def cluster_job(config, job_id, more=False):
setproctitle('%s %s' % (job_id, config.name))
proxy_port = 13000 + config.instance
hostname_no_instance = config.name.split(':')[0]
nice = compmake_config.cluster_nice #@UndefinedVariable
compmake_cmd = \
'nice -n %d compmake --hostname=%s --db=redis --redis_events --redis_host=localhost:%s --slave %s --save_progress=False\
make_single more=%s %s' % \
(nice, hostname_no_instance, proxy_port, get_namespace(), more, job_id)
redis_host = RedisInterface.host
redis_port = RedisInterface.port
if config.username:
connection_string = '%s@%s' % (config.username, config.host)
else:
connection_string = config.host
# TODO: make additional switches configurable
args = ['ssh', connection_string, '-X', '-R',
'%s:%s:%s' % (proxy_port, redis_host, redis_port),
'%s' % compmake_cmd]
if compmake_config.cluster_show_cmd: #@UndefinedVariable
print " ".join(args)
PIPE = subprocess.PIPE
p = subprocess.Popen(args, stdout=PIPE, stdin=PIPE, stderr=PIPE)
ret = p.wait()
if ret == RET_CODE_JOB_FAILED:
raise JobFailed('Job %s failed' % job_id)
if ret != 0:
raise HostFailed('Job %s: host failed: (line: "%s", ret=%s)' %
(job_id, " ".join(args), ret))
return ret
def make(job_id, more=False):
""" Makes a single job. Returns the user-object or raises JobFailed """
host = compmake_config.hostname #@UndefinedVariable
setproctitle(job_id)
# TODO: should we make sure we are up to date???
up, reason = up_to_date(job_id) #@UnusedVariable
cache = get_job_cache(job_id)
want_more = cache.state == Cache.MORE_REQUESTED
if up and not (more and want_more):
# print "%s is up to date" % job_id
assert is_job_userobject_available(job_id)
return get_job_userobject(job_id)
else:
# if up and (more and want_more): # XXX review the logic
# reason = 'want more'
# print "Making %s (%s)" % (job_id, reason)
computation = get_job(job_id)
assert(cache.state in [Cache.NOT_STARTED, Cache.IN_PROGRESS,
Cache.MORE_REQUESTED, Cache.DONE, Cache.FAILED])
if cache.state == Cache.NOT_STARTED:
previous_user_object = None
cache.state = Cache.IN_PROGRESS
if cache.state == Cache.FAILED:
previous_user_object = None
cache.state = Cache.IN_PROGRESS
elif cache.state == Cache.IN_PROGRESS:
if is_job_tmpobject_available(job_id):
previous_user_object = get_job_tmpobject(job_id)
else:
previous_user_object = None
elif cache.state == Cache.MORE_REQUESTED:
assert(is_job_userobject_available(job_id))
if is_job_tmpobject_available(job_id):
# resuming more computation
previous_user_object = get_job_tmpobject(job_id)
else:
# starting more computation
previous_user_object = get_job_userobject(job_id)
elif cache.state == Cache.DONE:
# If we are done, it means children have been updated
assert(not up)
previous_user_object = None
else:
assert(False)
# update state
cache.time_start = time()
cpu_start = clock()
set_job_cache(job_id, cache)
def progress_callback(stack):
publish('job-progress-plus', job_id=job_id, host=host, stack=stack)
init_progress_tracking(progress_callback)
num, total = 0, None
user_object = None
capture = OutputCapture(prefix=job_id,
echo_stdout=compmake_config.echo_stdout, #@UndefinedVariable
echo_stderr=compmake_config.echo_stderr) #@UndefinedVariable
try:
result = computation.compute(previous_user_object)
if type(result) == GeneratorType:
try:
while True:
next = result.next()
if isinstance(next, tuple):
if len(next) != 3:
raise CompmakeException('If computation yields a tuple, ' +
'should be a tuple with 3 elemnts.' +
'Got: %s' % str(next))
user_object, num, total = next
publish('job-progress', job_id=job_id, host=host,
done=None, progress=num, goal=total)
if compmake_config.save_progress: #@UndefinedVariable
set_job_tmpobject(job_id, user_object)
except StopIteration:
pass
else:
publish('job-progress', job_id=job_id, host='XXX',
done=1, progress=1, goal=1)
user_object = result
except KeyboardInterrupt:
# TODO: clear progress cache
# Save the current progress:
cache.iterations_in_progress = num
cache.iterations_goal = total
if user_object:
set_job_tmpobject(job_id, user_object)
set_job_cache(job_id, cache)
# clear progress cache
publish('job-interrupted', job_id=job_id, host=host)
raise JobInterrupted('Keyboard interrupt')
except Exception as e:
sio = StringIO()
print_exc(file=sio)
bt = sio.getvalue()
error("Job %s failed: %s" % (job_id, e))
error(bt)
mark_as_failed(job_id, e, bt)
# clear progress cache
publish('job-failed', job_id=job_id, host=host, reason=e)
raise JobFailed('Job %s failed: %s' % (job_id, e))
finally:
capture.deactivate()
# even if we send an error, let's save the output of the process
cache = get_job_cache(job_id)
cache.captured_stderr = capture.stderr_replacement.buffer.getvalue()
cache.captured_stdout = capture.stdout_replacement.buffer.getvalue()
set_job_cache(job_id, cache)
set_job_userobject(job_id, user_object)
if is_job_tmpobject_available(job_id):
# We only have one with yield
delete_job_tmpobject(job_id)
cache.state = Cache.DONE
cache.timestamp = time()
walltime = cache.timestamp - cache.time_start
cputime = clock() - cpu_start
# FIXME walltime/cputime not precise (especially for "more" computation)
cache.walltime_used = walltime
cache.cputime_used = cputime
cache.done_iterations = num # XXX not true
cache.host = compmake_config.hostname #@UndefinedVariable
set_job_cache(job_id, cache)
publish('job-succeeded', job_id=job_id, host=host)
# TODO: clear these records in other place
return user_object
def main():
setproctitle('compmake')
parser = OptionParser(version=version)
parser.add_option("--slave", action="store_true",
default=False, dest="slave",
help="[internal] Runs compmake in slave mode.")
parser.add_option("--redis_events", action="store_true",
default=False, dest="redis_events",
help="[internal] Relays events using Redis.")
config_populate_optparser(parser)
(options, args) = parser.parse_args()
initialize_backend()
# We load plugins after we parsed the configuration
from compmake import plugins #@UnusedImport
if options.redis_events:
if not compmake_config.db == 'redis': #@UndefinedVariable
error('Cannot use redis_events without redis.')
sys.exit(-2)
from compmake.storage.redisdb import RedisInterface
# register an handler that will capture all events
def handler(event):
RedisInterface.events_push(event)
remove_all_handlers()
register_handler("*", handler)
if not options.slave:
# XXX make sure this is the default
set_compmake_status(compmake_status_interactive)
# TODO: add command namespace
# TODO: add command "load"
if not args:
user_error('I expect at least one parameter (module name)')
sys.exit(-2)
module_name = args[0]
args = args[1:]
if module_name.endswith('.py') or (module_name.find('/') > 0):
warning('You passed a string "%s" which looks like a filename.' %
module_name)
module_name = module_name.replace('/', '.')
module_name = module_name.replace('.py', '')
warning('However, I need a module name. I will try with "%s".' %
module_name)
set_namespace(module_name)
compmake.is_it_time = True
try:
__import__(module_name)
except Exception as e:
error('Error while trying to import module "%s": %s' %
(module_name, e))
traceback.print_exc(file=sys.stderr)
sys.exit(-5)
# TODO: BUG: XXX: remove old jobs those in defined_this_section
else:
set_compmake_status(compmake_status_slave)
if not args:
user_error('I expect at least one parameter (namespace name)')
sys.exit(-2)
module_name = args.pop(0)
set_namespace(module_name)
if args:
try:
# XXX is this redudant?
# compmake_config.interactive = False
retcode = interpret_commands(args)
# print "Exiting with retcode %s" % retcode
sys.exit(retcode)
except UserError as e:
user_error(e)
sys.exit(-6)
else:
retcode = interactive_console()
sys.exit(retcode)