def run(self):
workers = []
LOG.info('Starting Queues')
libvirt = util.get_libvirt()
conn = libvirt.open(None)
present_cpus, _, _ = conn.getCPUMap()
while True:
try:
for w in copy.copy(workers):
if not w.is_alive():
w.join(1)
workers.remove(w)
if len(workers) < present_cpus / 2:
jobname, workitem = db.dequeue(config.NODE_NAME)
else:
workitem = None
if not workitem:
time.sleep(0.2)
continue
p = multiprocessing.Process(
target=handle, args=(jobname, workitem,),
name='%s-worker' % daemon.process_name('queues'))
p.start()
workers.append(p)
except Exception as e:
util.ignore_exception(daemon.process_name('queues'), e)
def _process_network_node_workitems(self):
jobname, workitem = db.dequeue('networknode')
try:
if not workitem:
time.sleep(0.2)
return
log_ctx = LOG.withField('workitem', workitem)
if not NetworkTask.__subclasscheck__(type(workitem)):
raise exceptions.UnknownTaskException(
'Network workitem was not decoded: %s' % workitem)
n = net.from_db(workitem.network_uuid())
if not n:
log_ctx.withNetwork(workitem.network_uuid()).warning(
'Received work item for non-existent network')
return
# NOTE(mikal): there's really nothing stopping us from processing a bunch
# of these jobs in parallel with a pool of workers, but I am not sure its
# worth the complexity right now. Are we really going to be changing
# networks that much?
if isinstance(workitem, DeployNetworkTask):
try:
n.create()
n.ensure_mesh()
db.add_event('network', workitem.network_uuid(),
'network node', 'deploy', None, None)
except exceptions.DeadNetwork as e:
log_ctx.withField('exception', e).warning(
'DeployNetworkTask on dead network')
elif isinstance(workitem, UpdateDHCPNetworkTask):
try:
n.create()
n.ensure_mesh()
n.update_dhcp()
db.add_event('network', workitem.network_uuid(),
'network node', 'update dhcp', None, None)
except exceptions.DeadNetwork as e:
log_ctx.withField('exception', e).warning(
'UpdateDHCPNetworkTask on dead network')
elif isinstance(workitem, RemoveDHCPNetworkTask):
n.remove_dhcp()
db.add_event('network', workitem.network_uuid(),
'network node', 'remove dhcp', None, None)
finally:
if jobname:
db.resolve('networknode', jobname)
def run(self):
logutil.info(None, 'Starting')
gauges = {
'updated_at': Gauge('updated_at', 'The last time metrics were updated')
}
last_metrics = 0
def update_metrics():
global last_metrics
stats = _get_stats()
for metric in stats:
if metric not in gauges:
gauges[metric] = Gauge(metric, '')
gauges[metric].set(stats[metric])
db.update_metrics_bulk(stats)
logutil.info(None, 'Updated metrics')
gauges['updated_at'].set_to_current_time()
while True:
try:
jobname, _ = db.dequeue(
'%s-metrics' % config.parsed.get('NODE_NAME'))
if jobname:
if time.time() - last_metrics > 2:
update_metrics()
last_metrics = time.time()
db.resolve('%s-metrics' % config.parsed.get('NODE_NAME'),
jobname)
else:
time.sleep(0.2)
if time.time() - last_metrics > config.parsed.get('SCHEDULER_CACHE_TIMEOUT'):
update_metrics()
last_metrics = time.time()
except Exception as e:
util.ignore_exception('resource statistics', e)