Bug 71076 – Old pywebsocket servers can cause NRWT to fail

• Component: Tools / Tests
• Status: RESOLVED FIXED

[Adam Roben (:aroben)]

Old pywebsocket servers can cause NRWT to fail. Here's an example:

http://build.webkit.org/builders/SnowLeopard%20Intel%20Leaks/builds/19784/steps/layout-test/logs/stdio

webkitpy.layout_tests.servers.http_server_base.ServerError: Port 8880 is already in use.

On that bot we see:

buildbot 14833   0.0  0.3  2532528  22040   ??  S    18Oct11  21:09.62 /usr/bin/python -u /Volumes/Data/WebKit-BuildSlave/snowleopard-intel-leaks/build/Tools/Scripts/webkitpy/thirdparty/autoinstalled/pywebsocket/mod_pywebsocket/standalone.py --server-host 127.0.0.1 --port 8880 --document-root /Volumes/Data/WebKit-BuildSlave/snowleopard-intel-leaks/build/LayoutTests/http/tests --scan-dir /Volumes/Data/WebKit-BuildSlave/snowleopard-intel-leaks/build/LayoutTests/http/tests/websocket/tests --cgi-paths /websocket/tests --log-file /Volumes/Data/WebKit-BuildSlave/snowleopard-intel-leaks/build/layout-test-results/pywebsocket.ws.log-18Oct2011-132521-err.txt --websock-handlers-map-file /Volumes/Data/WebKit-BuildSlave/snowleopard-intel-leaks/build/LayoutTests/http/tests/websocket/tests/handler_map.txt

That process is 9 days old!

ORWT doesn't fail as spectacularly in this case. It keeps trudging on:

http://build.webkit.org/builders/SnowLeopard%20Intel%20Leaks/builds/19781/steps/layout-test/logs/stdio

I think there are a few ways we can improve this:

1) Make this a non-fatal error in NRWT like it was in ORWT
2) Make kill-old-processes kill pywebsocket servers

[Ryosuke Niwa]

Can we kill that process automatically in http://trac.webkit.org/browser/trunk/Tools/BuildSlaveSupport/kill-old-processes ?

[Eric Seidel (no email)]

I'm open to making it non-fatal.  What does that mean though?  The tests just fail instead of NRWT failing?

[Adam Roben (:aroben)]

(In reply to comment #1)
> Can we kill that process automatically in http://trac.webkit.org/browser/trunk/Tools/BuildSlaveSupport/kill-old-processes ?

Yes, that would be good to do.

(In reply to comment #2)
> I'm open to making it non-fatal.  What does that mean though?  The tests just fail instead of NRWT failing?

It looks like we were getting away with using the old server. Obviously that could start to cause mysterious problems if we update the server and tests in a way that is incompatible with the old server.

[Dirk Pranke]

I think this has been resolved; we should be killing any old processes before we start the run now. Please re-open if we still see this (or file a new bug).