1.4.2 comes with a couple optimization which might explain the drop in field data cache and lucene memory (most notably here is the disabling of bloom filters by default - due to improvements in lucene they are not as needed as before and we can save on the memory).
The IO spike might come from https://github.com/elasticsearch/elasticsearch/pull/9683 , which is still under heavy discussion. Can you try running _hot_threads on you cluster and see whether you there's anything stats releated there? On Wednesday, February 11, 2015 at 5:57:11 PM UTC+1, George Stathis wrote: > > Hi there, > > We just completed this past Monday an ES upgrade from 1.1.1 to 1.4.2 and a > Java upgrade from 7 to 8. No other changes. Same boxes, same specs, no new > queries or filters, no increased requests. What I see in Marvel since the > upgrade is a large spike in Disk IO: > > > <https://lh6.googleusercontent.com/-i6OYBiqputk/VNuJC-47eII/AAAAAAAAA_g/TWxTcres-Uk/s1600/Screen%2BShot%2B2015-02-11%2Bat%2B11.37.58%2BAM.png> > > and a large dip in field data, filter cache and overall index memory > consumption: > > > <https://lh5.googleusercontent.com/-2DRu7af1npc/VNuJNqsvOQI/AAAAAAAAA_o/56trKYYMiQw/s1600/Screen%2BShot%2B2015-02-11%2Bat%2B11.37.47%2BAM.png> > > Should I be looking at the new circuit breaker > <http://www.elasticsearch.org/guide/en/elasticsearch/reference/current/index-modules-fielddata.html#circuit-breaker> > settings? > > Any pointers are welcome! > > -GS > > -- 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 elasticsearch+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/elasticsearch/d9ed771b-0c8b-4e32-863b-d33b1cf94da0%40googlegroups.com. For more options, visit https://groups.google.com/d/optout.
