Hi Madhur,
Shown described extreme case (not unusual though) and is not hard to
detect since effects will be catastrophic. You can use one of Solr
monitoring tools to see how GC (and other interrupting events such as
commits, segment merges, saturated network) affect Solr numbers. One
such tool is Sematext's SPM: http://sematext.com/spm/. It will give you
enough info to pinpoint issues and tune Solr and JVM.
Thanks,
Emir
--
Monitoring * Alerting * Anomaly Detection * Centralized Log Management
Solr & Elasticsearch Support * http://sematext.com/
On 25.07.2016 18:07, Shawn Heisey wrote:
On 7/24/2016 11:13 PM, Madhur Adlakha wrote:
To whom so ever it may concern,
I have been fetching certain Solr metrics and keeping a track of them
graphically on Grafana, there are a few queries I wanted to clear, as follows:
* How and which metrics of Solr are affected by the Garbage collector?
* Which metrics of garbage collector should we track that might have
implications on Solr?
Garbage collection can result in a complete pause of the Java virtual
machine. This pause can be quite long -- I've seen 15-20 seconds in a
single full GC pause in a running Solr 4.x install with no GC tuning.
When this happens, Java *only* does garbage collection. *ANYTHING* that
Solr is doing when the pause occurs will be delayed, so any performance
metric might be affected.
Pause time is what you need to track. When there are GC logs, you can
load those logs into GCViewer and get a lot of statistics.
https://github.com/chewiebug/GCViewer
Another good tool for calculating JVM pauses from *any* source is jHiccup.
https://www.azul.com/jhiccup/
The start script in Solr 5.0 and later has default GC tuning parameters
that work pretty well for typical heap sizes of a few gigabytes. If you
have a particularly large heap, you may need different GC tuning. I
have done some work on G1 tuning parameters.
https://wiki.apache.org/solr/ShawnHeisey#GC_Tuning_for_Solr
Thanks,
Shawn
