Learn about the concept of a Job in Pachyderm.

December 2, 2022


Note that Pachyderm uses two different scopes when referring to a job.

  • A “global” scope tracking down your entire provenance chain. Refer to GlobalID for more details.
  • And a “local” scope in which a job (also referred to as sub job) is an execution of one particular pipeline.

The following page details the latter.

Definition #

A Pachyderm job is an execution of a pipeline that triggers when new data is detected in an input repository.


When a commit is made to the input repo of a pipeline, jobs are created for all of the downstream pipelines of a DAG. Those jobs are not running yet; each one is waiting until the prior pipeline(s) that it depends on in your DAG produces their output, which then becomes the input for the waiting pipeline.

Each job runs your code against the current commit in a <repo>@<branch> and then submits the results to the output repository of the pipeline as a single output commit. A pipeline triggers a new job every time you submit new changes, a commit, into your input source.

Each job has an alphanumeric identifier (ID) that you can reference in the <pipeline>@<jobID> format.

You can obtain information about all jobs sharing the same ID (Global ID) by running list jobs <jobID> or restrict to a particular pipeline list jobs -p <pipeline>, or inspect jobs <pipeline>@<jobID> --raw.

Job Statuses #

Find a list of all possible job stages below and a state diagram detailing how a job transitions from one state to another.

Stage Description
CREATED An input commit exists, but the job has not been started by a worker yet.
STARTING The worker has allocated resources for the job (that is, the job counts towards parallelism), but it is still waiting on the inputs to be ready.
UNRUNNABLE The job could not be run, because one or more of its inputs is the result of a failed or unrunnable job. As a simple example, say that pipelines Y and Z both depend on the output from pipeline X. If pipeline X fails, both pipeline Y and Z will pass from STARTING to UNRUNNABLE to signify that they had to be cancelled because of upstream failures.
RUNNING The worker is processing datums.
EGRESS The worker has completed all the datums and is uploading the output to the egress endpoint.
FINISHING After all of the datum processing and egress (if any) is done, the job transitions to a finishing state where all of the post-processing tasks such as compaction are performed.
FAILURE The worker encountered too many errors when processing a datum.
KILLED The job timed out, or a user called StopJob
SUCCESS None of the bad stuff happened.

Below, the state transition diagram of a job:

Job State Diagram

List Jobs #

They are various ways to list jobs in Pachyderm, depending on the expected outcome:

  • The pachctl list jobs command returns the list of all global jobs.

  • The pachctl list jobs <jobID> command returns the list of all jobs sharing the same <jobID>.

  • Note that you can also track your jobs downstream as they complete by running pachctl wait jobs <jobID>.

  • The pachctl list jobs -p <pipeline> command returns the list of all the jobs run in a given pipeline.

Example #

pachctl list jobs -p edges

System Response:

ID                               PIPELINE STARTED      DURATION           RESTART PROGRESS  DL       UL       STATE
fd9454d06d8e4fa38a75c8cd20b39538 edges    20 hours ago 5 seconds          0       2 + 1 / 3 181.1KiB 111.4KiB success
5a78358d4b53494cbba4550428f2fe98 edges    20 hours ago 2 seconds          0       1 + 0 / 1 57.27KiB 22.22KiB success
7dcd77a2f7f34ff384a6096d1139e922 edges    20 hours ago Less than a second 0       0 + 0 / 0 0B       0B       success

For each (sub) job, Pachyderm shows the time the pipeline started, its duration, data downloaded and uploaded, the STATE of the pipeline execution, and the number of datums in the PROGRESS section. The format of the progress column is DATUMS PROCESSED + DATUMS SKIPPED / TOTAL DATUMS.

See Datum Processing States for details on Datum states.

Inspect Job #

The pachctl inspect jobs <pipeline>@<jobID> command enables you to view detailed information about a specific (sub)job in a given pipeline (state, number of datums processed/failed/skipped, data downloaded, uploaded, process time, image:tag used to transform your data, etc…). Along with checking the logs, it is especially useful when troubleshooting a failed job.

Example #

Add a --raw flag to output a detailed JSON version of the job.

pachctl inspect jobs edges@fd9454d06d8e4fa38a75c8cd20b39538 --raw

System Response:

    "job": {
        "pipeline": {
        "name": "edges"
        "id": "fd9454d06d8e4fa38a75c8cd20b39538"
    "pipeline_version": "1",
    "output_commit": {
        "branch": {
        "repo": {
            "name": "edges",
            "type": "user"
        "name": "master"
        "id": "fd9454d06d8e4fa38a75c8cd20b39538"
    "data_processed": "2",
    "data_skipped": "1",
    "data_total": "3",
    "stats": {
        "download_time": "0.113263653s",
        "process_time": "1.020472976s",
        "upload_time": "0.010323995s",
        "download_bytes": "185424",
        "upload_bytes": "114041"
    "state": "JOB_SUCCESS",
    "created": "2021-08-02T20:13:10.461841493Z",
    "started": "2021-08-02T20:13:32.870023561Z",
    "finished": "2021-08-02T20:13:38.691891860Z",
    "details": {
        "transform": {
        "image": "pachyderm/opencv:1.0",
        "cmd": [
        "input": {
        "pfs": {
            "name": "images",
            "repo": "images",
            "repo_type": "user",
            "branch": "master",
            "commit": "fd9454d06d8e4fa38a75c8cd20b39538",
            "glob": "/*"
        "salt": "27bbe39ccae54cc2976e3f960a2e1f94",
        "datum_tries": "3"