This page describes how to configure a Job. You can find a full sample configuration here.
A Job creates one or more tasks during its lifecycle. Each task corresponds to a single parallel and retry index for the Job.
The Job needs a template which describes how to create tasks. You can specify the task template under
.spec.taskTemplate, which supports the following fields (exactly one must be specified):
pod: Create the task as a Pod. This template is a
Currently, only the
pod task executor is supported.
For more details on the task executor interface, see Task Executors.
Running Adhoc Jobs
A Job can be created and started automatically by the JobConfig controller, but it can also be started on an adhoc basis.
Jobs created from a JobConfig will inherit its template, are subject to the JobConfig's concurrency policies, and can utilize the Job Options defined in the JobConfig's spec.
For more details, see Adhoc Execution.
When a Job is created, it may not necessarily start executing right away. This is defined by the
For more details, see Start Policy.
Timeouts and Retries
Furiko provides several mechanisms to impose timeouts and retrying of failed Jobs.
For more details see Timeouts and Retries.
Furiko exposes a
substitutions field, so that the JobController knows how to substitute context variables into the task template at runtime. This field is also used when creating a Job from a JobConfig, for example via Job Options.
All substitutions should be a key-value map of strings:
option.user-name: "John Smith"
For more details, see Context Variables.
Jobs should not live for too long after they are finished. Furiko imposes a time-to-live (TTL) on all Jobs after they are finished, so that the Kubernetes will not be too overloaded and controllers will be more responsive in general.
For more details see Garbage Collection.