Say where the whole failure is, not just the sentence about it #36

Merged

geusebio merged 1 commit from feature/failure-page-url-on-a-failed-deploy into main

2026-08-24 01:25:20 +00:00

geusebio commented

2026-08-24 01:10:05 +00:00

Owner

A failed deploy has had a page on the platform for as long as failures have been recorded: the phase that gave up, whose fault it was (FailurePhase::remedyHint() is blunt about which side of the fence the fix is on), the build log as it stood, and a button that shares all of it for 24 hours with somebody who cannot sign in. sc never mentioned it — it printed the one sentence of statusReason and stopped — so a reader had no way to know there was anywhere else to look.

Both places sc shows a bad landing now name it.

sc deploy, hung off the sentence it exits on:

Waiting for a build worker…
[S⠿C] Queued for 2s
Building…
  api | RUN go build -trimpath -o /out/api ./cmd/api
[S⠿C] Build failed after 11s

Error: Deployment #12 failed. kaniko exited 1
Why, in full:  https://someones.computer/failures/F-24GT1BQ7 (F-24GT1BQ7)

sc apps, under the row:

acme  (owner)
└── web
    ├── main  failed (#1)
    │   Deployment #1 failed. kaniko exited 1.
    │   F-24GT1BQ7 — https://someones.computer/failures/F-24GT1BQ7
    └── staging  running (#2)

deploy keeps the failure frame the progress stream sends just ahead of the end frame, and returns it with the sentence as one error — so the whole answer goes to stderr with the non-zero exit, rather than half of it landing on stdout.
apps prints it after the reason and independently of it: a revision can fail with nothing written against it, and the page is then the only account there is.
A platform that sends no failure — older than the frame, or a revision that did not land badly enough to have one — prints exactly what it printed before.

Bumps scclient to the commit carrying client.Failure (Grey.ooo/someones.computer_scclient#12).

Verification

go test ./... green; hack/lint.sh 0 issues; go vet clean.
hack/coverage.sh: internal/cli 85.8% against a floor of 80.7%; every other package at or above its floor.
Both renderings above are asserted as exact expected output in apps_test.go and deploy_progress_test.go, including the negative case (no frame → unchanged output).

The platform side that sends these fields is a separate PR on Grey.ooo/Someones.Computer; this half degrades cleanly against a platform without it.

A failed deploy has had a page on the platform for as long as failures have been recorded: the phase that gave up, whose fault it was (`FailurePhase::remedyHint()` is blunt about which side of the fence the fix is on), the build log as it stood, and a button that shares all of it for 24 hours with somebody who cannot sign in. `sc` never mentioned it — it printed the one sentence of `statusReason` and stopped — so a reader had no way to know there was anywhere else to look. Both places `sc` shows a bad landing now name it. **`sc deploy`**, hung off the sentence it exits on: ```text Waiting for a build worker… [S⠿C] Queued for 2s Building… api | RUN go build -trimpath -o /out/api ./cmd/api [S⠿C] Build failed after 11s Error: Deployment #12 failed. kaniko exited 1 Why, in full: https://someones.computer/failures/F-24GT1BQ7 (F-24GT1BQ7) ``` **`sc apps`**, under the row: ```text acme (owner) └── web ├── main failed (#1) │ Deployment #1 failed. kaniko exited 1. │ F-24GT1BQ7 — https://someones.computer/failures/F-24GT1BQ7 └── staging running (#2) ``` - **`deploy`** keeps the `failure` frame the progress stream sends just ahead of the end frame, and returns it *with* the sentence as one error — so the whole answer goes to stderr with the non-zero exit, rather than half of it landing on stdout. - **`apps`** prints it after the reason and independently of it: a revision can fail with nothing written against it, and the page is then the only account there is. - **A platform that sends no failure** — older than the frame, or a revision that did not land badly enough to have one — prints exactly what it printed before. Bumps scclient to the commit carrying `client.Failure` (Grey.ooo/someones.computer_scclient#12). ## Verification - `go test ./...` green; `hack/lint.sh` 0 issues; `go vet` clean. - `hack/coverage.sh`: `internal/cli` 85.8% against a floor of 80.7%; every other package at or above its floor. - Both renderings above are asserted as exact expected output in `apps_test.go` and `deploy_progress_test.go`, including the negative case (no frame → unchanged output). The platform side that sends these fields is a separate PR on `Grey.ooo/Someones.Computer`; this half degrades cleanly against a platform without it.

geusebio added 1 commit

2026-08-24 01:10:05 +00:00

Say where the whole failure is, not just the sentence about it

CI / build (pull_request) Successful in 4m31s

Details

635c543fd3

A failed deploy has had a page on the platform for as long as failures have
been recorded — the phase that gave up, whose fault it was, the build log as
it stood, and a button that shares all of it for a day with somebody who
cannot sign in. `sc` never mentioned it. It printed the one sentence of
`statusReason` and stopped, so a reader had no way to know there was
anywhere else to look, and the page went unread.

Both places `sc` shows a bad landing now name it:

    Deployment #12 failed. kaniko exited 1
    Why, in full:  https://someones.computer/failures/F-24GT1BQ7 (F-24GT1BQ7)

    └── web
        ├── main  failed (#1)
        │   Deployment #1 failed. kaniko exited 1.
        │   F-24GT1BQ7 — https://someones.computer/failures/F-24GT1BQ7

- `deploy` keeps the `failure` frame the progress stream sends just ahead of
  the end frame, and hangs it off the sentence it exits on — one error, so
  the whole answer lands on stderr with the non-zero exit rather than half
  of it on stdout.
- `apps` prints it under the row, after the reason and independent of it: a
  revision can fail with nothing written against it, and the page is then
  the only account there is.

A platform that sends no failure — one older than the frame, or a revision
that did not land badly enough to have one — prints exactly what it printed
before.

Bumps scclient to the commit carrying `client.Failure`.

geusebio changed title from ~~WIP: Say where the whole failure is, not just the sentence about it~~ to Say where the whole failure is, not just the sentence about it

2026-08-24 01:22:37 +00:00

geusebio merged commit 3a9f0cbafc into main

2026-08-24 01:25:20 +00:00

geusebio referenced this pull request from a commit

2026-08-24 01:25:20 +00:00

Merge pull request 'Say where the whole failure is, not just the sentence about it' (#36) from feature/failure-page-url-on-a-failed-deploy into main

geusebio referenced this pull request from a commit

2026-08-24 01:26:11 +00:00

Bump sc and scclient to the halves that read the failure link