Skip to content

Latest commit

 

History

History
98 lines (76 loc) · 6.22 KB

resources.md

File metadata and controls

98 lines (76 loc) · 6.22 KB

PipelineResources

⚠️ PipelineResources are deprecated.

Consider using replacement features instead. Read more in documentation and TEP-0074.

Note: PipelineResources have not been promoted to Beta in tandem with Pipeline's other CRDs. This means that the level of support for PipelineResources remains Alpha. PipelineResources are now deprecated.**

For Beta-supported alternatives to PipelineResources see the v1alpha1 to v1beta1 migration guide which lists each PipelineResource type and a suggested option for replacing it.

For more information on why PipelineResources are remaining alpha see the description of their problems, along with next steps, below.


Why Aren't PipelineResources in Beta?

The short answer is that they're not ready to be given a Beta level of support by Tekton's developers. The long answer is, well, longer:

  • Their behaviour can be opaque.

    They're implemented as a mixture of injected Task Steps, volume configuration and type-specific code in Tekton Pipeline's controller. This means errors from PipelineResources can manifest in quite a few different ways and it's not always obvious whether an error directly relates to PipelineResource behaviour. This problem is compounded by the fact that, while our docs explain each Resource type's "happy path", there never seems to be enough info available to explain error cases sufficiently.

  • When they fail they're difficult to debug.

    Several PipelineResources inject their own Steps before a Task's Steps. It's extremely difficult to manually insert Steps before them to inspect the state of a container before they run.

  • There aren't enough of them.

    The six types of existing PipelineResources only cover a tiny subset of the possible systems and side-effects we want to support with Tekton Pipelines.

  • Adding extensibility to them makes them really similar to Tasks:

    • User-definable Steps? This is what Tasks provide.
    • User-definable params? Tasks already have these.
    • User-definable "resource results"? Tasks have Task Results.
    • Sharing data between Tasks using PVCs? workspaces provide this for Tasks.
  • They make Tasks less reusable.

    • A Task has to choose the type of PipelineResource it will accept.
    • If a Task accepts a git PipelineResource then it's not able to accept a gcs PipelineResource from a TaskRun or PipelineRun even though both the git and gcs PipelineResources fetch files. They should technically be interchangeable: all they do is write files from somewhere remote onto disk. Yet with the existing PipelineResources implementation they aren't interchangeable.

They also present challenges from a documentation perspective:

  • Their purpose is ambiguous and it's difficult to articulate what the CRD is precisely for.
  • Four of the types interact with external systems (git, pull-request, gcs, gcs-build).
  • Five of them write files to a Task's disk (git, pull-request, gcs, gcs-build, cluster).
  • One tells the Pipelines controller to emit CloudEvents to a specific endpoint (cloudEvent).
  • One writes config to disk for a Task to use (cluster).
  • One writes a digest in one Task and then reads it back in another Task (image).
  • Perhaps the one thing you can say about the PipelineResource CRD is that it can create side-effects for your Tasks.

What's still missing

So what are PipelineResources still good for? We think we've identified some of the most important things:

  1. You can augment Task-only workflows with PipelineResources that, without them, can only be done with Pipelines.
    • For example, let's say you want to checkout a git repo for your Task to test. You have two options. First, you could use a git PipelineResource and add it directly to your test Task. Second, you could write a Pipeline that has a git-clone Task which checks out the code onto a PersistentVolumeClaim workspace and then passes that PVC workspace to your test Task. For a lot of users the second workflow is totally acceptable but for others it isn't. Some of the most notable reasons we've heard are:
      • Some users simply cannot allocate storage on their platform, meaning PersistentVolumeClaims are out of the question.
      • Expanding a single Task workflow into a Pipeline is labor-intensive and feels unnecessary.
  2. Despite being difficult to explain the whole CRD clearly each individual type is relatively easy to explain.
    • For example, users can build a pretty good "hunch" for what a git PipelineResource is without really reading any docs.
  3. Configuring CloudEvents to be emitted by the Tekton Pipelines controller.
    • Work is ongoing to get notifications support into the Pipelines controller which should hopefully be able to replace the cloudEvents PipelineResource.

For each of these there is some amount of ongoing work or discussion. We have deprecated PipelineResources and are actively working to cover the missing replacement features. Read more about the deprecation in TEP-0074.

For Beta-supported alternatives to PipelineResources see the v1alpha1 to v1beta1 migration guide which lists each PipelineResource type and a suggested option for replacing it.

Except as otherwise noted, the content of this page is licensed under the Creative Commons Attribution 4.0 License, and code samples are licensed under the Apache 2.0 License.