# lp-pinot-poc
- m
Mayank
10/03/2020, 10:29 PMNice @User - x
Xiang Fu
10/03/2020, 10:30 PMalso there is no latency bump up during segment build - m
Mayank
10/03/2020, 10:30 PMSeems like our settings worked very well - x
Xiang Fu
10/03/2020, 10:31 PMright - m
Mayank
10/03/2020, 10:31 PMđź‘Ťđź‘Ť - x
Xiang Fu
10/06/2020, 6:24 AMHi @User , I’ve tested Pinot on JDK 11 with ZGC, the p99 latency is comparatively better than jdk8+ g1gc Below is the graph of switching it. Just for your reference: the docker image I’m using is
theapachepinot/pinot:0.6.0-SNAPSHOT-1126cac9b-20201002-jdk11
isjvmOpts-javaagent:/opt/pinot/etc/jmx_prometheus_javaagent/jmx_prometheus_javaagent-0.12.0.jar=80:/opt/pinot/etc/jmx_prometheus_javaagent/configs/pinot.yml -Xms24G -Xmx24G -XX:+UnlockExperimentalVMOptions -XX:+UseZGC -XX:+UseNUMA -XX:ConcGCThreads=4 -XX:ParallelGCThreads=4 -Xlog:gc*:/opt/pinot/gc-pinot-server.log - x
Xiang Fu
10/06/2020, 6:25 AMHowever p999 is still around 1sec - a
Andrew First
10/12/2020, 1:24 PM@User that’s promising! how would i access the image via Helm? - x
Xiang Fu
10/12/2020, 3:36 PMyou can directly update the image tag in the helm script - x
Xiang Fu
10/12/2020, 3:37 PMCopy codeimage: repository: apachepinot/pinot tag: latest pullPolicy: IfNotPresent - x
Xiang Fu
10/12/2020, 3:37 PMhere you can change
totag0.6.0-SNAPSHOT-1126cac9b-20201002-jdk11đź‘Ť 1 - a
Andrew First
10/13/2020, 4:00 PMthis is the result with those settings. still degrades after a couple of hours - a
Andrew First
10/13/2020, 4:00 PMScreen Shot 2020-10-13 at 11.59.46 AM.png - a
Andrew First
10/13/2020, 4:01 PMScreen Shot 2020-10-13 at 12.01.40 PM.png - a
Andrew First
10/13/2020, 4:02 PMScreen Shot 2020-10-13 at 12.02.33 PM.png - x
Xiang Fu
10/13/2020, 5:31 PMhmm - x
Xiang Fu
10/13/2020, 5:31 PMhave you seen data movement happens to offline servers? - x
Xiang Fu
10/13/2020, 5:32 PMcan you break that number of segments in Table to be server level metrics - a
Andrew First
10/14/2020, 3:34 PMhm, i don’t see that happening - a
Andrew First
10/14/2020, 3:34 PMit’s all showing as REALTIME - a
Andrew First
10/14/2020, 3:35 PMi thought it was configured to move completed segments to offline - a
Andrew First
10/14/2020, 3:35 PMi also see that a lot of segments are in a “bad” state - a
Andrew First
10/14/2020, 3:35 PMScreen Shot 2020-10-14 at 11.31.50 AM.png - a
Andrew First
10/14/2020, 3:37 PMif i click one of the bad segments, there’s no further information explaining why - a
Andrew First
10/14/2020, 3:37 PMScreen Shot 2020-10-14 at 11.37.24 AM.png - k
Kishore G
10/14/2020, 3:42 PMits a minor UI bug - it shows bad when its getting converted from consuming to ONLINE - k
Kishore G
10/14/2020, 3:42 PM - x
Xiang Fu
10/14/2020, 4:08 PMIf all the segments are still on realtime nodes, then there is some issue there - x
Xiang Fu
10/14/2020, 4:08 PMfrom my benchmark, once segments are persisted, they will be moved to offline servers - x
Xiang Fu
10/14/2020, 4:09 PMdo you have controller log or we can do a zoom call to look at this