# Copyright (C) 2015-2021 Regents of the University of California
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import enum
import logging
import os
import shutil
import time
from abc import ABC, abstractmethod
from argparse import ArgumentParser, _ArgumentGroup
from collections.abc import Iterator
from contextlib import contextmanager
from threading import Condition
from typing import Any, ContextManager, NamedTuple, Optional, Union, cast
from toil.batchSystems.options import OptionSetter
from toil.bus import MessageBus, MessageOutbox
from toil.common import Config, Toil, cacheDirName
from toil.deferred import DeferredFunctionManager
from toil.fileStores.abstractFileStore import AbstractFileStore
from toil.job import JobDescription, ParsedRequirement, Requirer
from toil.lib.memoize import memoize
from toil.resource import Resource
logger = logging.getLogger(__name__)
# Value to use as exitStatus in UpdatedBatchJobInfo.exitStatus when status is not available.
EXIT_STATUS_UNAVAILABLE_VALUE = 255
[docs]
class BatchJobExitReason(enum.IntEnum):
FINISHED = 1
"""Successfully finished."""
FAILED = 2
"""Job finished, but failed."""
LOST = 3
"""Preemptable failure (job's executing host went away)."""
KILLED = 4
"""Job killed before finishing."""
ERROR = 5
"""Internal error."""
MEMLIMIT = 6
"""Job hit batch system imposed memory limit."""
MISSING = 7
"""Job disappeared from the scheduler without actually stopping, so Toil killed it."""
MAXJOBDURATION = 8
"""Job ran longer than --maxJobDuration, so Toil killed it."""
PARTITION = 9
"""Job was not able to talk to the leader via the job store, so Toil declared it failed."""
[docs]
@classmethod
def to_string(cls, value: int) -> str:
"""
Convert to human-readable string.
Given an int that may be or may be equal to a value from the enum,
produce the string value of its matching enum entry, or a stringified
int.
"""
try:
return cls(value).name
except ValueError:
return str(value)
[docs]
class UpdatedBatchJobInfo(NamedTuple):
jobID: int
exitStatus: int
"""
The exit status (integer value) of the job. 0 implies successful.
EXIT_STATUS_UNAVAILABLE_VALUE is used when the exit status is not available
(e.g. job is lost, or otherwise died but actual exit code was not reported).
"""
exitReason: Optional[BatchJobExitReason]
wallTime: Union[float, int, None]
# Information required for worker cleanup on shutdown of the batch system.
[docs]
class WorkerCleanupInfo(NamedTuple):
work_dir: Optional[str]
"""Work directory path (where the cache would go) if specified by user"""
coordination_dir: Optional[str]
"""Coordination directory path (where lock files would go) if specified by user"""
workflow_id: str
"""Used to identify files specific to this workflow"""
clean_work_dir: str
"""
When to clean up the work and coordination directories for a job ('always',
'onSuccess', 'onError', 'never')
"""
[docs]
class AbstractBatchSystem(ABC):
"""An abstract base class to represent the interface the batch system must provide to Toil."""
[docs]
@classmethod
@abstractmethod
def supportsAutoDeployment(cls) -> bool:
"""
Whether this batch system supports auto-deployment of the user script itself.
If it does, the :meth:`setUserScript` can be invoked to set the resource
object representing the user script.
Note to implementors: If your implementation returns True here, it should also override
"""
raise NotImplementedError()
[docs]
@classmethod
@abstractmethod
def supportsWorkerCleanup(cls) -> bool:
"""
Whether this batch system supports worker cleanup.
Indicates whether this batch system invokes
:meth:`BatchSystemSupport.workerCleanup` after the last job for a
particular workflow invocation finishes. Note that the term *worker*
refers to an entire node, not just a worker process. A worker process
may run more than one job sequentially, and more than one concurrent
worker process may exist on a worker node, for the same workflow. The
batch system is said to *shut down* after the last worker process
terminates.
"""
raise NotImplementedError()
[docs]
def setUserScript(self, userScript: Resource) -> None:
"""
Set the user script for this workflow.
This method must be called before the first job is
issued to this batch system, and only if :meth:`.supportsAutoDeployment` returns True,
otherwise it will raise an exception.
:param userScript: the resource object representing the user script
or module and the modules it depends on.
"""
raise NotImplementedError()
[docs]
def set_message_bus(self, message_bus: MessageBus) -> None:
"""
Give the batch system an opportunity to connect directly to the message
bus, so that it can send informational messages about the jobs it is
running to other Toil components.
"""
[docs]
@abstractmethod
def issueBatchJob(
self,
command: str,
job_desc: JobDescription,
job_environment: Optional[dict[str, str]] = None,
) -> int:
"""
Issues a job with the specified command to the batch system and returns
a unique job ID number.
:param command: the command to execute somewhere to run the Toil
worker process
:param job_desc: the JobDescription for the job being run
:param job_environment: a collection of job-specific environment
variables to be set on the worker.
:return: a unique job ID number that can be used to reference the newly
issued job
"""
raise NotImplementedError()
[docs]
@abstractmethod
def killBatchJobs(self, jobIDs: list[int]) -> None:
"""
Kills the given job IDs. After returning, the killed jobs will not
appear in the results of getRunningBatchJobIDs. The killed job will not
be returned from getUpdatedBatchJob.
:param jobIDs: list of IDs of jobs to kill
"""
raise NotImplementedError()
# FIXME: Return value should be a set (then also fix the tests)
[docs]
@abstractmethod
def getIssuedBatchJobIDs(self) -> list[int]:
"""
Gets all currently issued jobs
:return: A list of jobs (as job ID numbers) currently issued (may be
running, or may be waiting to be run). Despite the result being a
list, the ordering should not be depended upon.
"""
raise NotImplementedError()
[docs]
@abstractmethod
def getRunningBatchJobIDs(self) -> dict[int, float]:
"""
Gets a map of jobs as job ID numbers that are currently running (not
just waiting) and how long they have been running, in seconds.
:return: dictionary with currently running job ID number keys and how
many seconds they have been running as the value
"""
raise NotImplementedError()
[docs]
@abstractmethod
def getUpdatedBatchJob(self, maxWait: int) -> Optional[UpdatedBatchJobInfo]:
"""
Returns information about job that has updated its status (i.e. ceased
running, either successfully or with an error). Each such job will be
returned exactly once.
Does not return info for jobs killed by killBatchJobs, although they
may cause None to be returned earlier than maxWait.
:param maxWait: the number of seconds to block, waiting for a result
:return: If a result is available, returns UpdatedBatchJobInfo.
Otherwise it returns None. wallTime is the number of seconds (a strictly
positive float) in wall-clock time the job ran for, or None if this
batch system does not support tracking wall time.
"""
raise NotImplementedError()
[docs]
def getSchedulingStatusMessage(self) -> Optional[str]:
"""
Get a log message fragment for the user about anything that might be
going wrong in the batch system, if available.
If no useful message is available, return None.
This can be used to report what resource is the limiting factor when
scheduling jobs, for example. If the leader thinks the workflow is
stuck, the message can be displayed to the user to help them diagnose
why it might be stuck.
:return: User-directed message about scheduling state.
"""
# Default implementation returns None.
# Override to provide scheduling status information.
return None
[docs]
@abstractmethod
def shutdown(self) -> None:
"""
Called at the completion of a toil invocation.
Should cleanly terminate all worker threads.
"""
raise NotImplementedError()
[docs]
def setEnv(self, name: str, value: Optional[str] = None) -> None:
"""
Set an environment variable for the worker process before it is launched.
The worker process will typically inherit the environment of the machine
it is running on but this method makes it possible to override specific
variables in that inherited environment before the worker is launched.
Note that this mechanism is different to the one used by the worker
internally to set up the environment of a job. A call to this method affects
all jobs issued after this method returns. Note to implementors: This
means that you would typically need to copy the variables before enqueuing a job.
If no value is provided it will be looked up from the current environment.
"""
raise NotImplementedError()
[docs]
@classmethod
def add_options(cls, parser: Union[ArgumentParser, _ArgumentGroup]) -> None:
"""
If this batch system provides any command line options, add them to the given parser.
"""
[docs]
@classmethod
def setOptions(cls, setOption: OptionSetter) -> None:
"""
Process command line or configuration options relevant to this batch system.
:param setOption: A function with signature
setOption(option_name, parsing_function=None, check_function=None, default=None, env=None)
returning nothing, used to update run configuration as a side effect.
"""
[docs]
def getWorkerContexts(self) -> list[ContextManager[Any]]:
"""
Get a list of picklable context manager objects to wrap worker work in,
in order.
Can be used to ask the Toil worker to do things in-process (such as
configuring environment variables, hot-deploying user scripts, or
cleaning up a node) that would otherwise require a wrapping "executor"
process.
"""
return []
[docs]
class BatchSystemSupport(AbstractBatchSystem):
"""Partial implementation of AbstractBatchSystem, support methods."""
def __init__(
self, config: Config, maxCores: float, maxMemory: int, maxDisk: int
) -> None:
"""
Initialize initial state of the object.
:param toil.common.Config config: object is setup by the toilSetup script and
has configuration parameters for the jobtree. You can add code
to that script to get parameters for your batch system.
:param float maxCores: the maximum number of cores the batch system can
request for any one job
:param int maxMemory: the maximum amount of memory the batch system can
request for any one job, in bytes
:param int maxDisk: the maximum amount of disk space the batch system can
request for any one job, in bytes
"""
super().__init__()
self.config = config
self.maxCores = maxCores
self.maxMemory = maxMemory
self.maxDisk = maxDisk
self.environment: dict[str, str] = {}
if config.workflowID is None:
raise Exception("config.workflowID must be set")
else:
self.workerCleanupInfo = WorkerCleanupInfo(
work_dir=config.workDir,
coordination_dir=config.coordination_dir,
workflow_id=config.workflowID,
clean_work_dir=config.cleanWorkDir,
)
self._outbox: Optional[MessageOutbox] = None
[docs]
def check_resource_request(self, requirer: Requirer) -> None:
"""
Check resource request is not greater than that available or allowed.
:param requirer: Object whose requirements are being checked
:param str job_name: Name of the job being checked, for generating a useful error report.
:param str detail: Batch-system-specific message to include in the error.
:raise InsufficientSystemResources: raised when a resource is requested in an amount
greater than allowed
"""
try:
for resource, requested, available in [
("cores", requirer.cores, self.maxCores),
("memory", requirer.memory, self.maxMemory),
("disk", requirer.disk, self.maxDisk),
]:
assert requested is not None
if requested > available:
raise InsufficientSystemResources(requirer, resource, available)
# Handle accelerators in another method that can be overridden separately
self._check_accelerator_request(requirer)
except InsufficientSystemResources as e:
# Add more annotation info to the error
e.batch_system = self.__class__.__name__ or None
e.source = self.config.workDir if e.resource == "disk" else None
raise e
def _check_accelerator_request(self, requirer: Requirer) -> None:
"""
Raise an InsufficientSystemResources error if the batch system can't
provide the accelerators that are required.
If a batch system *can* provide accelerators, it should override this
to say so.
"""
if len(requirer.accelerators) > 0:
# By default we assume we can't fulfill any of these
raise InsufficientSystemResources(
requirer,
"accelerators",
[],
details=["The batch system does not support any accelerators."],
)
[docs]
def setEnv(self, name: str, value: Optional[str] = None) -> None:
"""
Set an environment variable for the worker process before it is launched. The worker
process will typically inherit the environment of the machine it is running on but this
method makes it possible to override specific variables in that inherited environment
before the worker is launched. Note that this mechanism is different to the one used by
the worker internally to set up the environment of a job. A call to this method affects
all jobs issued after this method returns. Note to implementors: This means that you
would typically need to copy the variables before enqueuing a job.
If no value is provided it will be looked up from the current environment.
:param name: the environment variable to be set on the worker.
:param value: if given, the environment variable given by name will be set to this value.
If None, the variable's current value will be used as the value on the worker
:raise RuntimeError: if value is None and the name cannot be found in the environment
"""
if value is None:
try:
value = os.environ[name]
except KeyError:
raise RuntimeError(f"{name} does not exist in current environment")
self.environment[name] = value
[docs]
def set_message_bus(self, message_bus: MessageBus) -> None:
"""
Give the batch system an opportunity to connect directly to the message
bus, so that it can send informational messages about the jobs it is
running to other Toil components.
"""
# We do in fact send messages to the message bus.
self._outbox = message_bus.outbox()
[docs]
@memoize
def get_batch_logs_dir(self) -> str:
"""
Get the directory where the backing batch system should save its logs.
Only really makes sense if the backing batch system actually saves logs
to a filesystem; Kubernetes for example does not. Ought to be a
directory shared between the leader and the workers, if the backing
batch system writes logs onto the worker's view of the filesystem, like
many HPC schedulers do.
"""
if self.config.batch_logs_dir:
# Use what is specified
if not os.path.isdir(self.config.batch_logs_dir):
# But if it doesn't exist, make it exist
os.makedirs(self.config.batch_logs_dir, exist_ok=True)
return self.config.batch_logs_dir
# And if nothing is specified use the workDir.
return Toil.getToilWorkDir(self.config.workDir)
[docs]
@staticmethod
def workerCleanup(info: WorkerCleanupInfo) -> None:
"""
Cleans up the worker node on batch system shutdown.
Also see :meth:`supportsWorkerCleanup`.
:param WorkerCleanupInfo info: A named tuple consisting of all the relevant information
for cleaning up the worker.
"""
logger.debug("Attempting worker cleanup")
assert isinstance(info, WorkerCleanupInfo)
assert info.workflow_id is not None
workflowDir = Toil.getLocalWorkflowDir(info.workflow_id, info.work_dir)
coordination_dir = Toil.get_local_workflow_coordination_dir(
info.workflow_id, info.work_dir, info.coordination_dir
)
DeferredFunctionManager.cleanupWorker(coordination_dir)
workflowDirContents = os.listdir(workflowDir)
AbstractFileStore.shutdownFileStore(
info.workflow_id, info.work_dir, info.coordination_dir
)
if info.clean_work_dir in ("always", "onSuccess", "onError"):
if workflowDirContents in ([], [cacheDirName(info.workflow_id)]):
logger.debug("Deleting workflow directory %s", workflowDir)
shutil.rmtree(workflowDir, ignore_errors=True)
if coordination_dir != workflowDir:
# No more coordination to do here either.
logger.debug("Deleting coordination directory %s", coordination_dir)
shutil.rmtree(coordination_dir, ignore_errors=True)
[docs]
class NodeInfo:
"""
The coresUsed attribute is a floating point value between 0 (all cores idle) and 1 (all cores
busy), reflecting the CPU load of the node.
The memoryUsed attribute is a floating point value between 0 (no memory used) and 1 (all memory
used), reflecting the memory pressure on the node.
The coresTotal and memoryTotal attributes are the node's resources, not just the used resources
The requestedCores and requestedMemory attributes are all the resources that Toil Jobs have reserved on the
node, regardless of whether the resources are actually being used by the Jobs.
The workers attribute is an integer reflecting the number of workers currently active workers
on the node.
"""
def __init__(
self,
coresUsed: float,
memoryUsed: float,
coresTotal: float,
memoryTotal: int,
requestedCores: float,
requestedMemory: int,
workers: int,
) -> None:
self.coresUsed = coresUsed
self.memoryUsed = memoryUsed
self.coresTotal = coresTotal
self.memoryTotal = memoryTotal
self.requestedCores = requestedCores
self.requestedMemory = requestedMemory
self.workers = workers
[docs]
class AbstractScalableBatchSystem(AbstractBatchSystem):
"""
A batch system that supports a variable number of worker nodes.
Used by :class:`toil.provisioners.clusterScaler.ClusterScaler`
to scale the number of worker nodes in the cluster
up or down depending on overall load.
"""
[docs]
@abstractmethod
def getNodes(
self, preemptible: Optional[bool] = None, timeout: int = 600
) -> dict[str, NodeInfo]:
"""
Returns a dictionary mapping node identifiers of preemptible or non-preemptible nodes to
NodeInfo objects, one for each node.
:param preemptible: If True (False) only (non-)preemptible nodes will be returned.
If None, all nodes will be returned.
"""
raise NotImplementedError()
[docs]
@abstractmethod
def nodeInUse(self, nodeIP: str) -> bool:
"""
Can be used to determine if a worker node is running any tasks. If the node is doesn't
exist, this function should simply return False.
:param nodeIP: The worker nodes private IP address
:return: True if the worker node has been issued any tasks, else False
"""
raise NotImplementedError()
[docs]
@abstractmethod
def ignoreNode(self, nodeAddress: str) -> None:
"""
Stop sending jobs to this node. Used in autoscaling
when the autoscaler is ready to terminate a node, but
jobs are still running. This allows the node to be terminated
after the current jobs have finished.
:param nodeAddress: IP address of node to ignore.
"""
raise NotImplementedError()
[docs]
@abstractmethod
def unignoreNode(self, nodeAddress: str) -> None:
"""
Stop ignoring this address, presumably after
a node with this address has been terminated. This allows for the
possibility of a new node having the same address as a terminated one.
"""
raise NotImplementedError()
[docs]
class InsufficientSystemResources(Exception):
def __init__(
self,
requirer: Requirer,
resource: str,
available: Optional[ParsedRequirement] = None,
batch_system: Optional[str] = None,
source: Optional[str] = None,
details: list[str] = [],
) -> None:
"""
Make a new exception about how we couldn't get enough of something.
:param requirer: What needed the resources. May have a .jobName string.
:param resource: The kind of resource requested (cores, memory, disk, accelerators).
:param requested: The amount requested.
:param available: The amount actually available.
:param batch_system: The batch system that could not provide the resource.
:param source: The place where the resource was to be gotten from. For disk, should be a path.
:param details: Any extra details about the problem that can be attached to the error.
"""
self.job_name: Optional[str] = str(requirer)
self.resource = resource
self.requested = cast(ParsedRequirement, getattr(requirer, resource))
self.available = available
self.batch_system = batch_system
self.source = source
self.details = details
[docs]
def __str__(self) -> str:
"""
Explain the exception.
"""
unit = "bytes of " if self.resource in ("disk", "memory") else ""
purpose = " for temporary space" if self.resource == "disk" else ""
qualifier = (
" free on {self.source}"
if self.resource == "disk" and self.source is not None
else ""
)
msg = []
if self.job_name is not None:
msg.append(f"The job {self.job_name} is requesting ")
else:
msg.append(f"Requesting ")
msg.append(f"{self.requested} {unit}{self.resource}")
msg.append(purpose)
if self.available is not None:
msg.append(
f', more than the maximum of {self.available} {unit}{self.resource}{qualifier} that {self.batch_system or "this batch system"} was configured with'
)
if self.resource in ("cores", "memory", "disk"):
msg.append(f", or enforced by --max{self.resource.capitalize()}")
else:
msg.append(", but that is not available")
msg.append(".")
if self.resource == "disk":
msg.append(
' Try setting/changing the toil option "--workDir" or changing the base temporary directory by setting TMPDIR.'
)
for detail in self.details:
msg.append(" ")
msg.append(detail)
return "".join(msg)
[docs]
class AcquisitionTimeoutException(Exception):
"""To be raised when a resource request times out."""
def __init__(
self,
resource: str,
requested: Union[int, float, set[int]],
available: Union[int, float, set[int]],
) -> None:
"""
Creates an instance of this exception that indicates which resource is insufficient for
current demands, as well as the resources requested and actually available.
:param str resource: string representing the resource type
:param requested: the resources requested that resulted in this exception
:param available: the resources actually available
"""
self.requested = requested
self.available = available
self.resource = resource
[docs]
class ResourcePool:
"""
Represents an integral amount of a resource (such as memory bytes).
Amounts can be acquired immediately or with a timeout, and released.
Provides a context manager to do something with an amount of resource
acquired.
"""
def __init__(
self, initial_value: int, resource_type: str, timeout: float = 5
) -> None:
super().__init__()
# We use this condition to signal everyone whenever some resource is released.
# We use its associated lock to guard value.
self.condition = Condition()
# This records how much resource is available right now.
self.value = initial_value
self.resource_type = resource_type
self.timeout = timeout
[docs]
def acquireNow(self, amount: int) -> bool:
"""
Reserve the given amount of the given resource.
Returns True if successful and False if this is not possible immediately.
"""
with self.condition:
if amount > self.value:
return False
self.value -= amount
self.__validate()
return True
[docs]
def acquire(self, amount: int) -> None:
"""
Reserve the given amount of the given resource.
Raises AcquisitionTimeoutException if this is not possible in under
self.timeout time.
"""
with self.condition:
startTime = time.time()
while amount > self.value:
if time.time() - startTime >= self.timeout:
# This means the thread timed out waiting for the resource.
raise AcquisitionTimeoutException(
resource=self.resource_type,
requested=amount,
available=self.value,
)
# Allow self.timeout seconds to get the resource, else quit
# through the above if condition. This wait + timeout is the
# last thing in the loop such that a request that takes longer
# than self.timeout due to multiple wakes under the threshold
# are still honored.
self.condition.wait(timeout=self.timeout)
self.value -= amount
self.__validate()
[docs]
def release(self, amount: int) -> None:
with self.condition:
self.value += amount
self.__validate()
self.condition.notify_all()
def __validate(self) -> None:
assert 0 <= self.value
[docs]
def __str__(self) -> str:
return str(self.value)
[docs]
def __repr__(self) -> str:
return "ResourcePool(%i)" % self.value
[docs]
@contextmanager
def acquisitionOf(self, amount: int) -> Iterator[None]:
self.acquire(amount)
try:
yield
finally:
self.release(amount)
[docs]
class ResourceSet:
"""
Represents a collection of distinct resources (such as accelerators).
Subsets can be acquired immediately or with a timeout, and released.
Provides a context manager to do something with a set of of resources
acquired.
"""
def __init__(
self, initial_value: set[int], resource_type: str, timeout: float = 5
) -> None:
super().__init__()
# We use this condition to signal everyone whenever some resource is released.
# We use its associated lock to guard value.
self.condition = Condition()
# This records what resources are available right now.
self.value = initial_value
self.resource_type = resource_type
self.timeout = timeout
[docs]
def acquireNow(self, subset: set[int]) -> bool:
"""
Reserve the given amount of the given resource.
Returns True if successful and False if this is not possible immediately.
"""
with self.condition:
if subset > self.value:
return False
self.value -= subset
return True
[docs]
def acquire(self, subset: set[int]) -> None:
"""
Reserve the given amount of the given resource.
Raises AcquisitionTimeoutException if this is not possible in under
self.timeout time.
"""
with self.condition:
startTime = time.time()
while subset > self.value:
if time.time() - startTime >= self.timeout:
# This means the thread timed out waiting for the resource.
raise AcquisitionTimeoutException(
resource=self.resource_type,
requested=subset,
available=self.value,
)
# Allow self.timeout seconds to get the resource, else quit
# through the above if condition. This wait + timeout is the
# last thing in the loop such that a request that takes longer
# than self.timeout due to multiple wakes under the threshold
# are still honored.
self.condition.wait(timeout=self.timeout)
self.value -= subset
[docs]
def release(self, subset: set[int]) -> None:
with self.condition:
self.value |= subset
self.condition.notify_all()
[docs]
def get_free_snapshot(self) -> set[int]:
"""
Get a snapshot of what items are free right now.
May be stale as soon as you get it, but you will need some kind of hint
to try and do an acquire.
"""
return set(self.value)
[docs]
def __str__(self) -> str:
return str(self.value)
[docs]
def __repr__(self) -> str:
return "ResourceSet(%s)" % self.value
[docs]
@contextmanager
def acquisitionOf(self, subset: set[int]) -> Iterator[None]:
self.acquire(subset)
try:
yield
finally:
self.release(subset)
