Skip to content
This repository was archived by the owner on Dec 18, 2018. It is now read-only.

WebKit bots having too many TIME_WAIT sockets #108

Closed
youennf opened this issue Nov 16, 2016 · 8 comments
Closed

WebKit bots having too many TIME_WAIT sockets #108

youennf opened this issue Nov 16, 2016 · 8 comments

Comments

@youennf
Copy link

youennf commented Nov 16, 2016

As shown in https://bugs.webkit.org/show_bug.cgi?id=159166#c21, some WebKit bots WPT servers (localhost-running) may have too many sockets in the TIME_WAIT state. This seems to prevent running the tests correctly for some runs.

Has anybody encountered the issue?
Maybe keep-alive should be implemented in wptserve to reduce the risk?

@youennf
Copy link
Author

youennf commented Nov 16, 2016

Looking at the code, it seems keep-alive is in action.
Not sure why we are seeing so much TIME_WAIT sockets then.

@youennf
Copy link
Author

youennf commented Nov 16, 2016

It might be that mote responses have no content-length header

@gsnedders
Copy link
Member

I spent a few hours last week trying to work out why under high load wptserve seems to sometimes hang, using ab—on OS X, it consistently stopped responding at c.16500, with all connections after that getting stuck in SYN_SENT. I didn't really get anywhere very useful, though! May well be related, though?

In principle, any response without Content-Length should have the connection closed by the server after the end of the response. If that's not true, we should fix that.

@youennf
Copy link
Author

youennf commented Nov 16, 2016

If more responses had content-length headers, we would enable a lot more reuse of sockets.
I am trying to check whether https://github.com/youennf/wptserve/tree/webkit-current-wptserver would reduce the issue.

@gsnedders
Copy link
Member

I have some memory of @jgraham or @Ms2ger having some reason why #86 didn't get merged.

That said, we shouldn't have them stuck in TIME_WAIT even without socket reuse.

@gsnedders
Copy link
Member

We've also had issues with w3c-test.org sometimes stopping responding, so I suspect there might be some underlying issue regardless.

@gsnedders
Copy link
Member

FWIW, I tried to look into why this happened and I couldn't figure out why it stopped working at the specific point it did looking at DTrace. That said, I've never tried to deal with anything at this level on OS X previously. Equally, I couldn't reproduce it on Linux artificially limiting number of sockets.

@wpt-issue-mover
Copy link
Collaborator

This issue has been moved to web-platform-tests/wpt#8369; please continue all discussion there.

@w3c w3c locked and limited conversation to collaborators Nov 20, 2017
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants