def main():
config.init(app.common_opts, sys.argv[1:])
application = app.setup_app()
host = CONF.bind_host
port = CONF.bind_port
workers = CONF.api_workers
if workers < 1:
LOG.warning(_LW("Wrong worker number, worker = %(workers)s"), workers)
workers = 1
LOG.info(_LI("Nova_APIGW on http://%(host)s:%(port)s with %(workers)s"), {
'host': host,
'port': port,
'workers': workers
})
service = wsgi.Server(CONF, 'Trio2o Nova_APIGW', application, host, port)
restapp.serve(service, CONF, workers)
LOG.info(_LI("Configuration:"))
CONF.log_opt_values(LOG, std_logging.INFO)
restapp.wait()
def wrapped(*args, **kwargs):
while True:
try:
return f(*args, **kwargs)
except db_exc.DBDeadlock:
LOG.warning(_LW("Deadlock detected when running "
"'%(func_name)s': Retrying..."),
dict(func_name=f.__name__))
# Retry!
time.sleep(0.5)
continue
def main():
config.init(xservice.common_opts, sys.argv[1:])
host = CONF.host
workers = CONF.workers
if workers < 1:
LOG.warning(_LW("Wrong worker number, worker = %(workers)s"), workers)
workers = 1
LOG.info(_LI("XJob Server on http://%(host)s with %(workers)s"), {
'host': host,
'workers': workers
})
xservice.serve(xservice.create_service(), workers)
LOG.info(_LI("Configuration:"))
CONF.log_opt_values(LOG, std_logging.INFO)
xservice.wait()
def quota_reserve(context, resources, quotas, deltas, expire,
until_refresh, max_age, project_id=None):
elevated = context.elevated()
with context.session.begin():
if project_id is None:
project_id = context.project_id
# Get the current usages
usages = _get_quota_usages(context, context.session, project_id)
# Handle usage refresh
refresh = False
work = set(deltas.keys())
while work:
resource = work.pop()
# Do we need to refresh the usage?
if resource not in usages:
usages[resource] = _quota_usage_create(elevated,
project_id,
resource,
0, 0,
until_refresh or None,
session=context.session)
refresh = True
elif usages[resource].in_use < 0:
# Negative in_use count indicates a desync, so try to
# heal from that...
refresh = True
elif usages[resource].until_refresh is not None:
usages[resource].until_refresh -= 1
if usages[resource].until_refresh <= 0:
refresh = True
elif max_age and usages[resource].updated_at is not None and (
(usages[resource].updated_at -
timeutils.utcnow()).seconds >= max_age):
refresh = True
if refresh:
# no actural usage refresh here
# refresh from the bottom pod
usages[resource].until_refresh = until_refresh or None
# Because more than one resource may be refreshed
# by the call to the sync routine, and we don't
# want to double-sync, we make sure all refreshed
# resources are dropped from the work set.
work.discard(resource)
# NOTE(Vek): We make the assumption that the sync
# routine actually refreshes the
# resources that it is the sync routine
# for. We don't check, because this is
# a best-effort mechanism.
# Check for deltas that would go negative
unders = [r for r, delta in deltas.items()
if delta < 0 and delta + usages[r].in_use < 0]
# Now, let's check the quotas
# NOTE(Vek): We're only concerned about positive increments.
# If a project has gone over quota, we want them to
# be able to reduce their usage without any
# problems.
overs = [r for r, delta in deltas.items()
if quotas[r] >= 0 and delta >= 0 and
quotas[r] < delta + usages[r].in_use + usages[r].reserved]
# NOTE(Vek): The quota check needs to be in the transaction,
# but the transaction doesn't fail just because
# we're over quota, so the OverQuota raise is
# outside the transaction. If we did the raise
# here, our usage updates would be discarded, but
# they're not invalidated by being over-quota.
# Create the reservations
if not overs:
reservations = []
for resource, delta in deltas.items():
reservation = _reservation_create(elevated,
str(uuid.uuid4()),
usages[resource],
project_id,
resource, delta, expire,
session=context.session)
reservations.append(reservation.uuid)
# Also update the reserved quantity
# NOTE(Vek): Again, we are only concerned here about
# positive increments. Here, though, we're
# worried about the following scenario:
#
# 1) User initiates resize down.
# 2) User allocates a new instance.
# 3) Resize down fails or is reverted.
# 4) User is now over quota.
#
# To prevent this, we only update the
# reserved value if the delta is positive.
if delta > 0:
usages[resource].reserved += delta
if unders:
LOG.warning(_LW("Change will make usage less than 0 for the following "
"resources: %s"), unders)
if overs:
usages = {k: dict(in_use=v['in_use'], reserved=v['reserved'])
for k, v in usages.items()}
raise exceptions.OverQuota(overs=sorted(overs), quotas=quotas,
usages=usages)
return reservations
def handle_args(*args, **kwargs):
if IN_TEST:
# NOTE(zhiyuan) job mechanism will cause some unpredictable
# result in unit test so we would like to bypass it. However
# we have problem mocking a decorator which decorates member
# functions, that's why we use this label, not an elegant
# way though.
func(*args, **kwargs)
return
ctx = args[1]
payload = kwargs['payload']
resource_id = payload[job_type]
db_api.new_job(ctx, job_type, resource_id)
start_time = datetime.datetime.now()
while True:
current_time = datetime.datetime.now()
delta = current_time - start_time
if delta.seconds >= CONF.worker_handle_timeout:
# quit when this handle is running for a long time
break
time_new = db_api.get_latest_timestamp(ctx, constants.JS_New,
job_type, resource_id)
time_success = db_api.get_latest_timestamp(
ctx, constants.JS_Success, job_type, resource_id)
if time_success and time_success >= time_new:
break
job = db_api.register_job(ctx, job_type, resource_id)
if not job:
# fail to obtain the lock, let other worker handle the job
running_job = db_api.get_running_job(
ctx, job_type, resource_id)
if not running_job:
# there are two reasons that running_job is None. one
# is that the running job has just been finished, the
# other is that all workers fail to register the job
# due to deadlock exception. so we sleep and try again
eventlet.sleep(CONF.worker_sleep_time)
continue
job_time = running_job['timestamp']
current_time = datetime.datetime.now()
delta = current_time - job_time
if delta.seconds > CONF.job_run_expire:
# previous running job expires, we set its status to
# fail and try again to obtain the lock
db_api.finish_job(ctx, running_job['id'], False,
time_new)
LOG.warning(
_LW('Job %(job)s of type %(job_type)s for '
'resource %(resource)s expires, set '
'its state to Fail'), {
'job': running_job['id'],
'job_type': job_type,
'resource': resource_id
})
eventlet.sleep(CONF.worker_sleep_time)
continue
else:
# previous running job is still valid, we just leave
# the job to the worker who holds the lock
break
# successfully obtain the lock, start to execute handler
try:
func(*args, **kwargs)
except Exception:
db_api.finish_job(ctx, job['id'], False, time_new)
LOG.error(
_LE('Job %(job)s of type %(job_type)s for '
'resource %(resource)s fails'), {
'job': job['id'],
'job_type': job_type,
'resource': resource_id
})
break
db_api.finish_job(ctx, job['id'], True, time_new)
eventlet.sleep(CONF.worker_sleep_time)