jobs

Legacy. This endpoint is read-only and disabled by default in new installations.

API endpoint base: https://pirca.arvadosapi.com/arvados/v1/jobs

Object type: 8i9sb

Example UUID: zzzzz-8i9sb-0123456789abcde

Resource

A job describes a work order to be executed by the Arvados cluster.

Each job has, in addition to the Common resource fields:

Attribute Type Description Notes
script string The filename of the job script. This program will be invoked by Crunch for each job task. It is given as a path to an executable file, relative to the /crunch_scripts directory in the Git tree specified by the repository and script_version attributes.
script_parameters hash The input parameters for the job. Conventionally, one of the parameters is called "input". Typically, some parameter values are collection UUIDs. Ultimately, though, the significance of parameters is left entirely up to the script itself.
repository string Git repository name or URL. Source of the repository where the given script_version is to be found. This can be given as the name of a locally hosted repository, or as a publicly accessible URL starting with git://, http://, or https://.
Examples:
yourusername/yourrepo
https://github.com/arvados/arvados.git
script_version string Git commit During a create transaction, this is the Git branch, tag, or hash supplied by the client. Before the job starts, Arvados updates it to the full 40-character SHA-1 hash of the commit used by the job.
See Specifying Git versions below for more detail about acceptable ways to specify a commit.
cancelled_by_client_uuid string API client ID Is null if job has not been cancelled
cancelled_by_user_uuid string Authenticated user ID Is null if job has not been cancelled
cancelled_at datetime When job was cancelled Is null if job has not been cancelled
started_at datetime When job started running Is null if job has not [yet] started
finished_at datetime When job finished running Is null if job has not [yet] finished
running boolean Whether the job is running
success boolean Whether the job indicated successful completion Is null if job has not finished
is_locked_by_uuid string UUID of the user who has locked this job Is null if job is not locked. The system user locks the job when starting the job, in order to prevent job attributes from being altered.
node_uuids array List of UUID strings for node objects that have been assigned to this job
log string Collection UUID Is null if the job has not finished. After the job runs, the given collection contains a text file with log messages provided by the arv-crunch-job task scheduler as well as the standard error streams provided by the task processes.
tasks_summary hash Summary of task completion states. Example: {"done":0,"running":4,"todo":2,"failed":0}
output string Collection UUID Is null if the job has not finished.
nondeterministic boolean The job is expected to produce different results if run more than once. If true, this job will not be considered as a candidate for automatic re-use when submitting subsequent identical jobs.
submit_id string Unique ID provided by client when job was submitted Optional. This can be used by a client to make the jobs.create method idempotent.
priority string
arvados_sdk_version string Git commit hash that specifies the SDK version to use from the Arvados repository This is set by searching the Arvados repository for a match for the arvados_sdk_version runtime constraint.
docker_image_locator string Portable data hash of the collection that contains the Docker image to use This is set by searching readable collections for a match for the docker_image runtime constraint.
runtime_constraints hash Constraints that must be satisfied by the job/task scheduler in order to run the job. See below.
components hash Name and uuid pairs representing the child work units of this job. The uuids can be of different object types. Example components hash: {"name1": "zzzzz-8i9sb-xyz...", "name2": "zzzzz-d1hrv-xyz...",}

Specifying Git versions

The script_version attribute and arvados_sdk_version runtime constraint are typically given as a branch, tag, or commit hash, but there are many more ways to specify a Git commit. The “specifying revisions” section of the gitrevisions manual page has a definitive list. Arvados accepts Git versions in any format listed there that names a single commit (not a tree, a blob, or a range of commits). However, some kinds of names can be expected to resolve differently in Arvados than they do in your local repository. For example, HEAD@{1} refers to the local reflog, and origin/master typically refers to a remote branch: neither is likely to work as desired if given as a Git version.

Runtime constraints

Key Type Description Implemented
arvados_sdk_version string The Git version of the SDKs to use from the Arvados git repository. See Specifying Git versions for more detail about acceptable ways to specify a commit. If you use this, you must also specify a docker_image constraint (see below). In order to install the Python SDK successfully, Crunch must be able to find and run virtualenv inside the container.
docker_image string The Docker image that this Job needs to run. If specified, Crunch will create a Docker container from this image, and run the Job’s script inside that. The Keep mount and work directories will be available as volumes inside this container. The image must be uploaded to Arvados using arv keep docker. You may specify the image in any format that Docker accepts, such as arvados/jobs, debian:latest, or the Docker image id. Alternatively, you may specify the portable data hash of the image Collection.
min_nodes integer
max_nodes integer
min_cores_per_node integer Require that each node assigned to this Job have the specified number of CPU cores
min_ram_mb_per_node integer Require that each node assigned to this Job have the specified amount of real memory (in MiB)
min_scratch_mb_per_node integer Require that each node assigned to this Job have the specified amount of scratch storage available (in MiB)
max_tasks_per_node integer Maximum simultaneous tasks on a single node
keep_cache_mb_per_task integer Size of file data buffer for per-task Keep directory ($TASK_KEEPMOUNT), in MiB. Default is 256 MiB. Increase this to reduce cache thrashing in situtations such as accessing multiple large (64+ MiB) files at the same time, or accessing different parts of a large file at the same time.
min_ram_per_task integer Minimum real memory (KiB) per task

Methods

See Common resource methods for more information about create, delete, get, list, and update.

Required arguments are displayed in green.

cancel

Cancel a job that is queued or running.

Arguments:

Argument Type Description Location Example
uuid string path

create

Create a new Job.

Arguments:

Argument Type Description Location Example
job object Job resource request body
minimum_script_version string Git branch, tag, or commit hash specifying the minimum acceptable script version (earliest ancestor) to consider when deciding whether to re-use a past job.1 query "c3e86c9"
exclude_script_versions array of strings Git commit branches, tags, or hashes to exclude when deciding whether to re-use a past job. query ["8f03c71","8f03c71"]
["badtag1","badtag2"]
filters array of arrays Conditions to find Jobs to reuse. query
find_or_create boolean Before creating, look for an existing job that has identical script, script_version, and script_parameters to those in the present job, has nondeterministic=false, and did not fail (it could be queued, running, or completed). If such a job exists, respond with the existing job instead of submitting a new one. query false

When a job is submitted to the queue using the create method, the script_version attribute is updated to a full 40-character Git commit hash based on the current content of the specified repository. If script_version cannot be resolved, the job submission is rejected.

1 See the note about specifying Git commits for more detail.

Specialized filters

Special filter operations are available for specific Job columns.

  • script_version in git REFSPEC, arvados_sdk_version in git REFSPEC
    Resolve REFSPEC to a list of Git commits, and match jobs with a script_version or arvados_sdk_version in that list. When creating a job and filtering script_version, the search will find commits between REFSPEC and the submitted job’s script_version; all other searches will find commits between REFSPEC and HEAD. This list may include parallel branches if there is more than one path between REFSPEC and the end commit in the graph. Use not in or not in git filters (below) to blacklist specific commits.
  • script_version not in git REFSPEC, arvados_sdk_version not in git REFSPEC
    Resolve REFSPEC to a list of Git commits, and match jobs with a script_version or arvados_sdk_version not in that list.
  • docker_image_locator in docker SEARCH
    SEARCH can be a Docker image hash, a repository name, or a repository name and tag separated by a colon (:). The server will find collections that contain a Docker image that match that search criteria, then match jobs with a docker_image_locator in that list.
  • docker_image_locator not in docker SEARCH
    Negate the in docker filter.

Reusing jobs

Because Arvados records the exact version of the script, input parameters, and runtime environment that was used to run the job, if the script is deterministic (meaning that the same code version is guaranteed to produce the same outputs from the same inputs) then it is possible to re-use the results of past jobs, and avoid re-running the computation to save time. Arvados uses the following algorithm to determine if a past job can be re-used:

  1. If find_or_create is false or omitted, create a new job and skip the rest of these steps.
  2. If filters are specified, find jobs that match those filters. If any filters are given, there must be at least one filter on the repository attribute and one on the script attribute: otherwise an error is returned.
  3. If filters are not specified, find jobs with the same repository and script, with a script_version between minimum_script_version and script_version inclusively (excluding excluded_script_versions), and a docker_image_locator with the latest Collection that matches the submitted job’s docker_image constraint. If the submitted job includes an arvados_sdk_version constraint, jobs must have an arvados_sdk_version between that refspec and HEAD to be found. This form is deprecated: use filters instead.
  4. If the found jobs include a completed job, and all found completed jobs have consistent output, return one of them. Which specific job is returned is undefined.
  5. If the found jobs only include incomplete jobs, return one of them. Which specific job is returned is undefined.
  6. If no job has been returned so far, create and return a new job.

Examples

Run the script “crunch_scripts/hash.py” in the repository “you” using the “master” commit. Arvados should re-use a previous job if the script_version of the previous job is the same as the current “master” commit. This works irrespective of whether the previous job was submitted using the name “master”, a different branch name or tag indicating the same commit, a SHA-1 commit hash, etc.

{
  "job": {
    "script": "hash.py",
    "repository": "you/you",
    "script_version": "master",
    "script_parameters": {
      "input": "c1bad4b39ca5a924e481008009d94e32+210"
    }
  },
  "find_or_create": true
}

Run using exactly the version “d00220fb38d4b85ca8fc28a8151702a2b9d1dec5”. Arvados should re-use a previous job if the “script_version” of that job is also “d00220fb38d4b85ca8fc28a8151702a2b9d1dec5”.

{
  "job": {
    "script": "hash.py",
    "repository": "you/you",
    "script_version": "d00220fb38d4b85ca8fc28a8151702a2b9d1dec5",
    "script_parameters": {
      "input": "c1bad4b39ca5a924e481008009d94e32+210"
    }
  },
  "find_or_create": true
}

Arvados should re-use a previous job if the “script_version” of the previous job is between “earlier_version_tag” and the “master” commit (inclusive), but not the commit indicated by “blacklisted_version_tag”. If there are no previous jobs matching these criteria, run the job using the “master” commit.

{
  "job": {
    "script": "hash.py",
    "repository": "you/you",
    "script_version": "master",
    "script_parameters": {
      "input": "c1bad4b39ca5a924e481008009d94e32+210"
    }
  },
  "minimum_script_version": "earlier_version_tag",
  "exclude_script_versions": ["blacklisted_version_tag"],
  "find_or_create": true
}

The same behavior, using filters:

{
  "job": {
    "script": "hash.py",
    "repository": "you/you",
    "script_version": "master",
    "script_parameters": {
      "input": "c1bad4b39ca5a924e481008009d94e32+210"
    }
  },
  "filters": [["script", "=", "hash.py"],
              ["repository", "=", "you/you"],
              ["script_version", "in git", "earlier_version_tag"],
              ["script_version", "not in git", "blacklisted_version_tag"]],
  "find_or_create": true
}

Run the script “crunch_scripts/monte-carlo.py” in the repository “you/you” using the current “master” commit. Because it is marked as “nondeterministic”, this job will not be considered as a suitable candidate for future job submissions that use the “find_or_create” feature.

{
  "job": {
    "script": "monte-carlo.py",
    "repository": "you/you",
    "script_version": "master",
    "nondeterministic": true,
    "script_parameters": {
      "input": "c1bad4b39ca5a924e481008009d94e32+210"
    }
  }
}

delete

Delete an existing Job.

Arguments:

Argument Type Description Location Example
uuid string The UUID of the Job in question. path

get

Gets a Job’s metadata by UUID.

Arguments:

Argument Type Description Location Example
uuid string The UUID of the Job in question. path

list

List jobs.

See common resource list method.

See the create method documentation for more information about Job-specific filters.

log_tail_follow

log_tail_follow jobs

Arguments:

Argument Type Description Location Example
uuid string path
buffer_size integer (default 8192) query

queue

Get the current job queue.

Arguments:

Argument Type Description Location Example
order string query
filters array query

This method is equivalent to the list method, except that the results are restricted to queued jobs (i.e., jobs that have not yet been started or cancelled) and order defaults to queue priority.

update

Update attributes of an existing Job.

Arguments:

Argument Type Description Location Example
uuid string The UUID of the Job in question. path
job object query

Previous: Crunch scripts Next: job_tasks

The content of this documentation is licensed under the Creative Commons Attribution-Share Alike 3.0 United States licence.
Code samples in this documentation are licensed under the Apache License, Version 2.0.