[
https://issues.apache.org/jira/browse/CASSANDRA-7394?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Aleksey Yeschenko updated CASSANDRA-7394:
-----------------------------------------
Description:
As of now, collectTimeOrderedData() will try to read from at least one sstable,
even if the memtable has a row tombstone with a higher timestamp than any max
timestamp in the sstable. If we initiate mostRecentRowTombstone with the value
from the memtables and not Long.MIN_VALUE, reading from the sstables can be
avoided entirely in this scenario.
Current tracking is also broken b/c we update mostRecentRowTombstone value with
the one read from the last sstable, unconditionally, even if it's a lesser
tombstone than we used to have tracked.
was:As of now, collectTimeOrderedData() will try to read from at least one
sstable, even if the memtable has a row tombstone with a higher timestamp than
any max timestamp in the sstable. If we initiate mostRecentRowTombstone with
the value from the memtables and not Long.MIN_VALUE, reading from the sstables
can be avoided entirely in this scenario.
Priority: Minor (was: Trivial)
Issue Type: Bug (was: Improvement)
Summary: Fix CollationController#collectTimeOrderedData()
mostRecentRowTombstone tracking (was: Optimize
CollationController#collectTimeOrderedData() to potentially skip the sstables
altogether)
> Fix CollationController#collectTimeOrderedData() mostRecentRowTombstone
> tracking
> --------------------------------------------------------------------------------
>
> Key: CASSANDRA-7394
> URL: https://issues.apache.org/jira/browse/CASSANDRA-7394
> Project: Cassandra
> Issue Type: Bug
> Reporter: Aleksey Yeschenko
> Assignee: Aleksey Yeschenko
> Priority: Minor
>
> As of now, collectTimeOrderedData() will try to read from at least one
> sstable, even if the memtable has a row tombstone with a higher timestamp
> than any max timestamp in the sstable. If we initiate mostRecentRowTombstone
> with the value from the memtables and not Long.MIN_VALUE, reading from the
> sstables can be avoided entirely in this scenario.
> Current tracking is also broken b/c we update mostRecentRowTombstone value
> with the one read from the last sstable, unconditionally, even if it's a
> lesser tombstone than we used to have tracked.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
