Re: [hibernate-dev] NoORM IRC meeting minutes
On 2 August 2017 at 13:28, Guillaume Smetwrote: > On Wed, Aug 2, 2017 at 1:21 PM, Sanne Grinovero wrote: >> >> This might work for HV but please don't do this generally? >> Concurrently running builds will fail because of conflicting resource >> usage, >> and we had plenty of cases in which some ports are needd which you >> didn't even know about. > > > At least for all the projects using our unified release scripts it shouldn't > be an issue: we don't run the tests in the release process (you're supposed > to check that all the CI jobs are green before releasing). > > It might still be an issue though if you start resources even if the tests > are not running. Might be the case with OGM and Search. Right, Search for example starts ES even if you skip tests (there's a JIRA to improve on that). But my point though is more about the unknowns: it's painful to figure out such things after the fact so we need to be careful with such jobs. > > -- > Guillaume ___ hibernate-dev mailing list hibernate-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev
Re: [hibernate-dev] NoORM IRC meeting minutes
On Wed, Aug 2, 2017 at 1:21 PM, Sanne Grinoverowrote: > This might work for HV but please don't do this generally? > Concurrently running builds will fail because of conflicting resource > usage, > and we had plenty of cases in which some ports are needd which you > didn't even know about. > At least for all the projects using our unified release scripts it shouldn't be an issue: we don't run the tests in the release process (you're supposed to check that all the CI jobs are green before releasing). It might still be an issue though if you start resources even if the tests are not running. Might be the case with OGM and Search. -- Guillaume ___ hibernate-dev mailing list hibernate-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev
Re: [hibernate-dev] NoORM IRC meeting minutes
On 2 August 2017 at 12:13, Guillaume Smetwrote: > The website jobs are not the only ones that need to be prioritized: the > release jobs should also be prioritized. Good point. > > When you do a release, it's really a pain to wait for all the other jobs to > finish. > > Currently, I set a weight of 2 for the BV/HV release jobs so that they can > also be prioritized. This might work for HV but please don't do this generally? Concurrently running builds will fail because of conflicting resource usage, and we had plenty of cases in which some ports are needd which you didn't even know about. > > No preference about the solution we choose, just wanted to mention it. > > On Wed, Aug 2, 2017 at 12:37 PM, Sanne Grinovero > wrote: >> >> Thank for the IRC logs. To comment on the Jenkins vs 5 Executors "trick": >> >> if it's too limiting +1 to remove this but please always allow a >> website release to allow "right now", sometimes it's really urgent and >> even waiting a couple of fast jobs is very stressful. >> >> A simple solution - and actually how it used to be in the very >> beginning - is to configure all website jobs to run on the master >> node, and no other jobs to allow running on the master node. >> Incidentally this simplifies security as this implies the website >> build doesn't have to be rsynced from the slaves, as it's being hosted >> from this same machine. >> >> The reason we changed this was that some more jobs were requiring to >> run on AWS, like some ORM tests needing to access the RDS hosted >> Oracle databases. With more slaves running on AWS now, that reason is >> no longer valid so we could return to the situation in which the >> master node is dedicated for website and super-light jobs exclusively. >> >> In most cases the reason to avoid one slave to run two (or more) full >> integration tests in parallel was isolation problems, from TCP port >> usage to using each other's local database; we could try Docker again >> if someone is willing to spend some time on such research, but bear in >> mind it's not suited for all our jobs - for example some need to run >> performance sensitive tasks, some others just don't have enough >> physical memory to run two jobs concurrently so we still need some >> form of coordination to avoid running multiple jobs on the same >> machine. >> >> Incidentally our "master node" was also scaled to a slightly higher >> machine class to be able to run those ORM tests (more RAM in >> particular); if we pick the simple road of having it build the >> websites at most, we could re-scale it down again. >> >> Thanks, >> Sanne >> >> >> >> On 1 August 2017 at 22:16, Guillaume Smet >> wrote: >> > Hi! >> > >> > Here are the minutes of today's NoORM meeting: >> > >> > 15:48 < jbott> Meeting ended Tue Aug 1 13:48:33 2017 UTC. Information >> > about MeetBot at >> >http://wiki.debian.org/MeetBot . (v 0.1.4) >> > 15:48 < jbott> Minutes: >> > >> > http://transcripts.jboss.org/meeting/irc.freenode.org/hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.html >> > 15:48 < jbott> Minutes (text): >> > >> > http://transcripts.jboss.org/meeting/irc.freenode.org/hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.txt >> > 15:48 < jbott> Log: >> > >> > http://transcripts.jboss.org/meeting/irc.freenode.org/hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.log.html >> > >> > Cheers, >> > >> > -- >> > Guillaume >> > ___ >> > hibernate-dev mailing list >> > hibernate-dev@lists.jboss.org >> > https://lists.jboss.org/mailman/listinfo/hibernate-dev >> ___ >> hibernate-dev mailing list >> hibernate-dev@lists.jboss.org >> https://lists.jboss.org/mailman/listinfo/hibernate-dev > > ___ hibernate-dev mailing list hibernate-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev
Re: [hibernate-dev] NoORM IRC meeting minutes
The website jobs are not the only ones that need to be prioritized: the release jobs should also be prioritized. When you do a release, it's really a pain to wait for all the other jobs to finish. Currently, I set a weight of 2 for the BV/HV release jobs so that they can also be prioritized. No preference about the solution we choose, just wanted to mention it. On Wed, Aug 2, 2017 at 12:37 PM, Sanne Grinoverowrote: > Thank for the IRC logs. To comment on the Jenkins vs 5 Executors "trick": > > if it's too limiting +1 to remove this but please always allow a > website release to allow "right now", sometimes it's really urgent and > even waiting a couple of fast jobs is very stressful. > > A simple solution - and actually how it used to be in the very > beginning - is to configure all website jobs to run on the master > node, and no other jobs to allow running on the master node. > Incidentally this simplifies security as this implies the website > build doesn't have to be rsynced from the slaves, as it's being hosted > from this same machine. > > The reason we changed this was that some more jobs were requiring to > run on AWS, like some ORM tests needing to access the RDS hosted > Oracle databases. With more slaves running on AWS now, that reason is > no longer valid so we could return to the situation in which the > master node is dedicated for website and super-light jobs exclusively. > > In most cases the reason to avoid one slave to run two (or more) full > integration tests in parallel was isolation problems, from TCP port > usage to using each other's local database; we could try Docker again > if someone is willing to spend some time on such research, but bear in > mind it's not suited for all our jobs - for example some need to run > performance sensitive tasks, some others just don't have enough > physical memory to run two jobs concurrently so we still need some > form of coordination to avoid running multiple jobs on the same > machine. > > Incidentally our "master node" was also scaled to a slightly higher > machine class to be able to run those ORM tests (more RAM in > particular); if we pick the simple road of having it build the > websites at most, we could re-scale it down again. > > Thanks, > Sanne > > > > On 1 August 2017 at 22:16, Guillaume Smet > wrote: > > Hi! > > > > Here are the minutes of today's NoORM meeting: > > > > 15:48 < jbott> Meeting ended Tue Aug 1 13:48:33 2017 UTC. Information > > about MeetBot at > >http://wiki.debian.org/MeetBot . (v 0.1.4) > > 15:48 < jbott> Minutes: > > http://transcripts.jboss.org/meeting/irc.freenode.org/ > hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.html > > 15:48 < jbott> Minutes (text): > > http://transcripts.jboss.org/meeting/irc.freenode.org/ > hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.txt > > 15:48 < jbott> Log: > > http://transcripts.jboss.org/meeting/irc.freenode.org/ > hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.log.html > > > > Cheers, > > > > -- > > Guillaume > > ___ > > hibernate-dev mailing list > > hibernate-dev@lists.jboss.org > > https://lists.jboss.org/mailman/listinfo/hibernate-dev > ___ > hibernate-dev mailing list > hibernate-dev@lists.jboss.org > https://lists.jboss.org/mailman/listinfo/hibernate-dev > ___ hibernate-dev mailing list hibernate-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev
Re: [hibernate-dev] NoORM IRC meeting minutes
Thank for the IRC logs. To comment on the Jenkins vs 5 Executors "trick": if it's too limiting +1 to remove this but please always allow a website release to allow "right now", sometimes it's really urgent and even waiting a couple of fast jobs is very stressful. A simple solution - and actually how it used to be in the very beginning - is to configure all website jobs to run on the master node, and no other jobs to allow running on the master node. Incidentally this simplifies security as this implies the website build doesn't have to be rsynced from the slaves, as it's being hosted from this same machine. The reason we changed this was that some more jobs were requiring to run on AWS, like some ORM tests needing to access the RDS hosted Oracle databases. With more slaves running on AWS now, that reason is no longer valid so we could return to the situation in which the master node is dedicated for website and super-light jobs exclusively. In most cases the reason to avoid one slave to run two (or more) full integration tests in parallel was isolation problems, from TCP port usage to using each other's local database; we could try Docker again if someone is willing to spend some time on such research, but bear in mind it's not suited for all our jobs - for example some need to run performance sensitive tasks, some others just don't have enough physical memory to run two jobs concurrently so we still need some form of coordination to avoid running multiple jobs on the same machine. Incidentally our "master node" was also scaled to a slightly higher machine class to be able to run those ORM tests (more RAM in particular); if we pick the simple road of having it build the websites at most, we could re-scale it down again. Thanks, Sanne On 1 August 2017 at 22:16, Guillaume Smetwrote: > Hi! > > Here are the minutes of today's NoORM meeting: > > 15:48 < jbott> Meeting ended Tue Aug 1 13:48:33 2017 UTC. Information > about MeetBot at >http://wiki.debian.org/MeetBot . (v 0.1.4) > 15:48 < jbott> Minutes: > http://transcripts.jboss.org/meeting/irc.freenode.org/hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.html > 15:48 < jbott> Minutes (text): > http://transcripts.jboss.org/meeting/irc.freenode.org/hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.txt > 15:48 < jbott> Log: > http://transcripts.jboss.org/meeting/irc.freenode.org/hibernate-dev/2017/hibernate-dev.2017-08-01-13.00.log.html > > Cheers, > > -- > Guillaume > ___ > hibernate-dev mailing list > hibernate-dev@lists.jboss.org > https://lists.jboss.org/mailman/listinfo/hibernate-dev ___ hibernate-dev mailing list hibernate-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/hibernate-dev