Jon Haddad created CASSANDRA-20107:
--------------------------------------
Summary: buffer sstable writes
Key: CASSANDRA-20107
URL: https://issues.apache.org/jira/browse/CASSANDRA-20107
Project: Cassandra
Issue Type: Improvement
Reporter: Jon Haddad
Watching the filesystem, we do this when writing an SSTable out:
{noformat}
root@cassandra2:~# xfsslower -p $(cassandra-pid) 0
Tracing XFS operations
TIME COMM PID T BYTES OFF_KB LAT(ms) FILENAME
20:13:26 CompactionExec 33477 W 16359 5607255 0.04 nb-1499-big-Data.db
20:13:26 CompactionExec 33477 W 1 5607271 0.00 nb-1499-big-Data.db
20:13:26 CompactionExec 33477 W 1 5607271 0.00 nb-1499-big-Data.db
20:13:26 CompactionExec 33477 W 1 5607271 0.00 nb-1499-big-Data.db
20:13:26 CompactionExec 33477 W 1 5607271 0.00 nb-1499-big-Data.db
{noformat}
This is silly. We should be internally buffering our writes. It's probably
not a ton of overhead, but if we're going to move to writing sstables with
direct IO, we have to do it anyways.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
