def run_multicore(fn, items, config, parallel=None):
"""Run the function using multiple cores on the given items to process.
"""
if len(items) == 0:
return []
if parallel is None or "num_jobs" not in parallel:
if parallel is None:
parallel = {
"type": "local",
"cores": config["algorithm"].get("num_cores", 1)
}
sysinfo = system.get_info({}, parallel)
parallel = resources.calculate(
parallel,
items,
sysinfo,
config,
parallel.get("multiplier", 1),
max_multicore=int(parallel.get("max_multicore", sysinfo["cores"])))
items = [
config_utils.add_cores_to_config(x, parallel["cores_per_job"])
for x in items
]
if joblib is None:
raise ImportError("Need joblib for multiprocessing parallelization")
out = []
for data in joblib.Parallel(parallel["num_jobs"],
batch_size=1,
backend="multiprocessing")(
joblib.delayed(fn)(x) for x in items):
if data:
out.extend(data)
return out
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 run_multicore(fn, items, config, parallel=None):
"""Run the function using multiple cores on the given items to process.
"""
if len(items) == 0:
return []
if parallel is None or "num_jobs" not in parallel:
if parallel is None:
parallel = {"type": "local", "cores": config["algorithm"].get("num_cores", 1)}
sysinfo = system.get_info({}, parallel)
parallel = resources.calculate(
parallel,
items,
sysinfo,
config,
parallel.get("multiplier", 1),
max_multicore=int(parallel.get("max_multicore", sysinfo["cores"])),
)
items = [config_utils.add_cores_to_config(x, parallel["cores_per_job"]) for x in items]
if joblib is None:
raise ImportError("Need joblib for multiprocessing parallelization")
out = []
for data in joblib.Parallel(parallel["num_jobs"])(joblib.delayed(fn)(x) for x in items):
if data:
out.extend(data)
return out
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")
def _calculate_resources(data, args, resources):
parallel = clargs.to_parallel(args)
config = data[0][0]['config']
config['resources'].update({resources['name']: {'memory': "%sg" % resources['mem'], 'cores': resources['cores']}})
parallel.update({'progs': [resources['name']]})
# parallel = log.create_base_logger(config, parallel)
# log.setup_local_logging(config, parallel)
log.setup_log(config, parallel)
dirs = {'work': os.path.abspath(os.getcwd())}
system.write_info(dirs, parallel, config)
sysinfo = system.machine_info()[0]
log.logger.info("Number of items %s" % len(data))
parallel = res.calculate(parallel, data, sysinfo, config)
log.logger.info(parallel)
# print parallel
# raise
return parallel
"bcbio_system.yaml")
except ValueError as err:
print(err)
print(
"WARNING: Attempting to read bcbio_system.yaml in the current directory."
)
system_config = "bcbio_system.yaml"
with open(system_config) as in_handle:
config = yaml.load(in_handle)
res = {'cores': args.cores_per_job}
config["algorithm"] = {"num_cores": args.cores_per_job}
config["resources"].update({'sambamba': res, 'samtools': res})
config["log_dir"] = os.path.join(os.path.abspath(os.getcwd()), "log")
parallel = clargs.to_parallel(args)
parallel.update({'progs': ['samtools', 'sambamba']})
parallel = log.create_base_logger(config, parallel)
log.setup_local_logging(config, parallel)
dirs = {'work': os.path.abspath(os.getcwd())}
system.write_info(dirs, parallel, config)
sysinfo = system.machine_info()[0]
samples = _get_samples_to_process(args.csv, out_dir, config,
args.force_single, args.separators)
parallel = resources.calculate(parallel, [samples], sysinfo, config)
with prun.start(parallel, samples, config, dirs) as run_parallel:
with profile.report("prepare bcbio samples", dirs):
samples = run_parallel("prepare_bcbio_samples", samples)
create_new_csv(samples, args)
system_config = args.galaxy
with open(system_config) as in_handle:
config = yaml.load(in_handle)
parallel = clargs.to_parallel(args)
parallel.update({'progs': args.progs})
dirs = {'work': os.path.abspath(os.getcwd())}
if args.sys_info.find(";") > -1:
info = args.sys_info.split(";")
sysinfo = {'cores': int(info[0]), 'memory': float(info[1])}
else:
if utils.file_exists(args.sys_info):
sysinfo = yaml.load(open(args.sys_info))[0]
print "system info %s" % sysinfo
samples = []
pipelines, config = _pair_samples_with_pipelines(args.yaml_file, config)
for s in pipelines:
samples = [item for item in pipelines[s]]
print "number of samples %s" % len(samples)
print "after calculate fn"
parallel = resources.calculate(parallel, samples, sysinfo, config)
print parallel
if args.fixed:
print "after fixed ipython fn"
parallel = ipython_fn(parallel, config)
else:
print "after ipython fn"
parallel = ipython_current(parallel, config)
print parallel
ipc_fn(parallel)
