[
https://issues.apache.org/jira/browse/CASSANDRA-6924?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13948631#comment-13948631
]
Tyler Hobbs commented on CASSANDRA-6924:
----------------------------------------
After a few unsuccessful bisect runs, it looks like this problem is caused by
1) dropping and recreating a keyspace too quickly, or 2) querying a secondary
index too quickly after creation. I'm not sure why these are more problematic
in 1.2.16-tentative than in 1.2.15. (Bisect roughly pointed at CASSANDRA-6700,
which doesn't make a lot of sense.)
With unique keyspace names each run, a 1s sleep after creating the index and
inserting the data was enough to guarantee correct results. When re-using the
same keyspace name, even a 30s sleep wasn't always enough (as I noted before).
That's annoying, but since it goes against recommended practices (and we have a
proper fix in 2.1?), I'm okay with marking this as invalid.
> Data Inserted Immediately After Secondary Index Creation is not Indexed
> -----------------------------------------------------------------------
>
> Key: CASSANDRA-6924
> URL: https://issues.apache.org/jira/browse/CASSANDRA-6924
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: Tyler Hobbs
> Fix For: 1.2.16
>
>
> The head of the cassandra-1.2 branch (currently 1.2.16-tentative) contains a
> regression from 1.2.15. Data that is inserted immediately after secondary
> index creation may never get indexed.
> You can reproduce the issue with a [pycassa integration
> test|https://github.com/pycassa/pycassa/blob/master/tests/test_autopacking.py#L793]
> by running:
> {noformat}
> nosetests tests/test_autopacking.py:TestKeyValidators.test_get_indexed_slices
> {noformat}
> from the pycassa directory.
> The operation order goes like this:
> # create CF
> # create secondary index
> # insert data
> # query secondary index
> If a short sleep is added in between steps 2 and 3, the data gets indexed and
> the query is successful.
> If a sleep is only added in between steps 3 and 4, some of the data is never
> indexed and the query will return incomplete results. This appears to be the
> case even if the sleep is relatively long (30s), which makes me think the
> data may never get indexed.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
