che
che copied to clipboard
eclipse-che
Git commit + push on workspace shutdown to backup unsaved workspace project changes
Is your enhancement related to a problem? Please describe
Using ephemeral storage can often provide faster workspace startup time (as storage does not have to be provisioned). However, there is the inherent disadvantage that any code changes to the project that the user is working on will not be persisted upon workspace shutdown. The user needs to commit and push their changes with git in order to retrieve them later on.
Describe the solution you'd like
It'd be nice to have a spec.devEnvironments.gitBackup: <bool> configuration option in the Che Cluster CR, that when enabled, will automatically commit all changes to the project in use, and push them to a unique branch, e.g. che-backup-<checked-out-branch-name>-YYYY-MM-DD-HH-MM-SS.
This could be implemented as a pre-stop event for workspace pods at the DevWorkspace Operator level.
There are some considerations that need to be made for this feature. Some that come to mind:
- A project can have multiple remotes, such as
originandfork. Since these remote names may be different on a per-devfile-basis, setting the default remote to "backup" to may be tricky. We may need something at the devfile or devworkspace level to specify which remote should be used for backups. - Using this feature should not prevent the use of other devfile
preStopevents (though these are currently unsupported by DWO/Che) - Enabling this feature should probably not be allowed when persistent workspace storage is used, as it will create an unnecessary commit of the user's work (which can be frustrating for keeping an organized git log) without providing any advantages (their changes would persist regardless of this feature).
- This feature should have no effect if the project being used is not a git project
Describe alternatives you've considered
As preStop events are not currently supported by DWO, there's no way to manually configure a git commit and push on workspace shutdown from a devfile. Users using ephemeral storage must remember to always manually commit and push their workspaces changes before their workspace shuts down.
Additional context
No response
@ibuziuk @l0rd @amisevsk CC'ing you all in case of any thoughts/input on this feature.
Looks good to me. I couple of comments:
- This is something that developer should be able to configure at single workspace level (as the ephemeral storage config).
- I would start with using
originas the remote (nooriginor no write privileges onoriginthen no backup is done) - The prefix of the backup branch name should not have hardcoded "che" in it, ideally it should be configurable
- You are not describing what happens when the developer restart the workspace: we should either automatically clone the last backup branch or ask the user (I prefer the first option as a first iteration as it looks the simplest to implement)
If we want to restore from a backup branch when possible, we should probably choose a deterministic name for the branch (e.g. <workspaceID>-<suffix> for some reasonable suffix). Otherwise, I don't know how we would automatically clone from that branch.
For determining the remote to check, we already determine which remote to clone from, so we could check if that branch exists and clone that instead of what's selected in the devworkspace, though this might cause confusion ("I've been pushing to branch X but now branch Y is checked out")
From a first pass, it looks like the configurable options we'd need to consider would be
- Enable/disable feature (default: disabled)
- Enable/disable restore from branch on restart (default: enabled?)
- (Maybe) configure branch name format.
- (Maybe) override remote to push to.
In addition, we could consider whether we want to allow configuration at a workspace level or a per-project level (or both). Attributes would be a good candidate for this since they allow structured data rather than just strings.
Implementation wise, we could potentially do everything required within DWO/project-clone itself, though if we implement it as a preStop hook we might have to consider collisions with preStop events, if we ever decide to implement those. Currently DWO will provision a preStop event handler when async storage is used, but this might be extended.
@l0rd @amisevsk Thank you both for the thoughtful input.
In addition, we could consider whether we want to allow configuration at a workspace level or a per-project level (or both). Attributes would be a good candidate for this since they allow structured data rather than just strings.
Ideally, I'd like to support configuring this at the workspace and project level. If we only support this on a per-project basis, then users who want to use this feature for all workspaces will have to copy/paste the attribute for all projects. If we only support it on a workspace basis, then users don't have the granularity to only use it for the project they are actively working on. However, for a first-iteration, I think configuration on a per-project basis would be fine.
Implementation wise, we could potentially do everything required within DWO/project-clone itself, though if we implement it as a preStop hook we might have to consider collisions with preStop events, if we ever decide to implement those. Currently DWO will provision a preStop event handler when async storage is used, but this might be extended.
I'm considering trying to implement support for multiple preStop events as a precursor to this feature. This would lay a foundation for this feature to build upon without colliding with the async storage preStop event handler (or normal preStop events). Additionally, supporting preStop events would probably allow for rough testing of this feature by configuring a devfile/devworkspace accordingly.
A project that may be interesting for this https://github.com/tkellogg/dura
This seems really cool & relevant to this feature. I think I should spend some time experimenting with dura to make sure it'll work nicely for our use case. My concern is ensuring the main workspace container image has the dura binary installed. However, the same concern exists for git: users could use a workspace container image that doesn't have git installed - so this concern might be outside our control.
Issues go stale after 180 days of inactivity. lifecycle/stale issues rot after an additional 7 days of inactivity and eventually close.
Mark the issue as fresh with /remove-lifecycle stale in a new comment.
If this issue is safe to close now please do so.
Moderators: Add lifecycle/frozen label to avoid stale mode.
@tolusha I'm actually (slowly) still working on this issue, though it's low priority since no one has actively requested for it.