You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@cassandra.apache.org by "Ryan King (JIRA)" <ji...@apache.org> on 2010/05/12 02:19:41 UTC

[jira] Created: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Cache capacity settings done via nodetool get reset on memtable flushes
-----------------------------------------------------------------------

                 Key: CASSANDRA-1079
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
             Project: Cassandra
          Issue Type: Bug
    Affects Versions: 0.6.1
            Reporter: Ryan King


In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.

The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Jonathan Ellis (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12866411#action_12866411 ] 

Jonathan Ellis commented on CASSANDRA-1079:
-------------------------------------------

It's not surprising if you remember that the conf is The Source Of Truth. :)

I'm big on "there should only be one way to do it" but restarting to test cache size effects is too painful.

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>    Affects Versions: 0.6.1
>            Reporter: Ryan King
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Updated: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Jonathan Ellis (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jonathan Ellis updated CASSANDRA-1079:
--------------------------------------

    Attachment: 1079.txt

patch to not recalculate cache capacity if it's been manually modified

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 0.6
>            Reporter: Ryan King
>             Fix For: 0.6.2
>
>         Attachments: 1079.txt
>
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Stu Hood (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12866445#action_12866445 ] 

Stu Hood commented on CASSANDRA-1079:
-------------------------------------

I think the disconnect is that Ryan and Jeff would like to be able to tune cache sizes on a cluster that is receiving writes, which is difficult when it is reset per-memtable.

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>    Affects Versions: 0.6.1
>            Reporter: Ryan King
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Ryan King (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12868279#action_12868279 ] 

Ryan King commented on CASSANDRA-1079:
--------------------------------------

Patch looks good to me.

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 0.6
>            Reporter: Ryan King
>            Assignee: Jonathan Ellis
>             Fix For: 0.6.2
>
>         Attachments: 1079.txt
>
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Ryan King (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12866483#action_12866483 ] 

Ryan King commented on CASSANDRA-1079:
--------------------------------------

Yes, "not persisting  between restarts" is expected, it currently gets reset when at every flush.

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 0.6
>            Reporter: Ryan King
>             Fix For: 0.6.2
>
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Ryan King (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12866630#action_12866630 ] 

Ryan King commented on CASSANDRA-1079:
--------------------------------------

It appears the problem is that SSTableTracker's replace method calls updateCacheSizes, which reads the cache settings from DatabaseDescriptor (which doesn't get updated by the jmx command).

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 0.6
>            Reporter: Ryan King
>             Fix For: 0.6.2
>
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Ryan King (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12866409#action_12866409 ] 

Ryan King commented on CASSANDRA-1079:
--------------------------------------

If we're not going to let the jmx-based setting live longer than one memtable, we should probably remove the ability to set it that way. The current behavior is too surprising.

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>    Affects Versions: 0.6.1
>            Reporter: Ryan King
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Updated: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Jonathan Ellis (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jonathan Ellis updated CASSANDRA-1079:
--------------------------------------

        Fix Version/s: 0.6.2
    Affects Version/s: 0.6
                           (was: 0.6.1)
          Component/s: Core

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 0.6
>            Reporter: Ryan King
>             Fix For: 0.6.2
>
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Jeff Hodges (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12866433#action_12866433 ] 

Jeff Hodges commented on CASSANDRA-1079:
----------------------------------------

??I'm big on "there should only be one way to do it" but restarting to test cache size effects is too painful.??

I'm confused. Did you mean "isn't"?

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>    Affects Versions: 0.6.1
>            Reporter: Ryan King
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Resolved: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Jonathan Ellis (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jonathan Ellis resolved CASSANDRA-1079.
---------------------------------------

    Resolution: Won't Fix

this is working as designed.  the config file is The Source Of Truth for settings it contains.

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>    Affects Versions: 0.6.1
>            Reporter: Ryan King
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Reopened: (CASSANDRA-1079) Cache capacity settings done via nodetool get reset on memtable flushes

Posted by "Jonathan Ellis (JIRA)" <ji...@apache.org>.
     [ https://issues.apache.org/jira/browse/CASSANDRA-1079?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Jonathan Ellis reopened CASSANDRA-1079:
---------------------------------------


Oh.  Sometimes I need to read things twice -- yes, it's a bug if it's not persisting between flushes.  (I'm saying that it's not supposed to persist b/t restarts, unless you change the conf file too.)

> Cache capacity settings done via nodetool get reset on memtable flushes
> -----------------------------------------------------------------------
>
>                 Key: CASSANDRA-1079
>                 URL: https://issues.apache.org/jira/browse/CASSANDRA-1079
>             Project: Cassandra
>          Issue Type: Bug
>          Components: Core
>    Affects Versions: 0.6
>            Reporter: Ryan King
>             Fix For: 0.6.2
>
>
> In an experiment we set cache capacities via nodetool. The config file had the KeyCache for this CF at 1000000, we set the RowCache to 10000000 via nodetool.
> The next time we flushed a memtable for that CF, the cache capacity settings got reverted to what is in the conf file. We repeated the experiment with the same results.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.