I'd be curious how they handle false positives... e.g. tasks that appear stuck (due to GC pauses, I/O stalls, etc.) vs truly dead ones. I have seen that overzealous cleanup can do more damage than letting a zombie linger. That being said, there is obviously an upper limit to letting zombies linger.
Really good article. We had been debating moving from a monolith to services in a distributed environment a while back, and I recommended real baby steps - lets not do full blown services but first break up some of the components so everything isn't deployed together. Guess what? Zombie tasks - albeit not that many, but tracking them down is a bear.
3 comments
[ 3.3 ms ] story [ 20.8 ms ] thread