Merge lp:~michael.nelson/launchpad/fix-buildd-slave-test into lp:launchpad

[Michael Nelson]

Don't swallow all OSError exceptions during two_stage_kill.

Recently I've seen a failure during ec2 test (and hence didn't land my branch) which today also appeared on buildbot:

https://lpbuildbot.canonical.com/builders/lp/builds/703/steps/shell_7/logs/summary

It seems that the recently updated two_stage_kill is returning before the process has terminated, and terminates instead during the s.info() call, which triggers the connection reset.

The recent changes to two_stage_kill can be seen here:
https://code.edge.launchpad.net/~jml/launchpad/sftp-poppy/+merge/21627
although I don't see any behaviour change.

I can force the error as in the test failure by doing:
http://pastebin.ubuntu.com/399226/

Pre-imp. with jml:
http://pastebin.ubuntu.com/399240/

As I don't know what exception might be being raised, I'm not sure how it can be tested (other than the current test-suite still passing).

[Eleanor Berger (intellectronica)]

review: Approve (code)

=== modified file 'lib/canonical/launchpad/daemons/tachandler.py'
--- lib/canonical/launchpad/daemons/tachandler.py	2010-03-18 19:18:34 +0000
+++ lib/canonical/launchpad/daemons/tachandler.py	2010-03-22 11:33:25 +0000
@@ -57,9 +57,10 @@
                 return result
             time.sleep(poll_interval)
         except OSError, e:
-            # Raised if the process is gone by the time we try to get the
-            # return value.
-            return
+            if e.errno in (errno.ESRCH, errno.ECHILD):
+                # Raised if the process is gone by the time we try to get the
+                # return value.
+                return
 
     # The process is still around, so terminate it violently.
     try: