Not really a review, but: For many years we've been using 10 seconds (scaled by timeout factor) as a duration long enough that a timeout is a real failure. Which is close to your own 20 seconds. Of course, no value is surely safe.
Probably, parallel testing infrastructure for timeouts should be a test library method. I do something similar in JSR166TestCase /** * Runs all the given actions in parallel, failing if any fail. * Useful for running multiple variants of tests that are * necessarily individually slow because they must block. */ void testInParallel(Action ... actions) { ExecutorService pool = Executors.newCachedThreadPool(); try (PoolCleaner cleaner = cleaner(pool)) { On Fri, Aug 30, 2019 at 6:23 AM Daniel Fuchs <daniel.fu...@oracle.com> wrote: > On 30/08/2019 13:54, Pavel Rappo wrote: > > Updated, > > > > http://cr.openjdk.java.net/~prappo/8151678/webrev.01/ > > > > Changes look good! > > best regards, > > -- daniel >