Enable coredumps on arm, assert for assumptions about sysctl coredump settings

tools/bots/fletch.py

 #!/usr/bin/python
 
 # Copyright (c) 2014, the Fletch project authors.  Please see the AUTHORS file
 # for details. All rights reserved. Use of this source code is governed by a
 # BSD-style license that can be found in the LICENSE file.
 
 """
 Buildbot steps for fletch testing
 """
 
@@ skipping 204 matching lines @@
       Run([GSUTIL, 'cp', "%s/%s" % (GCS_BUCKET, tarball), tarball])
 
     with bot.BuildStep('Unpack build tarball'):
       Run(['tar', '-xjf', tarball])
 
     # Run tests on all necessary configurations.
     configurations = GetBuildConfigurations(system, [mode], [arch], [False])
     for snapshot_run in [True, False]:
       for configuration in configurations:
         if not ShouldSkipConfiguration(snapshot_run, configuration):
+          build_conf = configuration['build_conf']
           build_dir = configuration['build_dir']
 
           # Sanity check we got build artifacts which we expect.
           assert os.path.exists(os.path.join(build_dir, 'fletch-vm'))
 
           # TODO(kustermann): This is hackisch, but our current copying of the
           # dart binary makes this a requirement.
           dart_arm = 'third_party/bin/linux/dart-arm'
           destination = os.path.join(build_dir, 'dart')
           shutil.copyfile(dart_arm, destination)
           shutil.copymode(dart_arm, destination)
 
-          StepTest(
-            configuration['build_conf'],
-            configuration['mode'],
-            configuration['arch'],
-            clang=configuration['clang'],
-            asan=configuration['asan'],
-            snapshot_run=snapshot_run,
-            debug_log=debug_log,
-            configuration=configuration)
+          def run():
+            StepTest(
+              build_conf,
+              configuration['mode'],
+              configuration['arch'],
+              clang=configuration['clang'],
+              asan=configuration['asan'],
+              snapshot_run=snapshot_run,
+              debug_log=debug_log,
+              configuration=configuration)
+
+          RunWithCoreDumpArchiving(run, build_dir, build_conf)
   finally:
     if os.path.exists(tarball):
       os.remove(tarball)
 
     # We always clobber this to save disk on the arm board.
     bot.Clobber(force=True)
 
 
 #### Buildbot steps helper
 
@@ skipping 144 matching lines @@
   def __exit__(self, *_):
     resource.setrlimit(resource.RLIMIT_CORE, self._old_limits)
 
 class CoredumpArchiver(object):
   def __init__(self, bucket, build_dir, conf):
     self._bucket = bucket
     self._build_dir = build_dir
     self._conf = conf
 
   def __enter__(self):
-    # TODO(kustermann): Assert that the core pattern is correctly set (at least
-    # on linux)
-    pass
+    if utils.GuessOS() == 'linux':
+      core_pattern = open('/proc/sys/kernel/core_pattern').read()
+      core_pattern_uses_pid = open('/proc/sys/kernel/core_uses_pid').read()
+
+      assert core_pattern == 'core'

[kustermann, 2015/08/25 09:08:25]

I'm aware that the pattern could be made to include pid and more, but the Vms
are configured this way ATM.

[kustermann, 2015/08/25 11:14:48]

Also account for the newline in the file now.

+      assert core_pattern_uses_pid == '1'

[ricow1, 2015/08/25 09:10:09]

should we gracefully fail here, and just tell people that coredumps is not
available until the pattern is set?

[kustermann, 2015/08/25 11:14:48]

I thought about it as well, but
* users are not running this code, only bots are
* if a bot doesn't have it enabled, we really need to know and change that

That's why I decided against it so far.

[If we add a new VM it can easily happen it doesn't have the bit, and we'll
never notice, which is bad.]

 
   def __exit__(self, *_):
     coredumps = self._find_coredumps()
     if coredumps:
       print 'Archiving coredumps: %s' % ', '.join(coredumps)
       sys.stdout.flush()
       self._archive(os.path.join(self._build_dir, 'fletch-vm'), coredumps)
 
   def _find_coredumps(self):
     # TODO(kustermann): Make this work for mac as well.
@@ skipping 112 matching lines @@
     print '@@@STEP_FAILURE@@@'
   else:
     print '@@@STEP_WARNINGS@@@'
   sys.stdout.flush()
 
 if __name__ == '__main__':
   # If main raises an exception we will get a very useful error message with
   # traceback written to stderr. We therefore intentionally do not catch
   # exceptions.
   Main()