[jira] [Created] (CASSANDRA-13024) Droppable Tombstone Ratio Calculation

["Stefan Podkowinski (JIRA)" <ji...@apache.org>]

Stefan Podkowinski created CASSANDRA-13024:
----------------------------------------------

             Summary: Droppable Tombstone Ratio Calculation
                 Key: CASSANDRA-13024
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-13024
             Project: Cassandra
          Issue Type: Bug
            Reporter: Stefan Podkowinski


Whenever no sstables can be compacted in a standard way, we currently try to compact sstables that would make worthwhile candidates by evaluation {{sstable.getEstimatedDroppableTombstoneRatio(gcBefore) <= tombstoneThreshold}}. You can find out more on how this is supposed to work and the various settings options by reading "[About Deletes and Tombstones in Cassandra|http://thelastpickle.com/blog/2016/07/27/about-deletes-and-tombstones.html#single-sstable-compaction]" by [~arodrime]. 

The bad news is that currently the ratio value will vary to a great degree depending on the data model and type of deletes, as the ratio will be created based on the number of tombstones and number of _columns_. Any kind of tombstone will be counted here, no matter if on column or partition level, which will give you very different ratios based on the number of columns in your table and how you delete the data. 

Considering a 0.20 default threshold for finding sstables with enough droppable tombstones, it only takes 3 columns in your table to never hit the threshold at all when using partition tombstones.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)