No error shown when workspace fails to start due to supervisor error

[WVerlaek]

Bug description

When a workspace fails to start due to supervisor exiting with an error (e.g. failing to setup the gitpod user in a custom image), the workspace immediately transitions to stopping and stopped, but doesn't show any error to the user:

Related: #12978 (comment) (has some more context as well)

Steps to reproduce

• Start workspace with a custom base image that doesn't have the gitpod user (e.g. hashicorp/terraform:1.2.9)
• supervisor fails to add the gitpod user and exits
• Workspace transitions to stopping -> stopped states without showing an error

Workspace affected

No response

Expected behavior

An error is shown to users if a workspace fails to start due to supervisor exiting with an error. If the error is for a missing gitpod user, the error message sent to the user should include a link to our docs on custom base images.

Example repository

No response

Anything else?

No response

[kylos101]

@akosyakov would you like Team IDE to own this one,or, is it one of those scenarios where maybe you'd prefer Team Workspace contribute to Supervisor? cc: @Furisto @utam0k @WVerlaek

[akosyakov]

@kylos101 if it is urgent for the workspace team because of PVC go ahead and create a PR 🙏

[kylos101]

👋 @akosyakov not urgent because of PVC (even when using object storage, the user gets an error message that is not helpful, so both failure modes are roughly the same). If I were to ask IDE team to do, could you estimate when you may be able to ingest as part of groundwork? That will help us know if we should do or not.

[kylos101]

Updated expected behavior to include a link if the gitpod user is missing.

[akosyakov]

We will consider it next Monday.

[kylos101]

For posterity, @WVerlaek is going to raise a PR, reference.

[WVerlaek]

For posterity, @WVerlaek is going to raise a PR, #12978 (comment).

Note that PR will solve a different issue that what's described here (it will fix a small bug in adduser for busybox-type images)

[kylos101]

Doh! Thank you, @WVerlaek .

Does this mean we still need help from IDE team? I ask because it's still in their inbox, and @akosyakov shared they were going to pick it up next week.

[WVerlaek]

@kylos101 yes if possible 🙏

Just raised the adduser fix here: #14438

[kylos101]

Perfect, thanks for clarifying @WVerlaek ! 💪 @akosyakov please let @WVerlaek know if you bump into any trouble or need help testing.

[kylos101]

Hi @mustard-mh , I tested #14650 with https://gitlab.com/kylos101/prebuild-experiment/-/blob/kylos101/customer-image/.gitpod.yml#L1, which should cause an error due to the gitpod user missing, but, the error message only mentions the exit status is 1. I was expecting a better error message to the user, for example, a message sharing that the gitpod user is missing. In other words, an exit status of 1 does not tell me what to fix so that my workspace can start.

Is that possible to change? Here is what I see now when trying to start a workspace from this branch:

@WVerlaek what sort of error message were you expecting to see? The workspace I tested with is https://kylos101-prebuildexperi-3ui0u5wx2ib.ws-us75.gitpod.io/.

[kylos101]

cc: @akosyakov reopened, and added back to inbox

[WVerlaek]

@kylos101 in that specific example (a terraform workspace image) the bug in #14438 gets triggered which fails to add the gitpod user, if that issue gets fixed it shouldn't error on a missing gitpod user anymore. supervisor will still fail to start the terraform-based workspace though once fixed, as it's missing some other required tools as well (it's an alpine image, which was mentioned in the issue that we don't really support)

[WVerlaek]

Maybe on workspace startup failure, if it's an alpine-based image we can somehow show that it's not something we support currently?

[kylos101]

Maybe on workspace startup failure, if it's an alpine-based image we can somehow show that it's not something we support currently?

@WVerlaek Can you create that as a separate issue for supervisor and inbox to IDE team for consideration?

[kylos101]

@kylos101 in that specific example (a terraform workspace image) the bug in #14438 gets triggered which fails to add the gitpod user, if that issue gets fixed it shouldn't error on a missing gitpod user anymore. supervisor will still fail to start the terraform-based workspace though once fixed, as it's missing some other required tools as well (it's an alpine image, which was mentioned in the issue that we don't really support)

@WVerlaek whoops, thank you! For this issue, could you update the description? For example, Steps to recreate should avoid using an alpine image as the base image.

@mustard-mh what happens if you try using a custom image that is not alpine based? For example, is the error exit status 1, or does the error message indicate something else - like the gitpod user is missing?

[mustard-mh]

@mustard-mh what happens if you try using a custom image that is not alpine based? For example, is the error exit status 1, or does the error message indicate something else - like the gitpod user is missing?

If you use custom image that is not alpine based, it will success, because even you don't have gitpod user, supervisor will create it, but it still can failed by some reason, e.g missing useradd or adduser

We will continue improve this error message in following PR