[
https://issues.apache.org/jira/browse/CASSANDRA-4332?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13294244#comment-13294244
]
Roland Gude commented on CASSANDRA-4332:
----------------------------------------
i think i understand the issue now:
It is dependant on Consistency Level/Replication and our usage of index queries
(which i guess several other people are using similar)
What we do is a index query and paging over the results (i.e. request 100 rows,
if we get 100 rows - fetch next 100 if we get less, stop)
Using consistency level one apparently there is a situation where the node
which started the query cannot satisfy the number of requested rows and simply
returns those he has - which is less then the expected number of rows. so
paging thinks there is nothing more and just stops.
Using consistency level ALL (or Quorum) the request seems to reach the node
which has the replica as well. That node now adds to the results the local
indexed rows starting from the given token. The results are merged and
returned. now - if first node cannot satisfy the number of requested rows - the
result will contain items from a node with another token range and paging can
continue on that other node.
If this is the case, this comes with several caveats.
In order to get consistent results from a paging index query, it is required to
either - page over all tokensranges from the cluster manually
or use consistency level ALL and make sure that replication layout gives you
adjacent token ranges (i.e. SimpleStrategy). Any other strategy might mess up
results, I guess.
does this sound at least half way reasonable?
> IndexRangeSliceQuery results contain IndexColumn even though it is not
> included in SliceRange
> ---------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-4332
> URL: https://issues.apache.org/jira/browse/CASSANDRA-4332
> Project: Cassandra
> Issue Type: Bug
> Affects Versions: 1.1.1
> Reporter: Roland Gude
> Attachments: system.log, system.log
>
>
> This is the magical reappearance of CASSANDRA-2964
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
