Link inputs and outputs without using volumes #617
Comments
bobcatfish
added
help wanted
|
One possible solution is to detect when this is happening, and schedule pods for Tasks with linked inputs/outputs onto the same nodes so they can share data directly on disk. This would be a bit of a scheduling challenge; we may need to look into using a custom scheduler. |
|
Hmm, I was sort of hoping that the k8s scheduler would be smart enough to allow us to schedule subsequent tasks on the same node as the previous one when there's a shared volume (in cases where this is possible). Maybe this is true and we just need to use the scheduler differently? If not a custom scheduler should do the trick. |
|
Not sure about this one, probably because caching (read persistence) is a core concept of ci/cd architectures. But did you looked at emptydir ? |
|
Issues go stale after 90d of inactivity. /lifecycle stale Send feedback to tektoncd/plumbing. |
|
Stale issues rot after 30d of inactivity. /lifecycle rotten Send feedback to tektoncd/plumbing. |
|
Rotten issues close after 30d of inactivity. /close Send feedback to tektoncd/plumbing. |
tekton-robot
added
lifecycle/stale
|
@tekton-robot: Closing this issue. In response to this:
Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository. |
|
These days the affinity assistant helps us a lot with this; also we are still trying to determine the fate of pipelineresources #1673 |
The testbuilder `EventListenerTriggerBinding` allows users to specify both `name` and `ref` fields. With #617, specifying both `name` and `ref` is no longer allowed. We fixed the builder usage for all the happy path tests to use `name` as an empty string but did not do it for tests that are verifying error paths (such as validation errors). So, these tests will now all fail at the same validation check making most of these tests useless (causing the drop in coverage in To fix, I changed the builder to only accept `ref` so that one cannot create invalid structs. Later, we can migrate our tests to not use builders at all and have our error tests verify the actual error string. Signed-off-by: Dibyo Mukherjee <dibyo@google.com>
This commit adds support for a more concise syntax for TriggerBindings embedded
in a Trigger or EventListener:
NEW SYNTAX:
```
bindings:
- ref: some-name
- name: commit_id # embedded binding
value: "$(body.head_commit_id)
```
OLD SYNTAX
```
bindings:
- name: some-name
spec:
params:
- name: commit_id # embedded binding
value: "$(body.head_commit_id)
```
The old embedded syntax is deprecated and Triggers will convert any old style
embedded bindings to the new syntax via the MutatingAdmissionWebhook.
BREAKING CHANGE: With this change, users cannot specify both `name` and `ref`
for a single binding. Use `ref` to refer to a TriggerBinding resource and
`name` for embedded bindings.
Fixes #617
Signed-off-by: Dibyo Mukherjee <dibyo@google.com>
Expected Behavior
When a user wants to link Pipeline Task outputs and inputs, this should:
Actual Behavior
There are currently two options for how outputs and inputs are linked:
Additional Info
The text was updated successfully, but these errors were encountered: