| Re: [jakartaee-tck-dev] committers: The jakartaeetck-nightly-run-master job will now prompt for the tests to run |
On Sat, Jun 27, 2020 at 3:36 PM Scott Marlow <smarlow@xxxxxxxxxx> wrote:Hi,The Jenkins jakartaeetck-nightly-run-master [1] job that runs the Full Platform TCK tests will now prompt for the list of tests to run. I made this change so we would have a way to run a subset of tests.For example, test run [2] was started with just "jpa_pmservlet jpa_puservlet jpa_stateful3 jpa_stateless3" tests.The full test run (https://ci.eclipse.org/jakartaee-tck/job/jakartaee-tck/job/master/729/console) that I started over the weekend seemed to get stuck in jpa_appmanagedNoTx, so I terminated that run early. I started a new run with only the various JPA tests (including jpa_appmanagedNoTx this time) and all of the JPA tests passed.I'm not yet sure of why this, perhaps we are running out of memory and thrashing. I'm thinking that we could increase the JNLP OS/VM memory size from memory: 2176Mi. This may mean that we are limited to fewer Jenkins VM instances during the test run since each one requires more memory which I think is okay.In install/jakartaee/bin/ts.jte, I do see ri.jvm.options set to -Xmx4096m which I think means the GlassFish process could have a larger Java heap than is available in the 2176 MB VM. We also are loading Derby as well.We previously were running GlassFish with a 2GB java heap but ran out of memory during ejb30 testing (as per 9a2200cdce976402d02c110124ac5d3836e5f577).I think we should first try leaving the GlassFish memory size at 4G and increase JNLP to memory: 6gb (should be more than enough).
Any other suggestions/thoughts?ScottI am not sure exactly of the meaning of the use of "_pmservlet", but assume that indicates which test vehicles to run the JPA tests with. The more obvious use is to specify the test folder path, as in "ejb30/bb" is the tests within the "src/com/sun/ts/tests/ejb30/bb" folder. This may be useful when you need to run a smaller subset of tests.Note that we are still seeing a lot of Jakarta Server Faces failures due to [3] that should be resolved by [4] when that lands in GlassFish 6.We need to better understand the other remaining test failures, perhaps we could try running only some of the failing tests to see if they still fail when run alone.Scott