Job

As explained in Machine Learning Platform Engineering

A Kubernetes job is a workload resource that runs a task to completion and terminates after the task succeeds.

A Kubernetes job runs a command that finishes, rather than keeping an application running like a deployment. It fits one-off work that must happen inside a cluster, such as registering Feast features before an ML pipeline retrieves them. section 5.1.1

Register features from inside the cluster

A locally run `feast apply` registered feature file sources using `localhost`, but MinIO is available to the workload through an internal cluster service, not a public endpoint. Register the features again with file sources pointing to MinIO directly, change the S3 endpoint override to `https://minio-service.kubeflow.svc.cluster.local:9000`, and run `feast apply` in a Kubernetes job. The job terminates after the command runs; feature retrieval can then use the nonlocal registry. section 5.1.1

A job can also be an internal deployment step. When a Bento deployment is submitted in Yatai, Yatai runs a Kubernetes job to build its Docker image. Once that job completes, the application’s API Server and Runner pods can start. The image-building job is temporary; the deployment pods serve requests. section 6.1.3

Figure 6.8 Yatai running a job to containerize the application and build an image

Shows the Yatai job that containerizes the application and builds its image before deployment pods launch. section 6.1.3

How Machine Learning Platform Engineering covers it

How deep it goes

Contrasts a job’s terminating command with a deployment and shows two integration points, but does not provide a job manifest.

Examples

Uses an in-cluster `feast apply` task and Yatai’s Bento image-builder job.

What it assumes

Assumes a Kubernetes cluster with MinIO and Feast or a Yatai deployment; introduces the job–deployment distinction directly.

What you can do after

Evaluate/understand when a terminating Kubernetes job fits cluster-side setup and how Yatai uses one before launching serving pods.

How it connects

Connects Kubernetes deployments and pods with Feast, MinIO, Yatai, and Docker image builds.

liveBookmanning.com

Read Machine Learning Platform Engineering on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Kubernetes in Action, Second EditionA Job is a Kubernetes resource that creates pods to run a finite task until a specified number of successful completions is reached.
• 

  liveBookmanning.com
  Kubernetes for DevelopersA Kubernetes Job is a workload resource that runs a finite task to completion rather than maintaining a continuously running service.
• 

  liveBookmanning.com
  API Design PatternsA job is an API resource that stores configuration for work and exposes a separate operation for executing that work.

Compare how 4 books explain job →

Related concepts

• resourcekind of · An API resource is a conceptual entity or target exposed through an interface and addressable for access or manipulation.
• cronjobproduced by · A Kubernetes CronJob is a workload resource that creates Jobs according to a specified recurring schedule.
• backoff limitconfigured by · A backoff limit is a Kubernetes Job setting that specifies the maximum number of retries before the Job is considered failed.
• restart policyconfigured by · A restart policy is a configuration rule specifying whether and when a container or workload is automatically restarted after termination or failure.
• Kubernetes controllercontrasts with · A Kubernetes controller is a control-loop component that observes resource state and takes action to move actual state toward desired state.
• work queue patternuses · The work queue pattern stores pending units of work for workers or processors to retrieve and process, often asynchronously.

This explainer was generated by Manning's knowledge graph from the text of Machine Learning Platform Engineering. The section links point to the parts of the book it draws on.