It seems a good amount of flakiness in the GTK and WPE layout tests is caused by the in-webprocess memory pressure monitor. You can see that in the following logs: https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r232868%20(7018)/results.html <-- the crash is our memory pressure killer https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r232867%20(7017)/results.html <-- second crash is our memory pressure killer https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r232866%20(7016)/results.html <-- the crash is our memory pressure killer https://build.webkit.org/results/GTK%20Linux%2064-bit%20Release%20(Tests)/r232864%20(7015)/results.html <-- first crash If it helped us find tests using too much memory, it might be useful, but it does not because the test that dies is probably just running in a process that previously leaked a bunch of memory... i.e. the test that is killed is probably not the test that has a real problem. The ideal solution would be to fix the cause of those leaks that accumulate over the course of those runs. But meanwhile we don't have that right fix a workaround that we can do now to avoid flakiness in our tests is to disable this memory pressure monitor when running with WTR.
Created attachment 342820 [details] Patch
Comment on attachment 342820 [details] Patch Thanks. I had been complaining about this all year. This is really only harmful when running layout tests.
Committed r232881: <https://trac.webkit.org/changeset/232881>