def __init__(
self,
workflow,
dag,
cores,
local_cores=1,
dryrun=False,
touch=False,
cluster=None,
cluster_status=None,
cluster_config=None,
cluster_sync=None,
drmaa=None,
drmaa_log_dir=None,
kubernetes=None,
container_image=None,
tibanna=None,
tibanna_sfn=None,
precommand="",
tibanna_config=False,
jobname=None,
quiet=False,
printreason=False,
printshellcmds=False,
keepgoing=False,
max_jobs_per_second=None,
max_status_checks_per_second=100,
latency_wait=3,
greediness=1.0,
force_use_threads=False,
assume_shared_fs=True,
keepincomplete=False,
):
""" Create a new instance of KnapsackJobScheduler. """
from ratelimiter import RateLimiter
self.cluster = cluster
self.cluster_config = cluster_config
self.cluster_sync = cluster_sync
self.dag = dag
self.workflow = workflow
self.dryrun = dryrun
self.touch = touch
self.quiet = quiet
self.keepgoing = keepgoing
self.running = set()
self.failed = set()
self.finished_jobs = 0
self.greediness = 1
self.max_jobs_per_second = max_jobs_per_second
self.keepincomplete = keepincomplete
self.global_resources = {
name: (sys.maxsize if res is None else res)
for name, res in workflow.global_resources.items()
}
self.resources = dict(self.global_resources)
use_threads = (force_use_threads or (os.name != "posix") or cluster
or cluster_sync or drmaa)
self._open_jobs = threading.Semaphore(0)
self._lock = threading.Lock()
self._errors = False
self._finished = False
self._job_queue = None
self._submit_callback = self._noop
self._finish_callback = partial(
self._proceed,
update_dynamic=not self.dryrun,
print_progress=not self.quiet and not self.dryrun,
)
self._local_executor = None
if dryrun:
self._executor = DryrunExecutor(
workflow,
dag,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
)
elif touch:
self._executor = TouchExecutor(
workflow,
dag,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
)
elif cluster or cluster_sync or (drmaa is not None):
if not workflow.immediate_submit:
# No local jobs when using immediate submit!
# Otherwise, they will fail due to missing input
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cores=local_cores,
keepincomplete=keepincomplete,
)
if cluster or cluster_sync:
if cluster_sync:
constructor = SynchronousClusterExecutor
else:
constructor = partial(
GenericClusterExecutor,
statuscmd=cluster_status,
max_status_checks_per_second=
max_status_checks_per_second,
)
self._executor = constructor(
workflow,
dag,
None,
submitcmd=(cluster or cluster_sync),
cluster_config=cluster_config,
jobname=jobname,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
assume_shared_fs=assume_shared_fs,
keepincomplete=keepincomplete,
)
if workflow.immediate_submit:
self._submit_callback = partial(
self._proceed,
update_dynamic=False,
print_progress=False,
update_resources=False,
handle_job_success=False,
)
else:
self._executor = DRMAAExecutor(
workflow,
dag,
None,
drmaa_args=drmaa,
drmaa_log_dir=drmaa_log_dir,
jobname=jobname,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cluster_config=cluster_config,
assume_shared_fs=assume_shared_fs,
max_status_checks_per_second=max_status_checks_per_second,
keepincomplete=keepincomplete,
)
elif kubernetes:
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cores=local_cores,
keepincomplete=keepincomplete,
)
self._executor = KubernetesExecutor(
workflow,
dag,
kubernetes,
container_image=container_image,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cluster_config=cluster_config,
keepincomplete=keepincomplete,
)
elif tibanna:
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
use_threads=use_threads,
latency_wait=latency_wait,
cores=local_cores,
keepincomplete=keepincomplete,
)
self._executor = TibannaExecutor(
workflow,
dag,
cores,
tibanna_sfn,
precommand=precommand,
tibanna_config=tibanna_config,
container_image=container_image,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
keepincomplete=keepincomplete,
)
else:
self._executor = CPUExecutor(
workflow,
dag,
cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
use_threads=use_threads,
latency_wait=latency_wait,
cores=cores,
keepincomplete=keepincomplete,
)
if self.max_jobs_per_second and not self.dryrun:
max_jobs_frac = Fraction(
self.max_jobs_per_second).limit_denominator()
self.rate_limiter = RateLimiter(max_calls=max_jobs_frac.numerator,
period=max_jobs_frac.denominator)
else:
# essentially no rate limit
self.rate_limiter = DummyRateLimiter()
self._user_kill = None
signal.signal(signal.SIGTERM, self.exit_gracefully)
self._open_jobs.release()
def __init__(
self,
workflow,
dag,
local_cores=1,
dryrun=False,
touch=False,
cluster=None,
cluster_status=None,
cluster_config=None,
cluster_sync=None,
drmaa=None,
drmaa_log_dir=None,
kubernetes=None,
container_image=None,
tibanna=None,
tibanna_sfn=None,
google_lifesciences=None,
google_lifesciences_regions=None,
google_lifesciences_location=None,
google_lifesciences_cache=False,
tes=None,
precommand="",
preemption_default=None,
preemptible_rules=None,
tibanna_config=False,
jobname=None,
quiet=False,
printreason=False,
printshellcmds=False,
keepgoing=False,
max_jobs_per_second=None,
max_status_checks_per_second=100,
latency_wait=3,
greediness=1.0,
force_use_threads=False,
assume_shared_fs=True,
keepincomplete=False,
keepmetadata=True,
scheduler_type=None,
scheduler_ilp_solver=None,
):
"""Create a new instance of KnapsackJobScheduler."""
from ratelimiter import RateLimiter
cores = workflow.global_resources["_cores"]
self.cluster = cluster
self.cluster_config = cluster_config
self.cluster_sync = cluster_sync
self.dag = dag
self.workflow = workflow
self.dryrun = dryrun
self.touch = touch
self.quiet = quiet
self.keepgoing = keepgoing
self.running = set()
self.failed = set()
self.finished_jobs = 0
self.greediness = 1
self.max_jobs_per_second = max_jobs_per_second
self.keepincomplete = keepincomplete
self.keepmetadata = keepmetadata
self.scheduler_type = scheduler_type
self.scheduler_ilp_solver = scheduler_ilp_solver
self._tofinish = []
self.global_resources = {
name: (sys.maxsize if res is None else res)
for name, res in workflow.global_resources.items()
}
if workflow.global_resources["_nodes"] is not None:
# Do not restrict cores locally if nodes are used (i.e. in case of cluster/cloud submission).
self.global_resources["_cores"] = sys.maxsize
self.resources = dict(self.global_resources)
use_threads = (
force_use_threads
or (os.name != "posix")
or cluster
or cluster_sync
or drmaa
)
self._open_jobs = threading.Semaphore(0)
self._lock = threading.Lock()
self._errors = False
self._finished = False
self._job_queue = None
self._submit_callback = self._noop
self._finish_callback = partial(
self._proceed,
update_dynamic=not self.dryrun,
print_progress=not self.quiet and not self.dryrun,
)
self._local_executor = None
if dryrun:
self._executor = DryrunExecutor(
workflow,
dag,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
)
elif touch:
self._executor = TouchExecutor(
workflow,
dag,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
)
elif cluster or cluster_sync or (drmaa is not None):
if not workflow.immediate_submit:
# No local jobs when using immediate submit!
# Otherwise, they will fail due to missing input
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cores=local_cores,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
if cluster or cluster_sync:
if cluster_sync:
constructor = SynchronousClusterExecutor
else:
constructor = partial(
GenericClusterExecutor,
statuscmd=cluster_status,
max_status_checks_per_second=max_status_checks_per_second,
)
self._executor = constructor(
workflow,
dag,
None,
submitcmd=(cluster or cluster_sync),
cluster_config=cluster_config,
jobname=jobname,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
assume_shared_fs=assume_shared_fs,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
if workflow.immediate_submit:
self._submit_callback = partial(
self._proceed,
update_dynamic=False,
print_progress=False,
update_resources=False,
handle_job_success=False,
)
else:
self._executor = DRMAAExecutor(
workflow,
dag,
None,
drmaa_args=drmaa,
drmaa_log_dir=drmaa_log_dir,
jobname=jobname,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cluster_config=cluster_config,
assume_shared_fs=assume_shared_fs,
max_status_checks_per_second=max_status_checks_per_second,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
elif kubernetes:
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cores=local_cores,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
self._executor = KubernetesExecutor(
workflow,
dag,
kubernetes,
container_image=container_image,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cluster_config=cluster_config,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
elif tibanna:
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
use_threads=use_threads,
latency_wait=latency_wait,
cores=local_cores,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
self._executor = TibannaExecutor(
workflow,
dag,
cores,
tibanna_sfn,
precommand=precommand,
tibanna_config=tibanna_config,
container_image=container_image,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
elif google_lifesciences:
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cores=local_cores,
)
self._executor = GoogleLifeSciencesExecutor(
workflow,
dag,
cores,
container_image=container_image,
regions=google_lifesciences_regions,
location=google_lifesciences_location,
cache=google_lifesciences_cache,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
preemption_default=preemption_default,
preemptible_rules=preemptible_rules,
)
elif tes:
self._local_executor = CPUExecutor(
workflow,
dag,
local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
cores=local_cores,
keepincomplete=keepincomplete,
)
self._executor = TaskExecutionServiceExecutor(
workflow,
dag,
cores=local_cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
latency_wait=latency_wait,
tes_url=tes,
container_image=container_image,
)
else:
self._executor = CPUExecutor(
workflow,
dag,
cores,
printreason=printreason,
quiet=quiet,
printshellcmds=printshellcmds,
use_threads=use_threads,
latency_wait=latency_wait,
cores=cores,
keepincomplete=keepincomplete,
keepmetadata=keepmetadata,
)
if self.max_jobs_per_second and not self.dryrun:
max_jobs_frac = Fraction(self.max_jobs_per_second).limit_denominator()
self.rate_limiter = RateLimiter(
max_calls=max_jobs_frac.numerator, period=max_jobs_frac.denominator
)
else:
# essentially no rate limit
self.rate_limiter = DummyRateLimiter()
# Choose job selector (greedy or ILP)
self.job_selector = self.job_selector_greedy
if scheduler_type == "ilp":
import pulp
if pulp.apis.LpSolverDefault is None:
logger.warning(
"Falling back to greedy scheduler because no default "
"solver is found for pulp (you have to install either "
"coincbc or glpk)."
)
else:
self.job_selector = self.job_selector_ilp
self._user_kill = None
try:
signal.signal(signal.SIGTERM, self.exit_gracefully)
except ValueError:
# If this fails, it is due to scheduler not being invoked in the main thread.
# This can only happen with --gui, in which case it is fine for now.
pass
self._open_jobs.release()