Project

General

Profile

Actions
History

Wiki » Proposals »

Containers API » History » Revision 9

« Previous | Revision 9/64 (diff) | Next »
Tom Clegg, 06/05/2015 07:05 PM

Jobs API (DRAFT)

Clients control JobRequests. The system controls Jobs, and assigns them to JobRequests. When the system has assigned a Job to a JobRequest, anyone with permission to read the JobRequest also has permission to read the Job.

A JobRequest describes job constraints which can have different interpretations over time. For example, a JobRequest with a {"kind":"git_tree","commit_range":"abc123..master",...} mount might be satisfiable by any of several different source trees, and this set of satisfying source trees can change when the repository's "master" branch is updated.

A Job is an unambiguously specified process. Git trees, data collections, docker images, etc. are specified using content addresses. A Job serves as a statement of exactly what computation will be attempted and, later, a record of what computation was done.

Use cases

Preview

Tell me how you would satisfy job request X. Which pdh/commits would be used? Is the satisfying job already started? finished?

Submit a previewed existing job

I'm happy with the already-running/finished job you showed me in "preview". Give me access to that job, its logs, and [when it finishes] its output.

Submit a previewed new job

I'm happy with the new job the "preview" response proposed to run. Run that job.

Submit a new job (disable reuse)

I don't want to use an already-running/finished job. Run a new job that satisfies my job request.

Submit a new duplicate job (disable reuse)

I'm happy with the already-running/finished job you showed me in "preview". Run a new job exactly like that one.

Select a job and associate it with my JobRequest

I'm not happy with the job you chose, but I know of another job that satisfies my request. Assuming I'm right about that, attach my JobRequest to the existing job of my choice.

Just do the right thing without a preview

Satisfy job request X one way or another, and tell me the resulting job's UUID.

JobRequest/Job life cycle

Illustrating job re-use and preview facility:
  1. Client CA creates a JobRequest JRA with priority=0.
  2. Server creates job JX and assigns JX to JRA, but does not try to run JX yet because max(priority)=0.
  3. Client CA presents JX to the user. "We haven't computed this result yet, so we'll have to run a new job. Is this OK?"
  4. Client CB creates a JobRequest JRB with priority=1.
  5. Server assigns JX to JRB and puts JX in the execution queue with priority=1.
  6. Client CA updates JRA with priority=2.
  7. Server updates JX with priority=2.
  8. Job JX starts.
  9. Client CA updates JRA with priority=0. (This is the "cancel" operation.)
  10. Server updates JX with priority=1. (JRB still wants this job to complete.)
  11. Job JX finishes.
  12. Clients CA and CB have permission to read JX (ever since JX was assigned to their respective JobRequests) as well as its progress indicators, output, and log.

"JobRequest" schema

Attribute Type Description Discussion Examples
uuid, owner_uuid, modified_by_client_uuid, modified_by_user_uuid string Usual Arvados model attributes
created_at, modified_at datetime Usual Arvados model attributes
name string Unparsed
description text Unparsed
job_uuid uuid The job that satisfies this job request.
Can be null if a suitable job has not yet been found or queued.
Assigned by the system: cannot be modified directly by clients.
If null, it can be changed by the system at any time.
If not null, it can be reset to null by a client if priority is zero.
mounts hash Objects to attach to the container's filesystem and stdin/stdout.
Keys starting with a forward slash indicate objects mounted in the container's filesystem.
Other keys are given special meanings here.
We use "stdin" instead of "/dev/stdin" because literally replacing /dev/stdin with a file would have a confusing effect on many unix programs. The stdin feature only affects the standard input of the first process started in the container; after that, the usual rules apply.		 
{
 "/input/foo":{
  "kind":"collection",
  "portable_data_hash":"d41d8cd98f00b204e9800998ecf8427e+0" 
 },
 "stdin":{
  "kind":"collection_file",
  "uuid":"zzzzz-4zz18-yyyyyyyyyyyyyyy",
  "path":"/foo.txt" 
 },
 "stdout":{
  "kind":"regular_file",
  "path":"/tmp/a.out" 
 }
}
runtime_permissions hash Restrict the job's access to the outside world (apart from its explicitly stated inputs and output).
Each key is the name of a capability, like "internet" or "API" or "clock". The corresponding value is true (the capability must be available in the job's runtime environment) or false (must not). If a key is omitted, availability of the corresponding capability is acceptable but not necessary.		 This is a generalized version of "enforce purity restrictions": it is not a claim that the job will be pure. Rather, it helps us control and track runtime restrictions, which can be helpful when reasoning about whether a given job was pure.
In the most basic implementation, no capabilities are defined, and the only acceptable value of this attribute is the empty hash.
(TC)This name isn't great, and conflicts with the "readable/writable" kind of permissions. Perhaps something along the lines of capabilities or interfaces?
(TC)Is this the same type of feature as requesting memory/disk/cores? Or are those resources assumed not to affect reproducibility?		 
{}
docker_image string Docker image repository and tag, docker image hash, collection UUID, or collection PDH. (TC)Could this be just another mount point, with target "docker_image"?
environment hash environment variables and values that should be set in the container environment (docker run --env). This augments and (when conflicts exists) overrides environment variables given in the image's Dockerfile.
cwd string initial working directory, given as an absolute path (in the container) or a path relative to the WORKDIR given in the image's Dockerfile. The default is ".". 
"/tmp"
command array of strings Command to execute in the container. Default is the CMD given in the image's Dockerfile.
(TC)Possible to specify a pipe, like "echo foo | tr f b"? Any shell variables supported? Or do you just use ["sh","-c","echo $PATH | wc"] if you want a shell?
output_path string Path to a directory or file inside the container that should be preserved as job's output when it finishes. For best performance, point output_path to a writable collection mount.
priority number Higher number means spend more resources (e.g., go ahead of other queued jobs, bring up more nodes).
Zero means a job should not be run on behalf of this request. (Clients are expected to submit JobRequests with zero priority in order to prevew the job that will be used to satisfy it.)		 0, 1000.5, -1
expires_at datetime After this time, priority is considered to be zero. If the assigned job is running at that time, the job may be cancelled to conserve resources. null, 2015-07-01T00:00:01Z

"Job" schema

Attribute Type Description Discussion Examples
uuid, owner_uuid, created_at, modified_at, modified_by_client_uuid, modified_by_user_uuid string Usual Arvados model attributes
state string 
"Queued" 
"Running" 
"Cancelled" 
"Failed" 
"Complete"
started_at, finished_at, log Same as current job
environment hash Must be equal to a JobRequest's environment in order to satisfy the JobRequest. (TC)We could offer a "resolve" process here like we do with mounts: e.g., hash values in the JobRequest environment could be resolved according to the given "kind". I propose we leave room for this feature but don't add it yet.
cwd, command, output_path string Must be equal to the corresponding values in a JobRequest in order to satisfy that JobRequest.
mounts hash Must contain the same keys as the JobRequest being satisfied. Each value must be within the range of values described in the JobRequest at the time the Job is assigned to the JobRequest.
runtime_permissions hash The types of access to the outside world (apart from its explicitly stated inputs and output) available to the job when it runs/ran.
Permission/access types will change over time and it may be hard/impossible to translate old types to new. Such cases may cause old Jobs to be inelegible for assignment to new JobRequests.
output string Portable data hash of the output collection.
pure boolean The job's output is thought to be dependent solely on its inputs, i.e., it is expected to produce identical output if repeated.
We want a feature along these lines, but "pure" seems to be a conclusion we can come to after examining various facts -- rather than a property of an individual job execution event -- and it probably needs something more subtle than a boolean.
docker_image_pdh string Portable data hash of a collection containing the docker image used to run the job. (TC) If docker image hashes can be verified efficiently, we can use the native docker image hash here instead of a collection PDH.
progress number A number between 0.0 and 1.0 describing the fraction of work done.
(TC)How does this relate to child tasks? E.g., is a job supposed to update this itself as its child tasks complete?
priority number Highest priority of all associated JobRequests

Mount types

The "mounts" hash is the primary mechanism for adding data to the container at runtime (beyond what is already in the container image).

Each value of the "mounts" hash is itself a hash, whose "kind" key determines the handler used to attach data to the container.

Mount type kind Expected keys Description Examples Discussion
Arvados data collection collection
"portable_data_hash" or "uuid" may be provided. If not provided, a new collection will be created. This is useful when "writable":true and the job's output_path is (or is a subdirectory of) this mount target.
"writable" may be provided with a true or false to indicate the path must (or must not) be writable. If not specified, the system can choose.
"path" may be provided, and defaults to "/".
At job startup, the target path will have the same directory structure as the given path within the collection. Even if the files/directories are writable in the container, modifications will not be saved back to the original collections when the job ends.		 
{
 "kind":"collection",
 "uuid":"...",
 "path":"/foo.txt" 
}

{
 "kind":"collection",
 "uuid":"..." 
}
Git tree git_tree
One of {"git-url", "repository_name", "uuid"} must be provided.
One of {"commit", "revisions"} must be provided.
"path" may be provided. The default path is "/".
At job startup, the target path will have the source tree indicated by the given revision. The .git metadata directory will not be available: typically the system will use git-archive rather than git-checkout to prepare the target directory.
If a value is given for "revisions", it will be resolved to a set of commits (as desribed in the "ranges" section of git-revisions(1)) and the job request will be satisfiable by any commit in that set.
If a value is given for "commit", it will be resolved to a single commit, and the tree resulting from that commit will be used.
"path" can be used to select a subdirectory or a single file from the tree indicated by the selected commit.
Note that multiple commits can resolve to the same tree: for example, the file/directory given in "path" might not have changed between commits A and B.		 
{
 "kind":"git_tree",
 "uuid":"zzzzz-s0uqq-xxxxxxxxxxxxxxx",
 "commit":"master" 
}

{
 "kind":"git_tree",
 "uuid":"zzzzz-s0uqq-xxxxxxxxxxxxxxx",
 "commit_range":"bugfix^..master",
 "path":"/crunch_scripts/grep" 
}
The resolved mount (found in the Job record) will have only the "kind" key and a "blob" or "tree" key indicating the 40-character hash of the git tree/blob used.
Temporary directory tmp
None
At job startup, the target path will be empty. When the job finishes, the content will be discarded. This will be backed by a memory-based filesystem where possible.		 
{
 "kind":"tmp",
}

(TC)Should add a "max size" feature, to help memfs-backed implementations.

Permissions

Users own JobRequests but the system owns Jobs. Users get permission to read Jobs by virtue of linked JobRequests.

API methods

Changes from the usual REST APIs:

arvados.v1.job_requests.create and .update

These methods can fail when objects referenced in the "mounts" hash do not exist, or the acting user has insufficient permission on them.

arvados.v1.job_requests.update

The job_uuid attribute is special:
  • It cannot be changed from null to non-null by a regular client.
  • It cannot be changed from non-null to null by system processes.
  • It can be reset from non-null to null by the system during a client-initiated update transaction that modifies attributes other than state and priority.
Apart from job_uuid, updates are restricted by the current state of the job request.
  • When state="Preview", all attributes can be updated.
  • When state="Request", only priority and state can be updated.
  • When state="Done", no attributes can be updated.
state cannot be null. The following state transitions are the only ones permitted.
  • Preview → Request
  • Preview → Done
  • Request → Done

arvados.v1.jobs.create and .update

These methods are not callable except by system processes.

arvados.v1.jobs.progress

This method permits specific types of updates while a job is running: update progress, record success/failure.

Q: [How] can a client submitting job B indicate it shouldn't run unless/until job A succeeds?

Debugging

Q: Need any infrastructure debug-logging controls in this API?

Q: Need any job debug-logging controls in this API? Or just use environment vars?

Scheduling and running jobs

Q: If two users submit identical pure jobs and ask to reuse existing jobs, whose token does the job get to use?
  • Should pure jobs be run as a pseudo-user that is given read access to the relevant objects for the duration of the job? (This would make it safer to share jobs -- see #5823)
Q: If two users submit identical pure jobs with different priority, which priority is used?
  • Choices include "whichever is greater" and "sum".

Q: If two users submit identical pure jobs and one cancels -- or one user submits two identical jobs and cancels one -- does the work stop, or continue? What do the job records look like after this?

Updated by Tom Clegg over 9 years ago · 64 revisions