[
https://issues.apache.org/jira/browse/CASSANDRA-3224?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jonathan Ellis updated CASSANDRA-3224:
--------------------------------------
Attachment: 3224.txt
This is a regression caused by CASSANDRA-3181.
The problem is that when CompactionTask calls submitBackground, the original
Task is not yet technically done, so LeveledCompactionStrategy returns "nothing
to do" to prevent multiple tasks running in parallel, which is not supported
for LCS. So after the first compaction runs five minutes in, that's all she
wrote.
Fixing the Task/Executor mess is out of scope for 1.0 (created CASSANDRA-3225
to address in 1.1) s
> LeveledCompactionStrategy is too complacent
> -------------------------------------------
>
> Key: CASSANDRA-3224
> URL: https://issues.apache.org/jira/browse/CASSANDRA-3224
> Project: Cassandra
> Issue Type: Bug
> Components: Core
> Affects Versions: 1.0.0
> Reporter: Brandon Williams
> Assignee: Benjamin Coverston
> Labels: compaction
> Fix For: 1.0.0
>
> Attachments: 3224.txt
>
>
> As the title says, it barely does anything. I inserted 50G worth of data
> with 1G heap and 99% overwrite ratio, and it only compacted twice:
> {noformat}
> INFO [CompactionExecutor:1] 2011-09-16 22:29:54,572 CompactionTask.java
> (line 118) Compacting
> [SSTableReader(path='/var/lib/cassandra/data/Keyspace1/Standard1-h-1-Data.db')]
> INFO [CompactionExecutor:1] 2011-09-16 22:29:58,606 CompactionTask.java
> (line 220) Compacted to
> [/var/lib/cassandra/data/Keyspace1/Standard1-h-2-Data.db,/var/lib/cassandra/data/Keyspace1/Standard1-h-4-Data.db,/var/lib/cassandra/data/Keyspace1/Standard1-h-5-Data.db,].
> 12,595,811 to 12,595,811 (~100% of original) bytes for 40,501 keys at
> 3.058122MBPS. Time: 3,928ms.
> INFO [CompactionExecutor:1] 2011-09-16 22:29:58,607 CompactionTask.java
> (line 222) CF Total Bytes Compacted: 12,595,811
> INFO [CompactionExecutor:3] 2011-09-16 22:29:58,889 CompactionTask.java
> (line 118) Compacting
> [SSTableReader(path='/var/lib/cassandra/data/Keyspace1/Standard1-h-4-Data.db'),
>
> SSTableReader(path='/var/lib/cassandra/data/Keyspace1/Standard1-h-2-Data.db'),
>
> SSTableReader(path='/var/lib/cassandra/data/Keyspace1/Standard1-h-5-Data.db'),
>
> SSTableReader(path='/var/lib/cassandra/data/Keyspace1/Standard1-h-3-Data.db')]
> INFO [CompactionExecutor:3] 2011-09-16 22:30:06,900 CompactionTask.java
> (line 220) Compacted to
> [/var/lib/cassandra/data/Keyspace1/Standard1-h-7-Data.db,/var/lib/cassandra/data/Keyspace1/Standard1-h-9-Data.db,/var/lib/cassandra/data/Keyspace1/Standard1-h-11-Data.db,/var/lib/cassandra/data/Keyspace1/Standard1-h-12-Data.db,/var/lib/cassandra/data/Keyspace1/Standard1-h-14-Data.db,/var/lib/cassandra/data/Keyspace1/Standard1-h-15-Data.db,].
> 28,374,396 to 28,374,396 (~100% of original) bytes for 91,236 keys at
> 3.380379MBPS. Time: 8,005ms.
> INFO [CompactionExecutor:3] 2011-09-16 22:30:06,901 CompactionTask.java
> (line 222) CF Total Bytes Compacted: 40,970,207
> {noformat}
> Resulting in the following levels:
> {noformat}
> L0: 4965
> L1: 6
> L2: 0
> L3: 0
> L4: 0
> L5: 0
> L6: 0
> L7: 0
> {noformat}
> This is obviously going to result in extremely poor read performance.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
