TestLangServer_workspaceExecuteCommand_modules_multiple fails on aarch64

[06kellyjac]

Server Version

v0.18.0

Terraform Version

None

Client Version

None

Terraform Configuration Files

None

Log Output

Gist contains: Just the error section, full aarch64 failing build, and a full x86 passing build for comparison

https://gist.github.com/06kellyjac/a5699a23c1745911eaf981a28c14a467

Expected Behavior

Tests should pass

Actual Behavior

TestLangServer_workspaceExecuteCommand_modules_multiple fails but only on aarch64

Steps to Reproduce

Either run the tests normally from this repo on an aarch64 machine (go test ./... or go test github.com/hashicorp/terraform-ls/internal/langserver/handlers)
Or clone my nixpkgs repo and use branch terraform-ls on an aarch64 machine and run nix-build -A terraform-ls

[radeksimko]

Hi @06kellyjac
Thanks for the report.

Unfortunately it seems that GitHub does not provide any ARM environment to run the tests in currently: https://docs.github.com/en/actions/using-github-hosted-runners/about-github-hosted-runners#supported-runners-and-hardware-resources which seems to be mostly blocked on lack of ARM support in Azure per actions/runner-images#2552 (comment)

We could in theory self host the ARM runners as per https://github.blog/changelog/2019-12-03-github-actions-self-hosted-runners-on-arm-architectures/ but that is obviously not as trivial (from ops perspective) as using GitHub-hosted runners.

I'm asking around internally (at HashiCorp) about that possibility, but generally until we have a way of testing this in CI it's unlikely that we'll find time to fix it.

That said if you can come up with a PR that fixes the test I'd be happy to review it!

[06kellyjac]

Might be able to set up the tests in actions using QEMU like deno does using this: https://github.com/denoland/rusty_v8/blob/main/.github/workflows/ci.yml#L81-L96

I imagine it'll need some adapting to work for golang rather than rust but I might play with it

[radeksimko]

I had a spare moment today to reproduce this on my Raspberry Pi and I have a hunch of what the root cause is.

I believe it's merely test data being limited to 3 combinations of OS & architecture:

https://github.com/hashicorp/terraform-ls/tree/main/internal/terraform/module/testdata/main-module-multienv/env/dev/.terraform/plugins

I reckon the best way forward would be to turn this test into more E2E test which actually calls Terraform and runs init on whatever platform the test is running on and creates the cache relevant for that platform, instead of maintaining cache in-tree.

[github-actions]

I'm going to lock this issue because it has been closed for 30 days ⏳. This helps our maintainers find and focus on the active issues.
If you have found a problem that seems similar to this, please open a new issue and complete the issue template so we can capture all the details necessary to investigate further.