[
https://issues.apache.org/jira/browse/CASSANDRA-2401?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jonathan Ellis updated CASSANDRA-2401:
--------------------------------------
Attachment: 2401.txt
I found *a* bug that could cause this: Cassandra will re-create a deleted index
entry if it gets a write with an obsolete timestamp, but the data row tombstone
will correctly suppress an update there.
I don't think this is the bug Tey Kar is hitting, though, because unless I'm
mistaken you won't get this NPE until after the data row tombstone is removed
by compaction after gc_grace_seconds. 4 days isn't enough to see that unless
you've tweaked gc_g_s.
Still, it's worth fixing. Patch attached. (Also adds an assert w/ more
information if/when another way of triggering this is found.)
> getColumnFamily() return null, which is not checked in ColumnFamilyStore.java
> scan() method, causing Timeout Exception in query
> -------------------------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-2401
> URL: https://issues.apache.org/jira/browse/CASSANDRA-2401
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Affects Versions: 0.7.0
> Environment: Hector 0.7.0-28, Cassandra 0.7.4, Windows 7, Eclipse
> Reporter: Tey Kar Shiang
> Assignee: Jonathan Ellis
> Fix For: 0.7.6
>
> Attachments: 2401.txt
>
>
> ColumnFamilyStore.java, line near 1680, "ColumnFamily data =
> getColumnFamily(new QueryFilter(dk, path, firstFilter))", the data is
> returned null, causing NULL exception in "satisfies(data, clause, primary)"
> which is not captured. The callback got timeout and return a Timeout
> exception to Hector.
> The data is empty, as I traced, I have the the columns Count as 0 in
> removeDeletedCF(), which return the null there. (I am new and trying to
> understand the logics around still). Instead of crash to NULL, could we
> bypass the data?
> About my test:
> A stress-test program to add, modify and delete data to keyspace. I have 30
> threads simulate concurrent users to perform the actions above, and do a
> query to all rows periodically. I have Column Family with rows (as File) and
> columns as index (e.g. userID, fileType).
> No issue on the first day of test, and stopped for 3 days. I restart the test
> on 4th day, 1 of the users failed to query the files (timeout exception
> received). Most of the users are still okay with the query.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
