Convert gnt-cluster

[ganeti-local] / qa / qa_daemon.py

diff --git a/qa/qa_daemon.py b/qa/qa_daemon.py
index 1fb3951..4817bfe 100644 (file)
--- a/qa/qa_daemon.py
+++ b/qa/qa_daemon.py
@@ -1,3 +1,6 @@
+#
+#
+
 # Copyright (C) 2007 Google Inc.
 #
 # This program is free software; you can redistribute it and/or modify
@@ -21,7 +24,6 @@
 """
 
 import time
-import subprocess
 
 from ganeti import utils
 from ganeti import constants
@@ -33,23 +35,6 @@ import qa_error
 from qa_utils import AssertEqual, StartSSH
 
 
-def _ResolveInstanceName(instance):
-  """Gets the full Xen name of an instance.
-
-  """
-  master = qa_config.GetMasterNode()
-
-  info_cmd = utils.ShellQuoteArgs(['gnt-instance', 'info', instance['name']])
-  sed_cmd = utils.ShellQuoteArgs(['sed', '-n', '-e', 's/^Instance name: *//p'])
-
-  cmd = '%s | %s' % (info_cmd, sed_cmd)
-  ssh_cmd = qa_utils.GetSSHCommand(master['primary'], cmd)
-  p = subprocess.Popen(ssh_cmd, shell=False, stdout=subprocess.PIPE)
-  AssertEqual(p.wait(), 0)
-
-  return p.stdout.read().strip()
-
-
 def _InstanceRunning(node, name):
   """Checks whether an instance is running.
 
@@ -72,7 +57,7 @@ def _XmShutdownInstance(node, name):
   master = qa_config.GetMasterNode()
 
   cmd = ['xm', 'shutdown', name]
-  AssertEqual(StartSSH(master['primary'],
+  AssertEqual(StartSSH(node['primary'],
                        utils.ShellQuoteArgs(cmd)).wait(), 0)
 
   # Wait up to a minute
@@ -85,38 +70,55 @@ def _XmShutdownInstance(node, name):
     raise qa_error.Error("xm shutdown failed")
 
 
-def _ResetWatcherDaemon(node):
+def _ResetWatcherDaemon():
   """Removes the watcher daemon's state file.
 
   Args:
     node: Node to be reset
   """
+  master = qa_config.GetMasterNode()
+
   cmd = ['rm', '-f', constants.WATCHER_STATEFILE]
-  AssertEqual(StartSSH(node['primary'],
+  AssertEqual(StartSSH(master['primary'],
                        utils.ShellQuoteArgs(cmd)).wait(), 0)
 
 
+def _RunWatcherDaemon():
+  """Runs the ganeti-watcher daemon on the master node.
+
+  """
+  master = qa_config.GetMasterNode()
+
+  cmd = ['ganeti-watcher', '-d']
+  AssertEqual(StartSSH(master['primary'],
+                       utils.ShellQuoteArgs(cmd)).wait(), 0)
+
+
+def PrintCronWarning():
+  """Shows a warning about the cron job.
+
+  """
+  msg = ("For the following tests it's recommended to turn off the"
+         " ganeti-watcher cronjob.")
+  print
+  print qa_utils.FormatWarning(msg)
+
+
 def TestInstanceAutomaticRestart(node, instance):
   """Test automatic restart of instance by ganeti-watcher.
 
-  Note: takes up to 6 minutes to complete.
   """
   master = qa_config.GetMasterNode()
-  inst_name = _ResolveInstanceName(instance)
+  inst_name = qa_utils.ResolveInstanceName(instance)
 
-  _ResetWatcherDaemon(node)
+  _ResetWatcherDaemon()
   _XmShutdownInstance(node, inst_name)
 
-  # Give it a bit more than five minutes to start again
-  restart_at = time.time() + 330
+  _RunWatcherDaemon()
+  time.sleep(5)
 
-  # Wait until it's running again
-  while time.time() <= restart_at:
-    if _InstanceRunning(node, inst_name):
-      break
-    time.sleep(15)
-  else:
-    raise qa_error.Error("Daemon didn't restart instance in time")
+  if not _InstanceRunning(node, inst_name):
+    raise qa_error.Error("Daemon didn't restart instance")
 
   cmd = ['gnt-instance', 'info', inst_name]
   AssertEqual(StartSSH(master['primary'],
@@ -126,28 +128,23 @@ def TestInstanceAutomaticRestart(node, instance):
 def TestInstanceConsecutiveFailures(node, instance):
   """Test five consecutive instance failures.
 
-  Note: takes at least 35 minutes to complete.
   """
   master = qa_config.GetMasterNode()
-  inst_name = _ResolveInstanceName(instance)
-
-  _ResetWatcherDaemon(node)
-  _XmShutdownInstance(node, inst_name)
+  inst_name = qa_utils.ResolveInstanceName(instance)
 
-  # Do shutdowns for 30 minutes
-  finished_at = time.time() + (35 * 60)
+  _ResetWatcherDaemon()
 
-  while time.time() <= finished_at:
-    if _InstanceRunning(node, inst_name):
-      _XmShutdownInstance(node, inst_name)
-    time.sleep(30)
+  for should_start in ([True] * 5) + [False]:
+    _XmShutdownInstance(node, inst_name)
+    _RunWatcherDaemon()
+    time.sleep(5)
 
-  # Check for some time whether the instance doesn't start again
-  check_until = time.time() + 330
-  while time.time() <= check_until:
-    if _InstanceRunning(node, inst_name):
-      raise qa_error.Error("Instance started when it shouldn't")
-    time.sleep(30)
+    if bool(_InstanceRunning(node, inst_name)) != should_start:
+      if should_start:
+        msg = "Instance not started when it should"
+      else:
+        msg = "Instance started when it shouldn't"
+      raise qa_error.Error(msg)
 
   cmd = ['gnt-instance', 'info', inst_name]
   AssertEqual(StartSSH(master['primary'],