✅ Deploy Preview for nebi-docs canceled.

Tested on the hetzner cluster. After restarting the JLab pod with sha-e81f609, created test11 and the path is correctly persisted:

test11 | /home/jovyan/.local/share/nebi/workspaces/test11-1a0a2608-70e7-4f20-88ab-e71772d658ba | ready

Earlier workspaces (test8, test10) still show empty path since they were created before the fix — expected.

@aktech the regression test is in and CI is green. Mind taking another look?

I started reviewing this but wanted to reproduce the error and found a new issue that prevented me from doing so. The linux nebi-desktop app crashes on startup. #350

Hey @Adam-D-Lewis, thanks for having a look, and good catch. I only ran this against NIC, so I wasn't aware that we needed to test this through the desktop version too. I wonder though, if testing on cloud would be enough as validation for this PR specifically, since the desktop probably would need to wait for upstream contribution or workaround to take effect first.

@viniciusdc Testing on cloud should be good enough to prove it works. I was trying to reproduce locally to be sure I understand the issue. With a recent fix, I was able to reproduce the issue today. I plan on reviewing this tomorrow. Thanks for your patience!

One concern on the error-handling branch: when SetWorkspacePath returns an error, the new code logs it but falls through — the create flow continues and SetWorkspaceStatus(ws.ID, models.WsStatusReady) still runs a few lines later. The user-observable result is identical to the bug this PR fixes: workspace shows ready in the UI but path is empty in the DB and the CLI reports "missing": true.

Suggest returning early (or transitioning to WsStatusFailed) after logging, matching the CreateWorkspace error handling four lines above:

if err := w.svc.SetWorkspacePath(ws.ID, resolvedPath); err != nil {
    w.logger.Error("failed to persist workspace path", "workspace_id", ws.ID, "resolved_path", resolvedPath, "error", err)
    w.svc.SetWorkspaceStatus(ws.ID, models.WsStatusFailed)
    return err
}
ws.Path = resolvedPath

It's a low-likelihood path in practice (the only SetWorkspacePath failure mode is the DB write itself), but the whole point of the fix is to make the "ready with empty path" state unreachable — leaving the silent-continue branch keeps the original symptom possible.

On the regression test — nice that it drives executeJob end-to-end and actually reaches db.Save(ws) via the real UpdateWorkspaceSize path. One thing worth tightening:

PackageManager: "test-pm-not-registered",

The test depends on SyncPackagesFromWorkspace and CreateVersionSnapshot failing with an unknown-manager error, and on the worker's current behavior of logging and continuing on those failures (w.logger.Error("Failed to sync packages", ...)). If either of those is later refactored to return early, fail-fast, or stop calling UpdateWorkspaceSize when the manager is invalid, this test stops covering the regression — and it does so silently, since it would still pass (the bug requires db.Save(ws) to actually run).

Two safer options:

1. Extend fakeExecutor (or add a sibling fake) to implement enough of the package-manager surface that SyncPackagesFromWorkspace/CreateVersionSnapshot succeed as no-ops on a registered manager name.
2. Use a registered manager ("pixi" or whatever the lightest one is) and inject test doubles for the two methods.

At minimum, a // NOTE: comment on the PackageManager line spelling out that the value is deliberately invalid and why the test still reaches the regression site would save the next person a debugging session.

I also have a related follow-up PR if you have time to review @viniciusdc. #353