Brandon Williams created CASSANDRA-7322:
-------------------------------------------
Summary: CqlPagingRecordReader returns partial or no data
Key: CASSANDRA-7322
URL: https://issues.apache.org/jira/browse/CASSANDRA-7322
Project: Cassandra
Issue Type: Bug
Components: Hadoop
Reporter: Brandon Williams
Assignee: Tyler Hobbs
Fix For: 2.1 rc1
In trying to solve the last problem with CASSANDRA-7241, this bisected to
CASSANDRA-6379. On the surface of course this doesn't seem right, so I
repeated the bisect (both of which were completely manual and tested by hand)
and landed at the same place, and since it did touch getSplits I think it's the
right candidate.
What I see is that CPRR gets all the splits, which with vnodes are mostly empty
since there's very little data in the test. Before CASSANDRA-6379 it iterates
them all and succeeds by eventually finding it. When it fails, however, it
only checks the first split and gives up, even if it does happen to get lucky
and find one row.
If I use data generated before CASSANDRA-6379, it works fine, but not with data
generated afterward. I suspect this isn't specific to CPRR but hadoop in
general, it's just been easier for me to spot with CPRR, perhaps due to the
nature of the test.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
