Bug: Stable release becomes outdated very quickly due to hard coding

[corneliusroemer]

What happened?

You appear to set a stableHelmVersion in run.ts. That's problematic because the stable version changes frequently so this action quickly ends up with an outdated stable version.

setup-helm/src/run.ts

Line 13 in df342a2

const stableHelmVersion = 'v3.13.3'

Currently, the "stable" version points at 3.13.3 which has not been the latest stable for 1.5 years (since January 2024): https://github.com/helm/helm/releases/tag/v3.14.0

You should probably call it defaultHelmVersion as stable is misleading.

Also, this version should be updated with every release to not have too much of a gap between actual upstream stable and this action's "stable".

Using old versions means that newly introduced syntax will not work, e.g. --skip-schema-validation, see #187

Version

• I am using the latest version

Runner

default

Relevant log output

NA

[corneliusroemer]

You probably shouldn't use that "stable" version as a hardcoded string in tests as well - that means updating it causes test failure...

setup-helm/src/run.test.ts

Line 91 in df342a2

expect(await run.getLatestHelmVersion()).toBe('v3.13.3')

[davidgamero]

the linked file mentions:

test('getLatestHelmVersion() - return the stable version of HELM when simulating a network error', async () => {

the variable linked is only used if the current stable version cannot be detected with a network call

the name "stableHelmVersion" could be ambiguous as you mentioned, and a PR would be welcome to fix it 😄

[thomastvedt]

This got worse when Helm 4 shipped, and it is worth separating from the staleness issue.

When this was filed, the hard-coded fallback kept users inside Helm 3, so the cost was a missing flag. Now latest resolves to Helm 4, and the same fallback moves a job back a whole major on a single failed request. The step stays green and the run fails later on something unrelated-looking, in our case Error: unknown flag: --rollback-on-failure.

I don't think we should fallback to another version. If fetching the latest version fails, then just fail? Or retry a few times first, but then fail with the actual error?