Details
-
Improvement
-
Status: Resolved
-
Normal
-
Resolution: Fixed
-
None
-
None
Description
The goal of this ticket is to reduce the amount of tombstones created from ExpiringColumns.
Currently tombstones will always stay a full gc_grace time, which is not neccessary for ExpiringColumns. We only need to ensure that ExpiringColumn and tombstone together live as long as gc_grace. If the ExpiringColumn's TTL>=gc_grace then we can create an already gcable tombstone and drop that instantly.
My initial proposal was to use the ExpiringColumns creation-timestamp as deletiontime for the tombstone, but Sylvain pointed out that we should not mix local and client timestamps. So I changed it to this:
public static Column create(ByteBuffer name, ByteBuffer value, long timestamp, int timeToLive, int localExpirationTime, int expireBefore, IColumnSerializer.Flag flag) { if (localExpirationTime >= expireBefore || flag == IColumnSerializer.Flag.PRESERVE_SIZE) return new ExpiringColumn(name, value, timestamp, timeToLive, localExpirationTime); // the column is now expired, we can safely return a simple tombstone return new DeletedColumn(name, localExpirationTime-timeToLive, timestamp); // return new DeletedColumn(name, localExpirationTime, timestamp); // old code }
This was discussed on the mailinglist: http://cassandra-user-incubator-apache-org.3065146.n2.nabble.com/repair-compaction-and-tombstone-rows-td7583481.html
Attachments
Issue Links
- is related to
-
CASSANDRA-4905 Repair should exclude gcable tombstones from merkle-tree computation
- Resolved
-
CASSANDRA-2698 Instrument repair to be able to assess it's efficiency (precision)
- Resolved
-
CASSANDRA-5398 Remove localTimestamp from merkle-tree calculation (for tombstones)
- Resolved