lucene-solr-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mark Miller <markrmil...@gmail.com>
Subject Re: Solr Startup CPU Spike
Date Tue, 09 Mar 2010 17:38:33 GMT
Ah - loading the fieldcache - do you have a *lot* of unique terms in the 
fields you are sorting/faceting on?

localhost:8983/solr/admin/luke is helpful for checking this.


-- 
- Mark

http://www.lucidimagination.com



On 03/09/2010 12:33 PM, John Williams wrote:
> Yonik,
>
> I have provided an image below gives details on what is causing the blocked http thread.
Is there any way to resolve this issue.
>
> Thanks,
> John
>
> --
> John Williams
> System Administrator
> 37signals
>
>    
>
>
>
> On Mar 9, 2010, at 10:41 AM, John Williams wrote:
>
>    
>> Yonik,
>>
>> I got yourkit setup to profile the Tomcat instance and as you will see in the graph
below all of the   http threads are blocked (red) until around 4:40. This is the point where
the instance becomes responsive and CPU usage drops. I have also ruled out GC being the issue
by using the GC monitoring in yourkit. Let me know your thoughts and if you have any questions.
>>
>> Thanks for your assistance.
>>
>> Thanks,
>> John
>>
>> --
>> John Williams
>> System Administrator
>> 37signals
>>
>> <Screen shot 2010-03-09 at 10.35.15 AM.png>
>> On Mar 8, 2010, at 5:28 PM, Yonik Seeley wrote:
>>
>>      
>>> On Mon, Mar 8, 2010 at 6:07 PM, John Williams<john@37signals.com>  wrote:
>>>        
>>>> Yonik,
>>>>
>>>> In all cases our "autowarmCount" is set to 0. Also, here is a link to our
config. http://pastebin.com/iUgruqPd
>>>>          
>>> Weird... on a quick glance, I don't see anything in your config that
>>> would cause work to be done on a commit.
>>> I expected something like autowarming, or rebuilding a spellcheck
>>> index, etc.  I assume this is happening even w/o any requests hitting
>>> the server?
>>>
>>> Could it be GC?  You could use -verbose:gc or jconsole to check if
>>> this corresponds to a big GC (which could naturally hit on an index
>>> change).  5 minutes is really excessive though, and I wouldn't expect
>>> it on startup.
>>>
>>> If it's not GC, perhaps the next step is to get some stack traces
>>> during the spike (or use a profiler) to figure out where the time is
>>> being spent.  And verify that the solrconfig.xml shown actually still
>>> matches the one you provided.
>>>
>>> -Yonik
>>> http://www.lucidimagination.com
>>>
>>>
>>>
>>>        
>>>> Thanks,
>>>> John
>>>>
>>>> --
>>>> John Williams
>>>> System Administrator
>>>> 37signals
>>>>
>>>> On Mar 8, 2010, at 4:44 PM, Yonik Seeley wrote:
>>>>
>>>>          
>>>>> Is this just autowarming?
>>>>> Check your autowarmCount parameters in solrconfig.xml
>>>>>
>>>>> -Yonik
>>>>> http://www.lucidimagination.com
>>>>>
>>>>> On Mon, Mar 8, 2010 at 5:37 PM, John Williams<john@37signals.com>
 wrote:
>>>>>            
>>>>>> Good afternoon.
>>>>>>
>>>>>> We have been experiencing an odd issue with one of our Solr nodes.
Upon startup or when bringing in a new index we get a CPU spike for 5 minutes or so. I have
attached a graph of this spike. During this time simple queries return without a problem but
more complex queries do not return. Here are some more details about the instance:
>>>>>>
>>>>>> Index Size: ~16G
>>>>>> Max Heap: 6144M
>>>>>> GC Option: -XX:+UseConcMarkSweepGC
>>>>>> System Memory: 16G
>>>>>>
>>>>>> We have a very similar instance to this one but with a much larger
index that we are not seeing this sort of issue.
>>>>>>
>>>>>> Your help is greatly appreciated. Let me know if you need any additional
information.
>>>>>>
>>>>>> Thanks,
>>>>>> John
>>>>>>
>>>>>> --
>>>>>> John Williams
>>>>>> System Administrator
>>>>>> 37signals
>>>>>>
>>>>>>
>>>>>>
>>>>>>              
>>>>
>>>>          
>>      
>    




Mime
  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message