schernic opened a new issue, #14327:
URL: https://github.com/apache/druid/issues/14327

   ### Description
   
   Provide a way to disable collection (not only emission) of query metrics. 
Alternatively, introduce sampling to the metric collection or make the 
collected metric types and dimensions configurable.
   
   ### Motivation
   
   For certain workloads, we have experienced an unusually high CPU load on our 
historical nodes. Often times the load increased to the point that queries 
started to time out and Zookeeper considered the node unavailable. Out of 
curiosity I used a profile to look into the CPU hot spots and found that metric 
generation was causing most of that load.  As we are not even emitting metrics 
I was looking for a way to disable metric collection altogether but found no 
way to configure that.
   Nevertheless, checking the code I found a way to disable metric collection 
during runtime using JMX in order to check if that would really help with the 
high CPU load. It reduced the load on that historical by 80% which makes me 
believe that some features around making it possible to disable collection of 
query metrics or dimension could be valuable addition for others as well. 
Actually, I don't see why metric collection would be enabled by default if no 
emitter was configured.
   
   Sampling or rate limiting of metric collections would also be useful options 
in cases like ours.
   
   


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to