Timeout reads and writes in the http server.

[tomwilkie]

Fixes #1916

[paulbellamy]

May need to adjust the timeouts in common/xfer/websocket.go as well, probably?

[tomwilkie]

Ah yes websockets!

On Wednesday, 12 October 2016, Paul Bellamy [email protected]
wrote:

@paulbellamy commented on this pull request.

May need to adjust the timeouts in common/xfer/websocket.go as well,
probably?

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
#1917 (review),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AAbGhcLtGAm_YUZS0c7s5LlgkD0b4p6dks5qzPjxgaJpZM4KTAQZ
.

[rade]

Won't this interact badly with persistent http connections?

[2opremio]

@tomwilkie Out of curiosity, how have you confirmed that this fixes #1916 ?

[rade]

And I am still waiting for an answer to my question re persistent connections.

[rade]

@ekimekim could you take over on this, e.g. address review comments, answer questions above, perhaps do a deploy to dev to test whether it solves #1916 (unless you can think of a better way to test this)?

[ekimekim]

I'm attempting to reproduce locally. So far what I've tried: making a connection which hangs (stops sending data) during the sending of a POST body. This causes goroutine count to rise until those connections are closed on the client side, at which point they drop again.

[ekimekim]

ok, good news and bad news.
good: timeout works. I leave 1000 connections hanging and the goroutines die after 90s
bad: it's broken websockets completely. when i try to actually use the scope ui i just get "re-connecting..."

[ekimekim]

Disregard the bad news above, it's an unrelated problem with my local setup since it's happening on master too (when accessing via /admin/scope). Connecting to scope.kube-system directly works fine with this PR.

[ekimekim]

ok, i'm now confident in my testing:

• works fine in normal browser usage
• websockets aren't affected, they remain open for > 90s
• manually testing with telnet - persistent connections remain open for > 90s
• manually testing with telnet - connections where I stop sending halfway through request headers are closed after 90s
  I think this is good to merge.