Bug 55579 – Websockets tests sometimes time out on apple-windows-5

• Component: Page Loading
• Status: RESOLVED CONFIGURATION CHANGED

[Adam Roben (:aroben)]

Websockets tests sometimes time out on Windows. Here are some examples:

http://build.webkit.org/builders/Windows%207%20Release%20%28Tests%29/builds/9943
http://build.webkit.org/builders/Windows%207%20Release%20%28Tests%29/builds/9944 (timed out)
http://build.webkit.org/builders/Windows%207%20Release%20%28Tests%29/builds/9945

http://build.webkit.org/builders/Windows%207%20Release%20%28Tests%29/builds/9922
http://build.webkit.org/builders/Windows%207%20Release%20%28Tests%29/builds/9923 (timed out)
http://build.webkit.org/builders/Windows%207%20Release%20%28Tests%29/builds/9926

In both cases, one test timed out while waiting for notifyDone to be called, and then another test failed soon after.

I don't think this started happening recently.

[Adam Roben (:aroben)]

<rdar://problem/9075090>

[Yuta Kitamura]

This timeouts seem to happen only on "apple-windows-5" bot. "apple-windows-6" is fine.

http://build.webkit.org/buildslaves/apple-windows-5
http://build.webkit.org/buildslaves/apple-windows-6

Sounds strange...

[Adam Roben (:aroben)]

Maybe Lucas can tell us what's different about apple-windows-5 vs. apple-windows-6. I seem to recall they have different versions of Python installed.

[Adam Roben (:aroben)]

apple-windows-5 has Python 2.6.5, while apple-windows-6 has Python 2.5.x.

[Adam Roben (:aroben)]

Lucas upgraded apple-windows-6 to Python 2.6.5 today. It'll be interesting to see if these failures start happening on that machine.

[Adam Roben (:aroben)]

This happened again: http://build.webkit.org/builders/Windows%207%20Release%20%28Tests%29/builds/10517

[Adam Roben (:aroben)]

(The Acid3 failure in that last URL is bug 55578.)

[Adam Roben (:aroben)]

Here's another one: http://build.webkit.org/results/Windows%207%20Release%20(Tests)/r81385%20(10528)/http/tests/websocket/tests/handshake-error-pretty-diff.html

[Adam Roben (:aroben)]

And another:

http://build.webkit.org/results/Windows%207%20Release%20(Tests)/r81571%20(10593)/http/tests/websocket/tests/url-no-trailing-slash-pretty-diff.html

[Adam Roben (:aroben)]

In fact, pretty much all of the recent flakiness on Windows 7 Release (Tests) has been due to websocket timeouts.

apple-windows-5 and apple-windows-6 both have Python 2.6.5 installed now, and the flakiness is still only happening on apple-windows-5. So there must be some other reason it's happening.

[Adam Roben (:aroben)]

http/tests/websocket/tests/multiple-connections.html timed out on SnowLeopard Intel Release (WebKit2 Tests); perhaps it's related?

http://build.webkit.org/results/SnowLeopard%20Intel%20Release%20%28WebKit2%20Tests%29/r81509%20(9838)/http/tests/websocket/tests/multiple-connections-pretty-diff.html

[Jessie Berlin]

After talking with Alexey on IRC, it is probably best if we just disable the http websocket tests on Windows for now until the flakiness can be resolved.

He also noted that the issues seem to stem from incorrect interaction with the loader thread.

[Jessie Berlin]

Skipped the websocket tests on Windows in http://trac.webkit.org/changeset/81976

[WebKit Review Bot]

http://trac.webkit.org/changeset/81976 might have broken Qt Linux Release

[Adam Roben (:aroben)]

See also bug 56625 and bug 57048, which are likely caused by the same underlying issue.

[Takeshi Yoshino]

FYI,

All WebSocket tests are still skipped for Windows platform.

http://trac.webkit.org/browser/trunk/LayoutTests/platform/win/TestExpectations
(Skipped file has been merged into TestExpectations)

[Takeshi Yoshino]

Summarizing Windows LayoutTests related bugs to this entry by adding "Depends on".

[Brent Fulgham]

Per -- can you see if we can unskip these on Windows?

[Per Arne Vollan]

Unskipped in <https://trac.webkit.org/changeset/201867>

[Ahmad Saleem]

Applicable for win-cairo or it was 'AppleWin' only and we can close this?

[Fujii Hironori]

I don't observe flaky timeout for WinCairo.