if args.noop:
print "Would have created sqs_queue with id: {}\nec2_args:".format(
run_id)
pprint(ec2_args)
ami = "ami-00000"
else:
run_summary, ami = launch_and_configure(ec2_args)
print
print "Summary:\n"
for run in run_summary:
print "{:<30} {:0>2.0f}:{:0>5.2f}".format(
run[0], run[1] / 60, run[1] % 60)
print "AMI: {}".format(ami)
if args.mongo_uri:
mongo_con.update_ami(ami)
mongo_con.update_deployment(ami)
finally:
print
if not args.no_cleanup and not args.noop:
if sqs_queue:
print "Cleaning up - Removing SQS queue - {}".format(run_id)
sqs.delete_queue(sqs_queue)
if instance_id:
print "Cleaning up - Terminating instance ID - {}".format(
instance_id)
# Check to make sure we have an instance id.
if instance_id:
ec2.terminate_instances(instance_ids=[instance_id])
__author__ = 'mhoyer'
import boto.sqs
from boto.sqs.message import Message
sqs = boto.sqs.connect_to_region("eu-west-1")
testqueue = sqs.create_queue('testqueue')
# write message
message = Message()
message.set_body("My first test message")
testqueue.write(message)
# read message
messages = testqueue.get_messages()
print messages[0].get_body()
sqs.delete_queue(testqueue)
message = 'Finished baking AMI {image_id} for {environment} {deployment} {play}.'.format(
image_id=ami,
environment=args.environment,
deployment=args.deployment,
play=args.play)
send_hipchat_message(message)
except Exception as e:
message = 'An error occurred building AMI for {environment} ' \
'{deployment} {play}. The Exception was {exception}'.format(
environment=args.environment,
deployment=args.deployment,
play=args.play,
exception=repr(e))
send_hipchat_message(message)
error_in_abbey_run = True
finally:
print
if not args.no_cleanup and not args.noop:
if sqs_queue:
print "Cleaning up - Removing SQS queue - {}".format(run_id)
sqs.delete_queue(sqs_queue)
if instance_id:
print "Cleaning up - Terminating instance ID - {}".format(
instance_id)
# Check to make sure we have an instance id.
if instance_id:
ec2.terminate_instances(instance_ids=[instance_id])
if error_in_abbey_run:
exit(1)
def wait_tasks():
queue_id = environ.get("SLURM_EC2_QUEUE_ID")
if queue_id is None:
print("SLURM_EC2_QUEUE_ID environment variable not set", file=stderr)
return 1
request_queue_name = "slurm-%s-request" % queue_id
response_queue_name = "slurm-%s-response" % queue_id
sqs = get_sqs()
request_queue = sqs.get_queue(request_queue_name)
response_queue = sqs.get_queue(response_queue_name)
times_empty = 0
while True:
msg = response_queue.read()
if msg is None:
# Are there pending requests?
attrs = request_queue.get_attributes()
in_flight = (int(attrs['ApproximateNumberOfMessages']) +
int(attrs['ApproximateNumberOfMessagesNotVisible']))
if in_flight == 0:
times_empty += 1
else:
times_empty = 0
# If we've not seen any responses and haven't found any unserved
# requests for MAX_TIMES_EMPTY polls, stop.
if times_empty >= MAX_TIMES_EMPTY:
break
if in_flight == 0:
attrs = response_queue.get_attributes()
long_poll_time = int(
attrs.get('ReceiveMessageWaitTimeSeconds', 0))
print("No tasks in flight... will wait %d more second(s)" %
((MAX_TIMES_EMPTY - times_empty) *
(SLEEP_TIME + long_poll_time)))
else:
print("%s task(s) in flight, but none are ready" %
(in_flight, ))
sleep(SLEEP_TIME)
continue
times_empty = 0
try:
# Decode the message as JSON
response = json_loads(msg.get_body())
id = response.get("id")
exit_code = response.get("exit_code")
log_dirs = [".", environ["HOME"], "/tmp", "/var/tmp"]
for log_dir in log_dirs:
filename = "%s/%s.json" % (log_dir, id)
try:
fd = open(filename, "w")
break
except IOError:
pass
else:
# Couldn't open a log file.
filename = "/dev/null"
fd = open(filename, "w")
fd.write(msg.get_body())
fd.close()
print("Task %s finished with exit code %s; logged to %s" %
(id, exit_code, filename))
except:
pass
msg.delete()
sqs.delete_queue(request_queue)
sqs.delete_queue(response_queue)
return 0
