE). This way the restored snapshot won't have the row. We would
> need to delete the row from SYSTEM.SEQUENCE after the upgrade is done
> (successfully or unsuccessfully).
> [~jamestaylor] - WDYT?
> FYI, [~lhofhansl] - this is probably a blocker for 4.8.1
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
