def start(parallel,
items,
config,
dirs=None,
name=None,
multiplier=1,
max_multicore=None):
"""Start a parallel cluster or machines to be used for running remote functions.
Returns a function used to process, in parallel items with a given function.
Allows sharing of a single cluster across multiple functions with
identical resource requirements. Uses local execution for non-distributed
clusters or completed jobs.
A checkpoint directory keeps track of finished tasks, avoiding spinning up clusters
for sections that have been previous processed.
multiplier - Number of expected jobs per initial input item. Used to avoid underscheduling
cores when an item is split during processing.
max_multicore -- The maximum number of cores to use for each process. Can be used
to process less multicore usage when jobs run faster on more single cores.
"""
if name:
checkpoint_dir = utils.safe_makedir(
os.path.join(dirs["work"], "checkpoints_parallel"))
checkpoint_file = os.path.join(checkpoint_dir, "%s.done" % name)
else:
checkpoint_file = None
sysinfo = system.get_info(dirs, parallel)
items = [x for x in items if x is not None] if items else []
parallel = resources.calculate(
parallel,
items,
sysinfo,
config,
multiplier=multiplier,
max_multicore=int(max_multicore or sysinfo.get("cores", 1)))
try:
if checkpoint_file and os.path.exists(checkpoint_file):
logger.info("run local -- checkpoint passed: %s" % name)
parallel["cores_per_job"] = 1
parallel["num_jobs"] = 1
yield multi.runner(parallel, config)
elif parallel["type"] == "ipython":
with ipython.create(parallel, dirs, config) as view:
yield ipython.runner(view, parallel, dirs, config)
elif parallel["type"] == "clusterk":
with clusterk.create(parallel) as queue:
yield clusterk.runner(queue, parallel)
else:
yield multi.runner(parallel, config)
except:
raise
else:
for x in ["cores_per_job", "num_jobs", "mem"]:
parallel.pop(x, None)
if checkpoint_file:
with open(checkpoint_file, "w") as out_handle:
out_handle.write("done\n")
def send_job(fn, data, args, resources=None):
"""decide if send jobs with ipython or run locally"""
utils.safe_makedir("checkpoint")
_check_items(data)
res = []
dirs = {'work': os.path.abspath(os.getcwd())}
config = data[0][0]['config']
if not resources:
resources = config_default
step = resources['name']
if 'mem' not in resources or 'cores' not in resources:
raise ValueError("resources without mem or cores keys: %s" % resources)
par = _calculate_resources(data, args, resources)
# args.memory_per_job = resources['mem']
# args.cores_per_job = resources['cores']
# log.setup_log(args)
log.logger.debug("doing %s" % step)
if par['type'] == "ipython" and not is_done(step):
with create(par, dirs, config) as view:
for sample in data:
res.append(view.apply_async(fn, sample[0], args))
res = wait_until_complete(res)
flag_done(step)
return res
for sample in data:
res.append([fn(sample[0], args)])
return res
def start(parallel, items, config, dirs=None, name=None, multiplier=1,
max_multicore=None):
"""Start a parallel cluster or machines to be used for running remote
functions.
Returns a function used to process, in parallel items with a given function.
Allows sharing of a single cluster across multiple functions with
identical resource requirements. Uses local execution for non-distributed
clusters or completed jobs.
A checkpoint directory keeps track of finished tasks, avoiding spinning up
clusters for sections that have been previous processed.
multiplier - Number of expected jobs per initial input item. Used to avoid
underscheduling cores when an item is split during processing.
max_multicore -- The maximum number of cores to use for each process. Can be
used to process less multicore usage when jobs run faster on more single
cores.
"""
if name:
checkpoint_dir = utils.safe_makedir(os.path.join(dirs["work"],
"checkpoints_parallel"))
checkpoint_file = os.path.join(checkpoint_dir, "%s.done" % name)
else:
checkpoint_file = None
sysinfo = system.get_info(dirs, parallel)
items = [x for x in items if x is not None] if items else []
max_multicore = int(max_multicore or sysinfo.get("cores", 1))
parallel = resources.calculate(parallel, items, sysinfo, config,
multiplier=multiplier,
max_multicore=max_multicore)
try:
view = None
if checkpoint_file and os.path.exists(checkpoint_file):
logger.info("run local -- checkpoint passed: %s" % name)
parallel["cores_per_job"] = 1
parallel["num_jobs"] = 1
parallel["checkpointed"] = True
yield multi.runner(parallel, config)
elif parallel["type"] == "ipython":
with ipython.create(parallel, dirs, config) as view:
yield ipython.runner(view, parallel, dirs, config)
elif parallel["type"] == "clusterk":
with clusterk.create(parallel) as queue:
yield clusterk.runner(queue, parallel)
else:
yield multi.runner(parallel, config)
except:
if view is not None:
ipython.stop(view)
raise
else:
for x in ["cores_per_job", "num_jobs", "mem"]:
parallel.pop(x, None)
if checkpoint_file:
with open(checkpoint_file, "w") as out_handle:
out_handle.write("done\n")