Commandline Options¶
A quick way to see all of Toil’s commandline options is by executing the following on a toil script:
$ toil example.py --help
For a basic toil workflow, Toil has one mandatory argument, the job store. All other arguments are optional.
The Job Store¶
Running toil scripts requires a filepath or url to a centralizing location for all of the files of the workflow.
This is Toil’s one required positional argument: the job store. To use the quickstart example,
if you’re on a node that has a large /scratch volume, you can specify that the jobstore be created there by
executing: python HelloWorld.py /scratch/my-job-store
, or more explicitly,
python HelloWorld.py file:/scratch/my-job-store
.
Syntax for specifying different job stores:
Local:
file:job-store-name
AWS:
aws:region-here:job-store-name
Google:
google:projectID-here:job-store-name
Different types of job store options can be found below.
Commandline Options¶
Core Toil Options Options to specify the location of the Toil workflow and turn on stats collation about the performance of jobs.
--workDir WORKDIR Absolute path to directory where temporary files generated during the Toil run should be placed. Standard output and error from batch system jobs (unless –noStdOutErr) will be placed in this directory. A cache directory may be placed in this directory. Temp files and folders will be placed in a toil-<workflowID> within workDir. The workflowID is generated by Toil and will be reported in the workflow logs. Default is determined by the variables (TMPDIR, TEMP, TMP) via mkdtemp. This directory needs to exist on all machines running jobs; if capturing standard output and error from batch system jobs is desired, it will generally need to be on a shared file system. When sharing a cache between containers on a host, this directory must be shared between the containers. --coordinationDir COORDINATION_DIR Absolute path to directory where Toil will keep state and lock files. When sharing a cache between containers on a host, this directory must be shared between the containers. --noStdOutErr Do not capture standard output and error from batch system jobs. --stats Records statistics about the toil workflow to be used by ‘toil stats’. --clean=STATE Determines the deletion of the jobStore upon completion of the program. Choices: ‘always’, ‘onError’,’never’, or ‘onSuccess’. The --stats option requires information from the jobStore upon completion so the jobStore will never be deleted with that flag. If you wish to be able to restart the run, choose ‘never’ or ‘onSuccess’. Default is ‘never’ if stats is enabled, and ‘onSuccess’ otherwise --cleanWorkDir STATE Determines deletion of temporary worker directory upon completion of a job. Choices: ‘always’, ‘onError’, ‘never’, or ‘onSuccess’. Default = always. WARNING: This option should be changed for debugging only. Running a full pipeline with this option could fill your disk with intermediate data. --clusterStats FILEPATH If enabled, writes out JSON resource usage statistics to a file. The default location for this file is the current working directory, but an absolute path can also be passed to specify where this file should be written. This option only applies when using scalable batch systems. --restart If --restart is specified then will attempt to restart existing workflow at the location pointed to by the --jobStore option. Will raise an exception if the workflow does not exist.
Logging Options Toil hides stdout and stderr by default except in case of job failure. Log levels in toil are based on priority from the logging module:
--logOff Only CRITICAL log levels are shown. Equivalent to --logLevel=OFF
or--logLevel=CRITICAL
.--logCritical Only CRITICAL log levels are shown. Equivalent to --logLevel=OFF
or--logLevel=CRITICAL
.--logError Only ERROR, and CRITICAL log levels are shown. Equivalent to --logLevel=ERROR
.--logWarning Only WARN, ERROR, and CRITICAL log levels are shown. Equivalent to --logLevel=WARNING
.--logInfo All log statements are shown, except DEBUG. Equivalent to --logLevel=INFO
.--logDebug All log statements are shown. Equivalent to --logLevel=DEBUG
.--logLevel=LOGLEVEL May be set to: OFF
(orCRITICAL
),ERROR
,WARN
(orWARNING
),INFO
, orDEBUG
.--logFile FILEPATH Specifies a file path to write the logging output to. --rotatingLogging Turn on rotating logging, which prevents log files from getting too big (set using --maxLogFileSize BYTESIZE
).--maxLogFileSize BYTESIZE The maximum size of a job log file to keep (in bytes), log files larger than this will be truncated to the last X bytes. Setting this option to zero will prevent any truncation. Setting this option to a negative value will truncate from the beginning. Default=62.5KiB Sets the maximum log file size in bytes ( --rotatingLogging
must be active).--log-dir DIRPATH For CWL and local file system only. Log stdout and stderr (if tool requests stdout/stderr) to the DIRPATH.
Batch System Options
--batchSystem BATCHSYSTEM The type of batch system to run the job(s) with, currently can be one of aws_batch, parasol, single_machine, grid_engine, lsf, mesos, slurm, tes, torque, htcondor, kubernetes. (default: single_machine) --disableAutoDeployment Should auto-deployment of the user script be deactivated? If True, the user script/package should be present at the same location on all workers. Default = False. --maxLocalJobs MAXLOCALJOBS For batch systems that support a local queue for housekeeping jobs (Mesos, GridEngine, htcondor, lsf, slurm, torque). Specifies the maximum number of these housekeeping jobs to run on the local system. The default (equal to the number of cores) is a maximum of concurrent local housekeeping jobs. --manualMemArgs Do not add the default arguments: ‘hv=MEMORY’ & ‘h_vmem=MEMORY’ to the qsub call, and instead rely on TOIL_GRIDGENGINE_ARGS to supply alternative arguments. Requires that TOIL_GRIDGENGINE_ARGS be set. --runCwlInternalJobsOnWorkers Whether to run CWL internal jobs (e.g. CWLScatter) on the worker nodes instead of the primary node. If false (default), then all such jobs are run on the primary node. Setting this to true can speed up the pipeline for very large workflows with many sub-workflows and/or scatters, provided that the worker pool is large enough. --coalesceStatusCalls Coalese status calls to prevent the batch system from being overloaded. Currently only supported for LSF. --statePollingWait STATEPOLLINGWAIT Time, in seconds, to wait before doing a scheduler query for job state. Return cached results if within the waiting period. Only works for grid engine batch systems such as gridengine, htcondor, torque, slurm, and lsf. --parasolCommand PARASOLCOMMAND The name or path of the parasol program. Will be looked up on PATH unless it starts with a slash. (default: parasol) --parasolMaxBatches PARASOLMAXBATCHES Maximum number of job batches the Parasol batch is allowed to create. One batch is created for jobs with a unique set of resource requirements. (default: 1000) --mesosEndpoint MESOSENDPOINT The host and port of the Mesos server separated by a colon. (default: <leader IP>:5050) --kubernetesHostPath KUBERNETES_HOST_PATH Path on Kubernetes hosts to use as shared inter-pod temp directory. --kubernetesOwner KUBERNETES_OWNER Username to mark Kubernetes jobs with. --kubernetesServiceAccount KUBERNETES_SERVICE_ACCOUNT Service account to run jobs as. --kubernetesPodTimeout KUBERNETES_POD_TIMEOUT Seconds to wait for a scheduled Kubernetes pod to start running. (default: 120s) --tesEndpoint TES_ENDPOINT The http(s) URL of the TES server. (default: http://<leader IP>:8000) --tesUser TES_USER User name to use for basic authentication to TES server. --tesPassword TES_PASSWORD Password to use for basic authentication to TES server. --tesBearerToken TES_BEARER_TOKEN Bearer token to use for authentication to TES server. --awsBatchRegion AWS_BATCH_REGION The AWS region containing the AWS Batch queue to submit to. --awsBatchQueue AWS_BATCH_QUEUE The name or ARN of the AWS Batch queue to submit to. --awsBatchJobRoleArn AWS_BATCH_JOB_ROLE_ARN The ARN of an IAM role to run AWS Batch jobs as, so they can e.g. access a job store. Must be assumable by ecs-tasks.amazonaws.com --scale SCALE A scaling factor to change the value of all submitted tasks’ submitted cores. Used in single_machine batch system. Useful for running workflows on smaller machines than they were designed for, by setting a value less than 1. (default: 1)
Data Storage Options Allows configuring Toil’s data storage.
--linkImports When using a filesystem based job store, CWL input files are by default symlinked in. Specifying this option instead copies the files into the job store, which may protect them from being modified externally. When not specified and as long as caching is enabled, Toil will protect the file automatically by changing the permissions to read-only. --moveExports When using a filesystem based job store, output files are by default moved to the output directory, and a symlink to the moved exported file is created at the initial location. Specifying this option instead copies the files into the output directory. Applies to filesystem-based job stores only. --disableCaching Disables caching in the file store. This flag must be set to use a batch system that does not support cleanup, such as Parasol. --caching BOOL Set caching options. This must be set to “false” to use a batch system that does not support cleanup, such as Parasol. Set to “true” if caching is desired.
Autoscaling Options Allows the specification of the minimum and maximum number of nodes in an autoscaled cluster, as well as parameters to control the level of provisioning.
--provisioner CLOUDPROVIDER The provisioner for cluster auto-scaling. This is the main Toil --provisioner option, and defaults to None for running on single_machine and non-auto-scaling batch systems. The currently supported choices are ‘aws’ or ‘gce’. --nodeTypes NODETYPES Specifies a list of comma-separated node types, each of which is composed of slash-separated instance types, and an optional spot bid set off by a colon, making the node type preemptible. Instance types may appear in multiple node types, and the same node type may appear as both preemptible and non-preemptible.
- Valid argument specifying two node types:
- c5.4xlarge/c5a.4xlarge:0.42,t2.large
- Node types:
- c5.4xlarge/c5a.4xlarge:0.42 and t2.large
- Instance types:
- c5.4xlarge, c5a.4xlarge, and t2.large
- Semantics:
- Bid $0.42/hour for either c5.4xlarge or c5a.4xlarge instances, treated interchangeably, while they are available at that price, and buy t2.large instances at full price
--minNodes MINNODES Minimum number of nodes of each type in the cluster, if using auto-scaling. This should be provided as a comma-separated list of the same length as the list of node types. default=0 --maxNodes MAXNODES Maximum number of nodes of each type in the cluster, if using autoscaling, provided as a comma-separated list. The first value is used as a default if the list length is less than the number of nodeTypes. default=10 --targetTime TARGETTIME Sets how rapidly you aim to complete jobs in seconds. Shorter times mean more aggressive parallelization. The autoscaler attempts to scale up/down so that it expects all queued jobs will complete within targetTime seconds. (Default: 1800) --betaInertia BETAINERTIA A smoothing parameter to prevent unnecessary oscillations in the number of provisioned nodes. This controls an exponentially weighted moving average of the estimated number of nodes. A value of 0.0 disables any smoothing, and a value of 0.9 will smooth so much that few changes will ever be made. Must be between 0.0 and 0.9. (Default: 0.1) --scaleInterval SCALEINTERVAL The interval (seconds) between assessing if the scale of the cluster needs to change. (Default: 60) --preemptibleCompensation PREEMPTIBLECOMPENSATION The preference of the autoscaler to replace preemptible nodes with non-preemptible nodes, when preemptible nodes cannot be started for some reason. Defaults to 0.0. This value must be between 0.0 and 1.0, inclusive. A value of 0.0 disables such compensation, a value of 0.5 compensates two missing preemptible nodes with a non-preemptible one. A value of 1.0 replaces every missing pre-emptable node with a non-preemptible one. --nodeStorage NODESTORAGE Specify the size of the root volume of worker nodes when they are launched in gigabytes. You may want to set this if your jobs require a lot of disk space. The default value is 50. --nodeStorageOverrides NODESTORAGEOVERRIDES Comma-separated list of nodeType:nodeStorage that are used to override the default value from --nodeStorage for the specified nodeType(s). This is useful for heterogeneous jobs where some tasks require much more disk than others. --metrics Enable the prometheus/grafana dashboard for monitoring CPU/RAM usage, queue size, and issued jobs. --assumeZeroOverhead Ignore scheduler and OS overhead and assume jobs can use every last byte of memory and disk on a node when autoscaling.
Service Options Allows the specification of the maximum number of service jobs in a cluster. By keeping this limited we can avoid nodes occupied with services causing deadlocks. (Not for CWL).
--maxServiceJobs MAXSERVICEJOBS The maximum number of service jobs that can be run concurrently, excluding service jobs running on preemptible nodes. default=9223372036854775807 --maxPreemptibleServiceJobs MAXPREEMPTIBLESERVICEJOBS The maximum number of service jobs that can run concurrently on preemptible nodes. default=9223372036854775807 --deadlockWait DEADLOCKWAIT Time, in seconds, to tolerate the workflow running only the same service jobs, with no jobs to use them, before declaring the workflow to be deadlocked and stopping. default=60 --deadlockCheckInterval DEADLOCKCHECKINTERVAL Time, in seconds, to wait between checks to see if the workflow is stuck running only service jobs, with no jobs to use them. Should be shorter than --deadlockWait. May need to be increased if the batch system cannot enumerate running jobs quickly enough, or if polling for running jobs is placing an unacceptable load on a shared cluster. default=30
Resource Options The options to specify default cores/memory requirements (if not specified by the jobs themselves), and to limit the total amount of memory/cores requested from the batch system.
--defaultMemory INT The default amount of memory to request for a job. Only applicable to jobs that do not specify an explicit value for this requirement. Standard suffixes like K, Ki, M, Mi, G or Gi are supported. Default is 2.0G --defaultCores FLOAT The default number of CPU cores to dedicate a job. Only applicable to jobs that do not specify an explicit value for this requirement. Fractions of a core (for example 0.1) are supported on some batch systems, namely Mesos and singleMachine. Default is 1.0 --defaultDisk INT The default amount of disk space to dedicate a job. Only applicable to jobs that do not specify an explicit value for this requirement. Standard suffixes like K, Ki, M, Mi, G or Gi are supported. Default is 2.0G --defaultAccelerators ACCELERATOR The default amount of accelerators to request for a job. Only applicable to jobs that do not specify an explicit value for this requirement. Each accelerator specification can have a type (gpu [default], nvidia, amd, cuda, rocm, opencl, or a specific model like nvidia-tesla-k80), and a count [default: 1]. If both a type and a count are used, they must be separated by a colon. If multiple types of accelerators are used, the specifications are separated by commas. Default is []. --defaultPreemptible BOOL Make all jobs able to run on preemptible (spot) nodes by default. --maxCores INT The maximum number of CPU cores to request from the batch system at any one time. Standard suffixes like K, Ki, M, Mi, G or Gi are supported. --maxMemory INT The maximum amount of memory to request from the batch system at any one time. Standard suffixes like K, Ki, M, Mi, G or Gi are supported. --maxDisk INT The maximum amount of disk space to request from the batch system at any one time. Standard suffixes like K, Ki, M, Mi, G or Gi are supported.
Options for rescuing/killing/restarting jobs. The options for jobs that either run too long/fail or get lost (some batch systems have issues!).
--retryCount RETRYCOUNT Number of times to retry a failing job before giving up and labeling job failed. default=1 --enableUnlimitedPreemptibleRetries If set, preemptible failures (or any failure due to an instance getting unexpectedly terminated) will not count towards job failures and --retryCount. --doubleMem If set, batch jobs which die due to reaching memory limit on batch schedulers will have their memory doubled and they will be retried. The remaining retry count will be reduced by 1. Currently only supported by LSF. default=False. --maxJobDuration MAXJOBDURATION Maximum runtime of a job (in seconds) before we kill it (this is a lower bound, and the actual time before killing the job may be longer). --rescueJobsFrequency RESCUEJOBSFREQUENCY Period of time to wait (in seconds) between checking for missing/overlong jobs, that is jobs which get lost by the batch system. Expert parameter.
Log Management Options
--maxLogFileSize MAXLOGFILESIZE The maximum size of a job log file to keep (in bytes), log files larger than this will be truncated to the last X bytes. Setting this option to zero will prevent any truncation. Setting this option to a negative value will truncate from the beginning. Default=62.5 K --writeLogs FILEPATH Write worker logs received by the leader into their own files at the specified path. Any non-empty standard output and error from failed batch system jobs will also be written into files at this path. The current working directory will be used if a path is not specified explicitly. Note: By default only the logs of failed jobs are returned to leader. Set log level to ‘debug’ or enable --writeLogsFromAllJobs to get logs back from successful jobs, and adjust --maxLogFileSize to control the truncation limit for worker logs. --writeLogsGzip FILEPATH Identical to --writeLogs except the logs files are gzipped on the leader. --writeMessages FILEPATH File to send messages from the leader’s message bus to. --realTimeLogging Enable real-time logging from workers to leader.
Miscellaneous Options
--disableChaining Disables chaining of jobs (chaining uses one job’s resource allocation for its successor job if possible). --disableJobStoreChecksumVerification Disables checksum verification for files transferred to/from the job store. Checksum verification is a safety check to ensure the data is not corrupted during transfer. Currently only supported for non-streaming AWS files --sseKey SSEKEY Path to file containing 32 character key to be used for server-side encryption on awsJobStore or googleJobStore. SSE will not be used if this flag is not passed. --setEnv NAME, -e NAME NAME=VALUE or NAME, -e NAME=VALUE or NAME are also valid. Set an environment variable early on in the worker. If VALUE is omitted, it will be looked up in the current environment. Independently of this option, the worker will try to emulate the leader’s environment before running a job, except for some variables known to vary across systems. Using this option, a variable can be injected into the worker process itself before it is started. --servicePollingInterval SERVICEPOLLINGINTERVAL Interval of time service jobs wait between polling for the existence of the keep-alive flag (default=60) --forceDockerAppliance Disables sanity checking the existence of the docker image specified by TOIL_APPLIANCE_SELF, which Toil uses to provision mesos for autoscaling. --statusWait INT Seconds to wait between reports of running jobs. (default=3600) --disableProgress Disables the progress bar shown when standard error is a terminal.
Debug Options Debug options for finding problems or helping with testing.
--debugWorker Experimental no forking mode for local debugging. Specifically, workers are not forked and stderr/stdout are not redirected to the log. (default=False) --disableWorkerOutputCapture Let worker output go to worker’s standard out/error instead of per-job logs. --badWorker BADWORKER For testing purposes randomly kill --badWorker proportion of jobs using SIGKILL. (Default: 0.0) --badWorkerFailInterval BADWORKERFAILINTERVAL When killing the job pick uniformly within the interval from 0.0 to --badWorkerFailInterval seconds after the worker starts. (Default: 0.01) --kill_polling_interval KILL_POLLING_INTERVAL Interval of time (in seconds) the leader waits between polling for the kill flag inside the job store set by the “toil kill” command. (default=5)
Restart Option¶
In the event of failure, Toil can resume the pipeline by adding the argument
--restart
and rerunning the python script. Toil pipelines (but not CWL
pipelines) can even be edited and resumed which is useful for development or
troubleshooting.
Running Workflows with Services¶
Toil supports jobs, or clusters of jobs, that run as services to other accessor jobs. Example services include server databases or Apache Spark Clusters. As service jobs exist to provide services to accessor jobs their runtime is dependent on the concurrent running of their accessor jobs. The dependencies between services and their accessor jobs can create potential deadlock scenarios, where the running of the workflow hangs because only service jobs are being run and their accessor jobs can not be scheduled because of too limited resources to run both simultaneously. To cope with this situation Toil attempts to schedule services and accessors intelligently, however to avoid a deadlock with workflows running service jobs it is advisable to use the following parameters:
--maxServiceJobs
: The maximum number of service jobs that can be run concurrently, excluding service jobs running on preemptible nodes.--maxPreemptibleServiceJobs
: The maximum number of service jobs that can run concurrently on preemptible nodes.
Specifying these parameters so that at a maximum cluster size there will be sufficient resources to run accessors in addition to services will ensure that such a deadlock can not occur.
If too low a limit is specified then a deadlock can occur in which toil can
not schedule sufficient service jobs concurrently to complete the workflow.
Toil will detect this situation if it occurs and throw a
toil.DeadlockException
exception. Increasing the cluster size
and these limits will resolve the issue.
Setting Options directly with the Toil Script¶
It’s good to remember that commandline options can be overridden in the Toil script itself. For example,
toil.job.Job.Runner.getDefaultOptions()
can be used to run toil with all default options, and in this example,
it will override commandline args to run the default options and always run with the “./toilWorkflow” directory
specified as the jobstore:
options = Job.Runner.getDefaultOptions("./toilWorkflow") # Get the options object
with Toil(options) as toil:
toil.start(Job()) # Run the script
However, each option can be explicitly set within the script by supplying arguments (in this example, we are setting
logLevel = "DEBUG"
(all log statements are shown) and clean="ALWAYS"
(always delete the jobstore) like so:
options = Job.Runner.getDefaultOptions("./toilWorkflow") # Get the options object
options.logLevel = "DEBUG" # Set the log level to the debug level.
options.clean = "ALWAYS" # Always delete the jobStore after a run
with Toil(options) as toil:
toil.start(Job()) # Run the script
However, the usual incantation is to accept commandline args from the user with the following:
parser = Job.Runner.getDefaultArgumentParser() # Get the parser
options = parser.parse_args() # Parse user args to create the options object
with Toil(options) as toil:
toil.start(Job()) # Run the script
Which can also, of course, then accept script supplied arguments as before (which will overwrite any user supplied args):
parser = Job.Runner.getDefaultArgumentParser() # Get the parser
options = parser.parse_args() # Parse user args to create the options object
options.logLevel = "DEBUG" # Set the log level to the debug level.
options.clean = "ALWAYS" # Always delete the jobStore after a run
with Toil(options) as toil:
toil.start(Job()) # Run the script