[ https://issues.apache.org/jira/browse/CASSANDRA-6826?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13932398#comment-13932398 ]
Bill Mitchell commented on CASSANDRA-6826: ------------------------------------------ Much as I found Sylvain's suggestion plausible, no, it does not explain this problem. After installing the Apache Cassandra 2.0.6 build, the first time I tried this, it still failed. Unfortunately, the problem is data or timing dependent. After seeing the failure on 2.0.6, I changed the test case to write all the rows into one partition, and that worked, so I changed it back to distributing the rows over 6 partitions, and this time that worked, too. So we were "lucky" that the first time I tried this, the failure did appear. (I should have noticed that CASSANDRA-6748 appeared only when a column was explicitly set to null. That was the behavior of my code about two weeks ago, before I discovered the issues around having a large number of tombstones in a wide row.) > Query returns different number of results depending on fetchsize > ---------------------------------------------------------------- > > Key: CASSANDRA-6826 > URL: https://issues.apache.org/jira/browse/CASSANDRA-6826 > Project: Cassandra > Issue Type: Bug > Components: Core > Environment: quad-core Windows 7 x64, single node cluster > Cassandra 2.0.5 > Reporter: Bill Mitchell > Assignee: Sylvain Lebresne > > I issue a query across the set of partitioned wide rows for one logical row, > where s, l, and partition specify the composite primary key for the row: > SELECT ec, ea, rd FROM sr WHERE s = ? and partition IN ? and l = ? ALLOW > FILTERING; > If I set fetchSize to only 1000 when the Cluster is configured, the query > sometimes does not return all the results. In the particular case I am > chasing, it returns a total of 98586 rows. If I increase the fetchsize to > 100000, all the 99999 actual rows are returned. This suggests there is some > problem with fetchsize re-establishing the position on the next segment of > the result set, at least when multiple partitions are being accessed. -- This message was sent by Atlassian JIRA (v6.2#6252)