question about jmx value (avgRequestsPerSecond) output from solr

[geeky2 <ge...@hotmail.com>]

hello all,

environment: centOS, solr 3.5, jboss 5.1

i have been using wily (a monitoring tool) to instrument our solr instances
in stress.

can someone help me to understand something about the jmx values being
output from solr?  please note - i am new to JMX.

problem / issue statement: for a given request handler (partItemDescSearch),
i see output from the jmx MBean for the metric avgRequestsPerSecond - AFTER
my test harness has completed and there is NO request activity to this
request handler - taking place (verified in solr log files).

example scenario during testing:  during a test run - the test harness will
fire requests at request handler (partItemDescSearch) and all numbers look
fine.   then after the test harness is done - the metric
avgRequestsPerSecond does not immediately drop to 0.  instead - it appears
as if JMX is somehow averaging this metric and gradually trending it
downward toward 0.

continual checking of this metric (in the JMX tree - see screen shot) shows
the number trending downward instead of a hard stop at 0.

is this behavior - just the way jmx works?

thanks mark

http://lucene.472066.n3.nabble.com/file/n3991616/test1.bmp 




--
View this message in context: http://lucene.472066.n3.nabble.com/question-about-jmx-value-avgRequestsPerSecond-output-from-solr-tp3991616.html
Sent from the Solr - User mailing list archive at Nabble.com.

Re: question about jmx value (avgRequestsPerSecond) output from solr

[Chris Hostetter <ho...@fucit.org>]

: example scenario during testing:  during a test run - the test harness will
: fire requests at request handler (partItemDescSearch) and all numbers look
: fine.   then after the test harness is done - the metric
: avgRequestsPerSecond does not immediately drop to 0.  instead - it appears
: as if JMX is somehow averaging this metric and gradually trending it
: downward toward 0.
: 
: continual checking of this metric (in the JMX tree - see screen shot) shows
: the number trending downward instead of a hard stop at 0.
: 
: is this behavior - just the way jmx works?

It's nothing special about JMX, it's just the way RequestHandlerBase 
computes this number -- it's the average number of requests per second 
that this handler has seen over it's entire lifetime -- ie: from the 
moment the SOlrCore was created/started until now.

it is quite simply...

	(numRequests*1000) / (currentTimeMillis()-handlerStart)

If you want an average over a finer granularity of time then that you can 
have your JMX client sample the numRequests stat from the hander at some 
interval (ie: record the value every minute, subtract the last value from 
the current value, divide by 60 and that will tell you the average number 
of reuqests per second, sampled per minute.



-Hoss