scheduler release hosts
-make scheduler really release hosts after verify failure. We decided this is the desired behavior, and it fixes a bug with metahosts failing verify.
-make frontend only reset hosts to Ready if they are "dead" (status Dead or Repair Failed). Otherwise hosts can get reset to ready while being repaired, and then two autoservs will run on the machine at once.
Signed-off-by: Steve Howard <showard@google.com>
git-svn-id: http://test.kernel.org/svn/autotest/trunk@1283 592f7852-d20e-0410-864c-8624ca9c26a4
diff --git a/scheduler/monitor_db b/scheduler/monitor_db
index e48052e..61e20cb 100755
--- a/scheduler/monitor_db
+++ b/scheduler/monitor_db
@@ -524,8 +524,7 @@
failure_tasks = self.get_failure_tasks()
- AgentTask.__init__(self, cmd, failure_tasks=failure_tasks,
- clear_queue_on_failure=False)
+ AgentTask.__init__(self, cmd, failure_tasks=failure_tasks)
def get_failure_tasks(self):
@@ -538,6 +537,8 @@
print "starting verify on %s" % (self.host.hostname)
if self.queue_entry:
self.queue_entry.set_status('Verifying')
+ self.queue_entry.clear_results_dir(
+ self.queue_entry.verify_results_dir())
self.host.set_status('Verifying')
@@ -559,11 +560,8 @@
def on_failure(self):
self.host.set_status('Failed Verify')
- # don't use queue_entry.requeue() here, because we don't want
- # a meta-host entry to release its host yet - that should only
- # happen after reverify fails
if self.queue_entry:
- self.queue_entry.set_status('Queued')
+ self.queue_entry.requeue()
def move_results(self):
@@ -593,31 +591,38 @@
class ReverifyTask(VerifyTask):
def __init__(self, queue_entry=None, host=None):
+ self.fail_queue_entry = None
if queue_entry:
- VerifyTask.__init__(self, queue_entry=queue_entry)
- else:
- VerifyTask.__init__(self, host=host)
- self.clear_queue_on_failure = True
+ if not host:
+ host = queue_entry.host
+ if not queue_entry.meta_host:
+ self.fail_queue_entry = queue_entry
+ # always construct VerifyTask without the queue_entry - we don't
+ # want to touch the queue entry unless we fail, in which case we
+ # just fail it (and only if it's a non-metahost)
+ VerifyTask.__init__(self, host=host)
def get_failure_tasks(self):
return []
- def prolog(self):
- VerifyTask.prolog(self)
- if self.queue_entry:
- self.queue_entry.clear_results_dir(
- self.queue_entry.verify_results_dir())
-
-
def on_failure(self):
self.host.set_status('Repair Failed')
- if self.queue_entry:
- self.queue_entry.handle_host_failure()
+ if self.fail_queue_entry:
+ self.fail_queue_entry.handle_host_failure()
-class VerifySynchronousMixin(object):
+class VerifySynchronousTask(VerifyTask):
+ def __init__(self, queue_entry):
+ VerifyTask.__init__(self, queue_entry = queue_entry)
+
+
+ def on_success(self):
+ VerifyTask.on_success(self)
+ self.on_pending()
+
+
def on_pending(self):
if self.queue_entry.job.num_complete() > 0:
# some other entry failed verify, and we've
@@ -631,31 +636,6 @@
self.agent.dispatcher.add_agent(agent)
-class VerifySynchronousTask(VerifyTask, VerifySynchronousMixin):
- def __init__(self, queue_entry):
- VerifyTask.__init__(self, queue_entry = queue_entry)
-
-
- def get_failure_tasks(self):
- return [RepairTask(self.host),
- ReverifySynchronousTask(self.queue_entry)]
-
-
- def on_success(self):
- VerifyTask.on_success(self)
- self.on_pending()
-
-
-class ReverifySynchronousTask(ReverifyTask, VerifySynchronousMixin):
- def __init__(self, queue_entry):
- ReverifyTask.__init__(self, queue_entry = queue_entry)
-
-
- def on_success(self):
- ReverifyTask.on_success(self)
- self.on_pending()
-
-
class QueueTask(AgentTask):
def __init__(self, job, queue_entries, cmd):
AgentTask.__init__(self, cmd)
@@ -1014,12 +994,10 @@
Called when this queue entry's host has failed verification and
repair.
"""
- if self.meta_host:
- self.requeue()
- else:
- self.set_status('Failed')
- if self.job.is_synchronous():
- self.job.stop_all_entries()
+ assert not self.meta_host
+ self.set_status('Failed')
+ if self.job.is_synchronous():
+ self.job.stop_all_entries()
def clear_results_dir(self, results_dir=None):