Track computations from autoruns and ensure they're all stopped on publish stop #38
Conversation
| # created from cursors inside the autorun will exist forever. | ||
| while trackedComputations.length | ||
| computation = trackedComputations.shift() | ||
| computation?.stop() |
There was a problem hiding this comment.
This is strange that this helps you, because handle is the same as computation and I already stop them? So the whole idea why I collect handles there is to clean them up.
There was a problem hiding this comment.
Maybe it is not the same, but then this is very strange and very problematic because it might point to a bug in server-side tracker implementation.
There was a problem hiding this comment.
@mitar would a reproduction repo help? I think the issue is that the computation isn't cleaned up ever because the publication isn't active nor is a computation returned from my autorun in this case. If I return a computation it does work, but can't do that since the pub needs to return a cursor
There was a problem hiding this comment.
Well your current implementation works only if a computation is returned from the autorun, but not if a cursor is.
Given meteorhacks:unblock is the only way to reproduce this, it seems like a race condition but I really don't know.
There was a problem hiding this comment.
Well your current implementation works only if a computation is returned from the autorun, but not if a cursor is.
No, result of calling Tracker.autorun is always a handle/computation of this autorun. It does not matter what internally is returned.
There was a problem hiding this comment.
Also, not sure why you want to use unblock, BTW. There are some strange side effects and it uses a global variable I am pretty scared of seeing.
|
Also, please add a test which deterministically shows this problem and then how it is fixed. |
|
I've written some tests to ensure Meteor's multiplexers aren't left hanging around. Originally my approach was to track the computations from the subscription in the test, but I found it much cleaner to do a count of the multiplexers before + after to prove it no longer leaks. If you comment out Let me know your thoughts - happy to change approach if you think there's a better way. |
|
I think I found the problem. I will merge this and update stuff a bit. Thanks for finding this, reporting it, and making a PR. |
Fixes #37