jvanneman commented on PR #3891:
URL: https://github.com/apache/solr/pull/3891#issuecomment-3725805435

   > I do have a slight worry that maybe an errant request that takes way too 
long for reasons specific to the request
   
   @dsmiley I think as long as folks are using `timeAllowed` and set the order 
of timeouts as socketTimeout > timeAllowed, then this will only trigger when 
processing time + in flight time + jetty queue time is > socketTimeout. In 
practice what i think would usually happen is that a request would be stuck in 
the jetty queue for a long time before it's processed and the cumulative 
processing time + queue time would trigger socketTimeout before timeAllowed is 
up. In which case I think it's not a bad idea to mark the server as 
unresponsive because it means there's enough requests queued to degrade 
performance. 
   
   One edge case to this might be [two phase requests like 
grouping](https://github.com/apache/solr/blob/c011ac05859b4efe3913b0c6bf8ce73a01f28daf/solr/core/src/java/org/apache/solr/search/grouping/distributed/requestfactory/TopGroupsShardRequestFactory.java#L90).
 It could plausibly make sense to have socketTime < timeAllowed in which the 
socketTimeout servces as the upper bound for phase 1 and leaves the request 
some room to complete phase 2. This is more likely to results in the scenario 
you describe. It might be worth a different mechanism to restrict first phase 
time in those types of requests though, I don't think setting socketTimeout to 
accomplish that makes sense.
   
   Updated the PR based on your suggestions!


-- 
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