[
https://issues.apache.org/jira/browse/CASSANDRA-9079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14389499#comment-14389499
]
Vinay Chella commented on CASSANDRA-9079:
-----------------------------------------
This discussion leads to another interesting point, maybe setting these
properties at column family level would give much more granular control, since
cluster could have different column families with different behavior for TTL
and tombstone data in terms of the payload and gc_grace_seconds. Disabling
“tombstoneThresholds” at cluster level might lead to CASSANDRA-6117 if there is
any column family in the cluster, which is having moderate/high payload and
large number of tombstones.
> "respectTombstoneThresholds" should be configurable
> ----------------------------------------------------
>
> Key: CASSANDRA-9079
> URL: https://issues.apache.org/jira/browse/CASSANDRA-9079
> Project: Cassandra
> Issue Type: Improvement
> Components: Core
> Reporter: Vinay Chella
> Labels: tombstone
> Attachments: 9079.txt
>
>
> Currently there is no way to control/ configure the functionality of "drop
> queries exceeding a configurable number of tombstones". Making this
> functionality configurable through yaml helps use case like:
> We have several production clusters with large number of tombstones due to
> aggressive TTLs with very small payloads, and low gc_grace_seconds. Most of
> the read queries on these column families are ending up in drop operations or
> excessive warn messages due to "tombstone_failure_threshold" and
> "tombstone_warn_threshold". One cannot deterministically disable these
> tombstone thresholds by setting high values. Hence, making
> "SliceQueryFilter.respectTombstoneThresholds()" configurable through yaml
> would be the better approach.
> Column families with small payloads (couple hundred bytes) having large
> number of tombstones would not lead to "Avoid death-by-tombstone by default
> (CASSANDRA-6117)", which is the reason for observed drop operations. We see
> drop operations before compactions catch up, and clean those tombstones with
> both STCS and LCS. Configurable "respectTombstoneThresholds()" will help in
> reducing tombstone warnings noise.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
