Batch job workloads

What is the lifecycle of a batch job?

When you submit a batch job, it runs to completion. Typically, batch jobs retrieve input data, do computational work, and store the results in persistent data stores. When the batch job is completed, resources that are used to run the job are removed and no cost is incurred for any stand-by resources.

Isolation

Code Engine is a multi-tenant, regional service where tenants share the same network and compute infrastructure. In particular, the network and compute infrastructure are shared resources and some management components are common to all tenants. However, tenants and their workloads are isolated from each other by using Code Engine projects. Code Engine prevents communication between projects, providing isolation to your applications inside a multi-tenant environment. In addition, there are access controls that are performed on a resource level to allow only authorized users to perform certain operations on project resources, such as applications or other Code Engine workloads.

For more information about workload isolation, see Code Engine workload isolation.

Logging

When you work with Code Engine jobs in the console with logging enabled, logs are forwarded to an IBM Log Analysis service instance that is associated with your Code Engine project. In the Log Analysis service instance, the logs are indexed, enabling full-text search through all generated messages and convenient querying based on specific fields. See Getting logs for jobs.

System event information can also be helpful to troubleshoot problems when you run jobs. You can view system event information with the CLI. See Getting system event information for jobs.

For more information, see Viewing logs.

Queuing

Submitted batch jobs are automatically queued and are in a pending state until dispatched. Batch jobs are run based on the available compute resources as defined by your Code Engine project. Batch jobs that are in a pending status can be removed from the queue. You can monitor batch jobs that are pending, running, or completed by using the Code Engine console or command-line interface.

Retries

Each job instance runs to completion. However, the workload code might encounter an error during the job run. When a job instance completes with a nonzero return code, Code Engine restarts the job instance. With Code Engine, you can specify to limit the number of retries to avoid restarting failed job instances.

Running batch jobs

Whether your code exists as source in a local file or in a Git repository, or your code is a container image that exists in a public or private registry, Code Engine provides a streamlined way for you to run your code as a job.

You can create and run batch jobs in Code Engine in the following ways:

• Run an existing container image. Create a job and provide a reference to your image to use when you submit the job. For an example, see Create and run a job.
• Start with source code. If you are starting with source code that is located in a Git repository or on your local workstation, you can point to the location of your source and Code Engine takes care of building the image for you. See Create a job from repository source code and Create a job from local source code.

For more information, see Working with jobs.

Scaling

A job in Code Engine (batch job) consists of one or more job instances. While job instances run independent of each other, they run the same code. Suppose you have a database with 100 records to analyze. You can run your job such that each job instance analyzes 10 records each. For example, the first job instance analyzes records 0 - 9, the second job instance can analyze records 10 - 19, and so on. In this example, each job instance receives the system provided job input parameter JOB_INDEX as an environment variable that you can use to calculate the range of database records to analyze by each job instance. The number of job instances is specified when the batch job is submitted.

When you run a batch job, during the startup time, job instances might linger in a pending state as the instances are being started. Pending times for job instances can vary due to the system and network infrastructure, as the system adjusts to fulfill the resource demands related to this run of the job. If the system infrastructure is responding to a high usage demand, the result might be a multi-minute delay of the startup for jobs because job instances are provisioned just-in-time.

Status

A batch job is in the Succeeded status when all job run instances are completed.

A batch job is in Failed status after one or more job run instances reaches the retry limit. Also, if a job takes too long to complete, the job is in Failed status whenever the maximum execution time is reached. For more information, see job status.

Submitting similar batch jobs

When you create a job, you can specify workload configuration information that is used each time that the job is run. When you use a common set of configuration information for batch jobs, with Code Engine, you can specify additional parameters with the job submission to overwrite the batch job parameters for this specific job submission.

Triggering batch jobs with events

Batch jobs can be submitted automatically based on events, such as periodic timers, Object Storage events, or Kafka topics.

Suppose you want to trigger your batch job automatically based on a subscription to an IBM Cloud Object Storage bucket that generates events whenever a file changes or is added to the Object Storage bucket. Review the Code Engine cos-event-job sample for information about how to create Object Storage events and use the events to trigger batch jobs.

For more information about using eventing with Code Engine jobs, see the following topics: Working with the Periodic timer (cron) event producer, Working with the IBM Cloud Object Storage event producer, and Working with the Kafka event producer.

Looking for more code examples? Check out the Samples for IBM Cloud Code Engine GitHub repo.