Build https://builds.apache.org/job/TomEE/job/master-build-full/776/ finished after 4 hr 25 min, still not much improvement from the average build time with the MAVEN_OPTS-Xmx1536m https://builds.apache.org/job/TomEE/job/master-build-full/buildTimeTrend
I reverted MAVEN OPTS to -Xmx1536m and removed the maven property help:system in the job https://ci-builds.apache.org/job/TomEE/job/master-build-full/ El vie, 14 oct 2022 a las 12:29, Cesar Hernandez (<cesargu...@gmail.com>) escribió: > Anyway long story short: yes, our build could be faster with better >> hardware. Just not very clear what that hardware should be. > > > We don't know what is running under the hood in > https://ci-builds.apache.org/ for on-demand workers, but I do see other > projects that have specific controller types: > https://jenkins-ccos.apache.org/ > From the numbers so far, it seems that anything bigger than AWS t2.xlarge > (4 vCpus, 16 GB ram) can speed by half the current timing executor in the > apache infra is taking > https://builds.apache.org/job/TomEE/job/master-build-full/buildTimeTrend > > Based on Jonathan's feedback, I reviewed, and the current > https://builds.apache.org/job/TomEE/job/master-build-full is using in > MAVEN_OPTS = -Xmx1536m > I wonder if the result shared in this thread have specific MAVEN_OPTS > values, mine are = -Xms2g -Xmx8g > > I updated https://builds.apache.org/job/TomEE/job/builder-info/ to show > the worker memory, and it seems we have 18 GB free out of 94 GB. [1], but > this is distributed amount of other projects running in that worker, so > there is no guarantee we will always have X or Y GB of RAM available. > I don't have access to check the runner's limits peer node in the entire > builds.apache.org. and > https://cwiki.apache.org/confluence/display/INFRA/Jenkins+node+labels > documentation seems to be outdated. > > > Today's build already finished > https://builds.apache.org/job/TomEE/job/master-build-full/774/ and took ~ > 4 hour. > I updated https://builds.apache.org/job/TomEE/job/master-build-full/ with > the following changes: > > a) Changed MAVEN_OPTS-Xmx1536m to MAVEN_OPTS=-Xmx4096m > b) Changed -U --show-version --fail-at-end clean install -Pall-adapters > -Dfile.encoding=UTF-8 to -U help:system --show-version --fail-at-end clean > install -Pall-adapters -Dfile.encoding=UTF-8 > The only change is I added help:system maven argument to maven to > double-check the OTPS are read. > c) Build https://builds.apache.org/job/TomEE/job/master-build-full/776/ > is currently running with a) and b) applied, I don't expect to see major > improvements even if we are now using 4GB and the load on the shared server > is unpredictable. > > After building > https://builds.apache.org/job/TomEE/job/master-build-full/776/ I plan to > revert to previous job configurations if need it. > > > [1] > > free -g > total used free shared buff/cache > available > Mem: 94 10 18 0 64 > 82 > > > > > > > El vie, 14 oct 2022 a las 8:18, Jonathan Gallimore (< > jonathan.gallim...@gmail.com>) escribió: > >> Taking the commands very literally, the timings for 2 machines here are >> below. I'm curious though - are there any JAVA_OPTS or MAVEN_OPTS we >> should >> use here (e.g. max heap)? The slower machine here seems to be running at >> an >> absolute crawl 2/3rds of the way through the build here with the default >> settings. The result below is after a reboot and MAVEN_OPTS="-Xmx4g". >> >> Laptop (WSL2/Ubuntu on Windows 11, i7-12700H, 32GB Ram): >> >> [INFO] >> ------------------------------------------------------------------------ >> [INFO] BUILD FAILURE >> [INFO] >> ------------------------------------------------------------------------ >> [INFO] Total time: 01:36 h >> [INFO] Finished at: 2022-10-14T11:10:08+01:00 >> [INFO] >> ------------------------------------------------------------------------ >> >> Dell Optiplex 7040 "Server" machine (Ubuntu 20.04.5 LTS, i7-6700T, 16GB >> Ram): >> >> [INFO] BUILD FAILURE >> [INFO] >> ------------------------------------------------------------------------ >> [INFO] Total time: 01:57 h >> [INFO] Finished at: 2022-10-14T15:16:58+01:00 >> [INFO] >> ------------------------------------------------------------------------ >> >> On Fri, Oct 14, 2022 at 12:13 AM David Blevins <david.blev...@gmail.com> >> wrote: >> >> > Alright, here are some build times for the runs I've done. Results are >> > quite varied. >> > >> > c1.xlarge : [INFO] Total time: 03:59 h >> > t2.medium : [INFO] Total time: 05:05 h >> > t2.xlarge : [INFO] Total time: 02:30 h >> > t3.medium : [INFO] Total time: 03:21 h >> > my.laptop : [INFO] Total time: 01:56 h >> > >> > Specs for the AWS instances are public. Here are the stats for my >> laptop: >> > >> > MacBook Pro (16-inch, 2019) >> > 2.4 GHz 8-Core Intel Core i9 >> > Memory 64 GB 2667 MHz DDR4 >> > >> > I did try some other AWS instance types, but anything with 2GB or less >> > failed due to out of memory issues. I'd be interested to try some other >> > types, particularly the AMD-based types. >> > >> > Anyway long story short: yes, our build could be faster with better >> > hardware. Just not very clear what that hardware should be. >> > >> > >> > -David >> > >> > >> > > On Oct 11, 2022, at 3:05 PM, David Blevins <david.blev...@gmail.com> >> > wrote: >> > > >> > > All, >> > > >> > > I'm collecting some stats on how long it takes to run our full build >> > exactly as Jenkins does. The goal is to work with them to see if we can >> > get some better hardware -- I assume that will require donations, etc. >> > > >> > > If you'd like to help in collecting data, here's the script I'm >> running: >> > > >> > > - curl >> > >> https://gist.githubusercontent.com/dblevins/b39cc3300bcdd89b426ca33b87b5452b/raw/7c68d4df71e9246c8bf2d0a741f8b145ca5d0820/buildtime.sh >> > | bash >> > > >> > > Send the time reported in the build.log along with your system >> > information (os, number of cores, if you disk is an SSD, etc) >> > > >> > > >> > > -- >> > > David Blevins >> > > http://twitter.com/dblevins >> > > http://www.tomitribe.com >> > > >> > >> > >> > > > -- > Atentamente: > César Hernández. > -- Atentamente: César Hernández.