[
https://issues.apache.org/jira/browse/CASSANDRA-12709?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15523685#comment-15523685
]
Nate McCall commented on CASSANDRA-12709:
-----------------------------------------
Specifically:
{noformat}
// An expired tombstone will be immediately discarded in memory,
and needn't be counted.
// Neither should be any cell shadowed by a range- or a partition
tombstone.
if (cell.getLocalDeletionTime() < gcBefore ||
!columnCounter.count(cell, tester))
continue;
{noformat}
Yeah, that's not cool from an operator's perspective (and actually answers a
question or two I've had recently with tombstone issues on 2.1 and 2.2
clusters).
This behavior works as one would expect again in 3.x in that if it's a
tombstone, we count it:
https://github.com/apache/cassandra/blob/trunk/src/java/org/apache/cassandra/db/ReadCommand.java#L483-L484
A patch will have to include a {{NEWS.txt}} entry as we will be modifying
production behavior and will for sure catch some people unaware.
> Nothing logged when lots of expired tombstones in a slice query
> ---------------------------------------------------------------
>
> Key: CASSANDRA-12709
> URL: https://issues.apache.org/jira/browse/CASSANDRA-12709
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Reporter: nicolas ginder
> Priority: Minor
>
> Currently Cassandra only logs a warning and an error for non expired
> tombstones, see org.apache.cassandra.db.filter.SliceQueryFilter. Expired
> tombstones used to be included in the tombstone count but this has been
> removed by CASSANDRA-9299.
> It is unclear that expired tombstones are not counted anymore, making
> tombstone_warn_threshold and tombstone_failure_threshold property names
> misleading.
> Currently the only way to get some visibility is by running
> SSTableMetadataViewer to get the droppable tombstone ratio.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
