Remove use of at-async in SSLStream close #48
Conversation
Should fix #44. I can't quite remember why we called at-async from close to close the underlying IO, but it shouldn't make that much of a difference to run this inline w/o at-async and makes this not cause problems when run in finalizers.
|
Or @ancapdev, it sounds like you've also seen the issue; can you try this branch and report back? |
I can give it a go. My failures were on large quarterly batch jobs, using a lot of resources, and then failing once every couple of days. It's hard to test reliably without a quicker reproducible. Does the JuliaWeb team run any kind of soak testing? |
|
I've not been able to come up with a smaller reproducible that triggers the issue, but the changes here can't make things worse, so maybe we can get that in and monitor? A couple of questions
|
| end | ||
| @async try | ||
| try | ||
| Base.close(ssl.io) |
There was a problem hiding this comment.
Won't close acquire the iolock, and therefore is not permitted in a finalizer either?
3 participants
Should fix #44. I can't quite remember why we called at-async from close to close the underlying IO, but it shouldn't make that much of a difference to run this inline w/o at-async and makes this not cause problems when run in finalizers.