Repository navigation
parallel/test-listen-fd-cluster is flaky #1931
Description
Activity
The test has a few problems:
- top-level exits immediately after killing its child (called "parent"), and the cluster master. This is a race, it can exit before the kills take effect. It should probably kill its direct child, then wait for it.
- it doesn't effectively protect against child process leaking, it should attach an ipc control channel when it spawns the parent, and it spawns the master, so that those process can do
process.on('disconnect', process.exit), preventing them living beyond their parent - it has one too many process, the "parent" seems wholly unnecessary
I'll see if I can fix, I'm about to get on a plane.
@sam-github we just found a similar behaviour in
test-listen-fd-server. Would you mind having a look?@jbergstroem,
test-listen-fd-server.jswas a duplicate oftest-listen-fd-cluster.js, I pushed a rework of it onto #1944Is this still an issue? Or was this fixed by @sam-github's commits on June 10?
I figure we can close this for now, we'll reopen if we see another failure.
This test leaves trailing processes every now and then on our jenkins buildbots. Since a lot of tests reuses common.PORT, it will make following runs on that host fail.
When failing, it also times out, as seen on most hosts here: https://jenkins-iojs.nodesource.com/job/iojs+any-pr+multi/788/