[
https://issues.apache.org/jira/browse/PHOENIX-881?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13942565#comment-13942565
]
James Taylor commented on PHOENIX-881:
--------------------------------------
There are some bugs here in MetaDataEndPointImpl, but the fix you're proposing
isnt' quite right.
The scan done by buildDeletedTables is only looking for a *newer* table than
clientTimeStamp. The scan after that in doDropTable is looking for a table up
to the clientTimestamp. This covers the corner case of going "back in time" and
deleting a table older than clientTimestamp, but not the most recent table (as
we already looked for that in the first call to buildTable and didn't find it).
Look closely at the scan time range used in both scans.
But you are right, we should not be assuming that table is not null, as there's
the potential for an NPE here. That code crept in later.
How about if I take a crack at fixing it and writing a unit test for it as well?
+1 on everything else, but please remove the clearCache call as it'll be called
already (unless this is still causing problems).
For the zookeeper quorum, HBase has a single clientPort property and another
hbase.zookeeper.quorum property. It's been our experience that the port
information is not expected to be in the hbase.zookeeper.quorum property, but
instead the clientPort property applies to all the hosts listed in the
hbase.zookeeper.quorum property. I don't know what happens if you include the
port info in the hbase.zookeeper.quorum property and have a clientPort as well.
Have you actually tried this and confirmed it works?
> end2end test fails to run against a live cluster using end2endtest.py
> ---------------------------------------------------------------------
>
> Key: PHOENIX-881
> URL: https://issues.apache.org/jira/browse/PHOENIX-881
> Project: Phoenix
> Issue Type: Bug
> Affects Versions: 3.0.0, 4.0.0
> Reporter: Jeffrey Zhong
> Assignee: Jeffrey Zhong
> Attachments: phoenix-881.patch
>
>
> In most IT tests, we're using constant PHOENIX_JDBC_URL instead of
> BaseConnectedQueryIT#getUrl() which will return correct url depending where
> the test is running against.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
