[
https://issues.apache.org/jira/browse/CASSANDRA-10509?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14958464#comment-14958464
]
Stefania commented on CASSANDRA-10509:
--------------------------------------
It seems we get an extra row in the count if we cross the page boundary. For
example we can reproduce this problem about 50% of the times with as little as
1000 entries if we set the page size to 1000 using
{{self.session.default_fetch_size = 1000}}. Using the current dtest value of
100K entries is slower but also more reliable in reproducing this problem.
{{AbstractQueryPager.fetchPage}} retrieves an extra live row the last time it
is called. It has a mechanism to exclude the first row if it is the same as the
last row in the previous page by calling {{containsPreviousLast}}, which is
implemented by the sub-classes. {{RangeNamesQueryPager.containsPreviousLast}}
however always returns false because the corresponding {{queryNextPage}} uses
{{ExcludingBounds}} to set the range in the read command. Therefore, the last
queried key should never be included. However, as far as I can see
{{ExcludingBounds}} is serialized as {{Bounds}}, which does include the
endpoints. So, first we query (MIN, MIN) to get the entire range, then the next
page will query (LAST_KEY,MIN), where LAST_KEY is the key of the last partition
retrieved by the previous page, but if the last key is not local we are
actually querying \[LAST_KEY, MIN\] and we retrieve the partition for LAST_KEY
again.
It is not clear why it could not be reproduced with {{SELECT *}}.
Tentative [patch|https://github.com/stef1927/cassandra/tree/10509-2.2] attached.
http://cassci.datastax.com/view/Dev/view/stef1927/job/stef1927-10509-2.2-dtest
http://cassci.datastax.com/view/Dev/view/stef1927/job/stef1927-10509-2.2-testall
> Fix dtest cqlsh_tests.cqlsh_copy_tests.CqlshCopyTest.test_bulk_round_trip
> -------------------------------------------------------------------------
>
> Key: CASSANDRA-10509
> URL: https://issues.apache.org/jira/browse/CASSANDRA-10509
> Project: Cassandra
> Issue Type: Sub-task
> Reporter: Paulo Motta
> Assignee: Stefania
> Fix For: 2.2.x
>
>
> Test failing on 2.2 after fixing CASSANDRA-10507:
> http://cassci.datastax.com/view/Dev/view/stef1927/job/stef1927-10507-2.2-dtest/lastCompletedBuild/testReport/cqlsh_tests.cqlsh_copy_tests/CqlshCopyTest/test_bulk_round_trip/
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
