[ 
https://issues.apache.org/jira/browse/CASSANDRA-6944?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14087409#comment-14087409
 ] 

graham sanderson commented on CASSANDRA-6944:
---------------------------------------------

Perhaps all these comments belong in CASSANDRA-6945 after all, but just 
finishing train of thought... things now seem a bit strange when threads are 
racing.

liveRatio and liveRatioComputedAt T can be passed from memtable X to a 
subsequent memtable Y before liveRatio for X at T is computed. liveRatio may 
then (I think) be computed for X or Y or Z (depending on where threads are at 
when MeteringRunnable instances for T executes) -  not sure which is 
preferable, but if it is X, then the liveRatio will not be applied to its 
successors (who may then not update their live ratio for some time)


> liveRatio jumps to max when Memtable is empty
> ---------------------------------------------
>
>                 Key: CASSANDRA-6944
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-6944
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>         Environment: SUSE Linux Enterprise 11 (64-bit)
>            Reporter: Erik Hansen
>            Assignee: Aleksey Yeschenko
>            Priority: Minor
>              Labels: memtables
>
> liveRatio calculation on an empty memtable results in a value of Infinity 
> since memtable.currentSize=0.  Infinity then gets capped at the liveRatio max 
> of 64.
> {noformat}
> WARN [MemoryMeter:1] 2014-03-19 09:26:59,483 Memtable.java (line 441) setting 
> live ratio to maximum of 64.0 instead of Infinity
> INFO [MemoryMeter:1] 2014-03-19 09:26:59,485 Memtable.java (line 452) 
> CFS(Keyspace='system', ColumnFamily='compactions_in_progress') liveRatio is 
> 64.0 (just-counted was 64.0).  calculation took 7ms for 0 cells
> {noformat}
> Jumping liveRatio to the max value based on an empty Memtable leads to more 
> frequent flushing than may be necessary.
> CASSANDRA-4243 previously addressed this issue, but was resolved as fixed by 
> CASSANDRA-3741.  It does not appear this issue has been fixed as of 2.0.5



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Reply via email to