You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@accumulo.apache.org by Roshan Punnoose <ro...@gmail.com> on 2016/06/08 13:28:59 UTC
Accumulo Ingest JMX metric
I am using Accumulo 1.6.5 and trying to monitor ingest rates through JMX
endpoints. The ingest reported through JMX does not seem to match with the
ingest rates reported on the accumulo monitor. Looking at the source code,
I noticed that the TabletServerMBean is counting all the entries in memory
for online tablets on that TabletServer:
result += tablet.getNumEntriesInMemory()
Is this is a good metric to use for ingest rates? Just want a good way to
measure ingest rates as I change configurations and run the continuous
ingest.
Thanks!
Roshan
Re: Accumulo Ingest JMX metric
Posted by Dylan Hutchison <dh...@cs.washington.edu>.
Hi Roshan,
We fixed the JMX stat call for the ingest rate. It will go into 1.6.6
when it is released.
Cheers, Dylan
On Wed, Jun 8, 2016 at 8:26 AM, Josh Elser <jo...@gmail.com> wrote:
> Boss. Thanks, Roshan.
>
> Roshan Punnoose wrote:
>
>> Just filed: https://issues.apache.org/jira/browse/ACCUMULO-4334
>>
>> Thanks Josh!
>>
>> On Wed, Jun 8, 2016 at 11:13 AM Josh Elser <josh.elser@gmail.com
>> <ma...@gmail.com>> wrote:
>>
>> Yeah, this looks wrong.
>>
>> The numbers the monitor gets are from the TabletServer polling each
>> Tablet
>>
>>
>> https://github.com/apache/accumulo/blob/1.6/server/tserver/src/main/java/org/apache/accumulo/tserver/TabletServer.java#L3676-L3677
>>
>> The JMX metrics reported by the TServer should also likely do the
>> same.
>>
>> Care to file a JIRA issue?
>>
>> Roshan Punnoose wrote:
>> > I am using Accumulo 1.6.5 and trying to monitor ingest rates
>> through JMX
>> > endpoints. The ingest reported through JMX does not seem to match
>> with
>> > the ingest rates reported on the accumulo monitor. Looking at the
>> source
>> > code, I noticed that the TabletServerMBean is counting all the
>> entries
>> > in memory for online tablets on that TabletServer:
>> >
>> > result+=tablet.getNumEntriesInMemory()
>> >
>> >
>> > Is this is a good metric to use for ingest rates? Just want a
>> good way
>> > to measure ingest rates as I change configurations and run the
>> > continuous ingest.
>> >
>> > Thanks!
>> > Roshan
>>
>>
Re: Accumulo Ingest JMX metric
Posted by Josh Elser <jo...@gmail.com>.
Boss. Thanks, Roshan.
Roshan Punnoose wrote:
> Just filed: https://issues.apache.org/jira/browse/ACCUMULO-4334
>
> Thanks Josh!
>
> On Wed, Jun 8, 2016 at 11:13 AM Josh Elser <josh.elser@gmail.com
> <ma...@gmail.com>> wrote:
>
> Yeah, this looks wrong.
>
> The numbers the monitor gets are from the TabletServer polling each
> Tablet
>
> https://github.com/apache/accumulo/blob/1.6/server/tserver/src/main/java/org/apache/accumulo/tserver/TabletServer.java#L3676-L3677
>
> The JMX metrics reported by the TServer should also likely do the same.
>
> Care to file a JIRA issue?
>
> Roshan Punnoose wrote:
> > I am using Accumulo 1.6.5 and trying to monitor ingest rates
> through JMX
> > endpoints. The ingest reported through JMX does not seem to match
> with
> > the ingest rates reported on the accumulo monitor. Looking at the
> source
> > code, I noticed that the TabletServerMBean is counting all the
> entries
> > in memory for online tablets on that TabletServer:
> >
> > result+=tablet.getNumEntriesInMemory()
> >
> >
> > Is this is a good metric to use for ingest rates? Just want a
> good way
> > to measure ingest rates as I change configurations and run the
> > continuous ingest.
> >
> > Thanks!
> > Roshan
>
Re: Accumulo Ingest JMX metric
Posted by Roshan Punnoose <ro...@gmail.com>.
Just filed: https://issues.apache.org/jira/browse/ACCUMULO-4334
Thanks Josh!
On Wed, Jun 8, 2016 at 11:13 AM Josh Elser <jo...@gmail.com> wrote:
> Yeah, this looks wrong.
>
> The numbers the monitor gets are from the TabletServer polling each Tablet
>
>
> https://github.com/apache/accumulo/blob/1.6/server/tserver/src/main/java/org/apache/accumulo/tserver/TabletServer.java#L3676-L3677
>
> The JMX metrics reported by the TServer should also likely do the same.
>
> Care to file a JIRA issue?
>
> Roshan Punnoose wrote:
> > I am using Accumulo 1.6.5 and trying to monitor ingest rates through JMX
> > endpoints. The ingest reported through JMX does not seem to match with
> > the ingest rates reported on the accumulo monitor. Looking at the source
> > code, I noticed that the TabletServerMBean is counting all the entries
> > in memory for online tablets on that TabletServer:
> >
> > result+=tablet.getNumEntriesInMemory()
> >
> >
> > Is this is a good metric to use for ingest rates? Just want a good way
> > to measure ingest rates as I change configurations and run the
> > continuous ingest.
> >
> > Thanks!
> > Roshan
>
Re: Accumulo Ingest JMX metric
Posted by Josh Elser <jo...@gmail.com>.
Yeah, this looks wrong.
The numbers the monitor gets are from the TabletServer polling each Tablet
https://github.com/apache/accumulo/blob/1.6/server/tserver/src/main/java/org/apache/accumulo/tserver/TabletServer.java#L3676-L3677
The JMX metrics reported by the TServer should also likely do the same.
Care to file a JIRA issue?
Roshan Punnoose wrote:
> I am using Accumulo 1.6.5 and trying to monitor ingest rates through JMX
> endpoints. The ingest reported through JMX does not seem to match with
> the ingest rates reported on the accumulo monitor. Looking at the source
> code, I noticed that the TabletServerMBean is counting all the entries
> in memory for online tablets on that TabletServer:
>
> result+=tablet.getNumEntriesInMemory()
>
>
> Is this is a good metric to use for ingest rates? Just want a good way
> to measure ingest rates as I change configurations and run the
> continuous ingest.
>
> Thanks!
> Roshan