Did INFRA give any hint as to when they might provide the bigger VMs? On Mon, Apr 17, 2023 at 8:24 PM Sai Boorlagadda <sai.boorlaga...@gmail.com> wrote:
> I went ahead and merged github workflow jobs that tests > WAN, CQ, Assembly and Managment distributed tests. > > Free workers VMs has 2 cores and tuning any sort of > parameters isn't speeding up geode-core DUnits. > > Talking to infra team found that infra is working on providing > self-hosted (sponsored by infra) that are much bigger VMs. > > So until such VMs are available I am going to find if there are any > alternate solution. > > On Thu, 13 Apr 2023 at 15:08, Kirk Lund <kl...@apache.org> wrote: > > > I see that there is at least one person concerned about DUnit tests > > requiring longer timeouts. This is the current situation with an unknown > > number of the DUnit tests. One possibility is to move the worst offenders > > to a new src set within geode-core and then give that its own job with a > > larger timeout. The longer term solution is to fix or even rewrite some > of > > those tests. Excluding them is also not a viable option as we risk > > losing important test coverage that way. I agree that some of these tests > > need a lot more help than tweaking overall job timeout values, but > without > > a lot more time commitment from contributors that might not be an > > option for some time. > > > > -Kirk > > > > On Thu, Apr 13, 2023 at 3:02 PM Kirk Lund <kl...@apache.org> wrote: > > > > > Is the coreDistributedTests the only dunit job that currently takes too > > > long? If it is we may want to split that into more than one job. > > > > > > -Kirk > > > > > > On Wed, Apr 12, 2023 at 7:58 PM Sai Boorlagadda < > > sai.boorlaga...@gmail.com> > > > wrote: > > > > > >> All, > > >> > > >> There is an upper bound for job execution time on free workers (set > to 6 > > >> hours max[1]), which can be configured beyond 6hrs with a self-hosted > > >> worker. All of our pipeline jobs are using `--max-workers` to > > parallelize > > >> gradle tasks but `testMaxParallelForks` is left to default which is > > (1/4th > > >> of the available CPU cores), so primarily due to running only a single > > >> test > > >> in each parallel fork geode-core distribution tests are taking more > > than 6 > > >> hours. Other than finding a solution for core distributed tests, most > > >> DUnit > > >> tests are passed[2] by splitting them into individual jobs (WAN, CQ, > > >> Lucene, assembly, management). > > >> > > >> Will reach out to infra team and trying playing with `--max-workers` > to > > >> parallelize more tests than having to run parallel tests with in a > fork > > >> would be options. > > >> > > >> I am going to wait for few days to get answers from infra team before > I > > >> can > > >> create a PR to add at least the passing DUnits. > > >> > > >> [1] > > >> > > >> > > > https://docs.github.com/en/actions/learn-github-actions/usage-limits-billing-and-administration > > >> [2] https://github.com/apache/geode/actions/runs/4639012912 > > >> > > >> Sai > > >> > > > > > >
