Christoffer,

How much JVM heap are you giving ES and what are the size of the sets? 
 According to 
this 
http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/index-modules-fielddata.html
 
it looks like in 1.4 you will be able to control the circuit breaker more 
via config.  However, depending on your data set size I am guessing you are 
still going to have to worry what you can allocate to the ES heap since 
that page seems to indicate the circuit breakers are defaulted to 
reasonably high %.

I am trying to look into the scalability characteristics of this feature 
myself because it is iterating for some goals I have, but I don't see any 
information about how it scales or what it is bound by.  In my case I would 
like to be able to analyse foreground sets of 10s to 100s of thousands of 
documents against a bg set of millions.  Without finding anything 
documented your #s might give me an idea if my use is crazy or reasonable 
prior getting some testing done with it.

Kevin

On Friday, September 5, 2014 3:19:13 AM UTC-5, Christoffer Vig wrote:
>
> The significant terms aggregation is a really great feature that allows 
> for some really interesting data analysis. We quite often experience out of 
> memory errors, "CircuitBreakingException: Data too large, data would be 
> larger than limit"
> Which is not hard to understand, due to the amount of data and the speed 
> requirements. 
>
> I think it would be interesting if it was possible to "trade off" speed to 
> allow deeper analysis. To run significant terms, and possibly other 
> aggregations, allow them to run for as long as needed, just to return some 
> (presumably correct) results. 
>
>

-- 
You received this message because you are subscribed to the Google Groups 
"elasticsearch" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
To view this discussion on the web visit 
https://groups.google.com/d/msgid/elasticsearch/00fb6efa-e869-4672-afd6-673c995f1506%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to