Still problems with windows buildbot

Simon Heimberg simohe at besonet.ch
Thu Feb 13 13:56:53 CST 2014


About windows builds, on http://hgbuildbot.kublai.com/waterfall the 
following is visible:
  * After windows buildbot slave has been rebootet, the failures on 
"default" went down to 4 [1].
  * Then a job for "stable" was run. There 13 failures happened [2].
  * Then a job for "default" was run. There were 15 failures [3] now.

My analysis of this results:
* The first failures on "default" [1] are real failures (no failures 
because of the environment).
* The failures on "stable"[2]  are some real failures plus those 
stopping and restarting hg serve.
   * All but 4 failures on the next build on "default" [3] are because 
of hg serve processes hanging around from the run on "stable" (blocking 
the tcp ports).
So:
  * No process from "default" has blocked a process for the next run.
   => Killing the processes on "default" seems to work. (The patch on 
mercurial.win32 works.)
  * Killing the processes on "stable" does not work, stop fails. Of 
course, this is not fixed there.
  * Because the number of tests is the same on "default" and "stable", 
the processes from test-x from "stable" block those from test-x on 
"default".

What can we do to improve this situation? Some suggestions:
  * Apply the patch on mercurial.win32 from "default" branch onto 
"stable" branch. (And reboot windows slave again.)
  * Set different values for "HGTEST_PORT" for "default" and "stable" 
runs. An offset of 3 should already improve the situation, because not 
all tests use the 3 ports reserved for them. (Maybe a good idea anyway, 
for having less influence. But the stable builds will still fail.)
  * ???


[1] 
http://hgbuildbot.kublai.com/builders/Windows%202008%20R2%20hg%20tests/builds/98
[2] 
http://hgbuildbot.kublai.com/builders/Windows%202008%20R2%20hg%20tests%20%28stable%29/builds/60
[3] 
http://hgbuildbot.kublai.com/builders/Windows%202008%20R2%20hg%20tests/builds/99


More information about the Mercurial-devel mailing list