Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

`vp run`: an option to keep going after a task fails (today one failure kills unrelated running tasks and their results are lost)

Aperta
  • Qualcosa non va?
#822 0 commenti 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
4/5
Tempo stimato
3-5 giorni
Idoneità per principianti
35/100
Tipo di issue
Funzionalità
Chiarezza
Abbastanza chiara
Stato di attività
Attiva
Stack tecnologico
cli, rust
Ambito
cli, devtools

Direzione di ricerca

Start in crates/vt/src/session/execute/scheduler.rs, where a failure cancels the shared token (around lines 229-231), and in cache_update.rs, where cancelled runs are skipped for caching (around lines 48-49). A --continue flag would need to stop that cancellation, skip only tasks that depend on a failure, and set a non-zero exit code at the end. Done means a fixture with independent failing and passing packages finishes the passing ones, caches them, and exits non-zero. The issue is a feature with a design choice (which continue modes to support), so confirm the scope with maintainers before starting.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

Summary

When a task fails, vp run cancels the whole run. Tasks that are running and do not depend on the failed task are killed, tasks not yet started never run, and none of them gets a cache entry. There is no option to change this.

This asks for one: keep running every task whose dependencies succeeded, skip only the tasks that depend on a failure, and exit non-zero at the end.

What happens today

A workspace of four packages with no dependencies between them. a fails after 2 s; b, c and d take 8 s.

// packages/a/package.json
{ "name": "a", "scripts": { "check": "node -e \"setTimeout(() => process.exit(1), 2000)\"" } }
// packages/b/package.json (c and d are the same)
{ "name": "b", "scripts": { "check": "node -e \"setTimeout(() => console.log(1), 8000)\"" } }

$ vp run --cache -r check
vp run: 0/4 cache hit (0%), 4 failed.

$ vp run --last-details
  [1] a#check ... ✗ (exit code: 1)
  [2] c#check ... ✗ (exit code: 137)
  [3] d#check ... ✗ (exit code: 137)
  [4] b#check ... ✗ (exit code: 137)

The run ends after about 2 s. b, c and d were healthy and independent of a. They are reported as failed, and the next run starts them from nothing. (vp v1.0.0, macOS.)

In the source this is unconditional. The scheduler cancels the shared token on any failure (scheduler.rs#L229-L231), and a run cancelled that way is not cached (cache_update.rs#L48-L49).

Why it matters
  • One failure per run. In a recursive test or lint over many packages, the first failure hides every other one. A change that breaks three packages takes three runs to fix, and in CI each run is a round trip.
  • A flake costs the whole run. One test that times out kills every suite still running. Their results are lost, so the retry re-executes work that was about to pass.
  • The killed tasks read as failures. They show exit code: 137 beside the real failure and the summary counts them as failed, so the reader has to work out which one was the cause.
  • The workaround is expensive. To get a result per package, a CI job has to start one vp run <package>#<task> process per package. Those processes then replay the same dependency tasks at the same time, and a cache hit restores a task's outputs over the existing files while another process is reading them. So the job also has to run the shared dependencies once up front and pass --ignore-depends-on to every process.
Prior art
Tool Option Its documentation
GNU Make -k, --keep-going "Keep going when some targets can't be made."
Turborepo turbo run --continue[=never|dependencies-successful|always] Default never. With dependencies-successful: "turbo will cancel dependent tasks. Tasks whose dependencies have succeeded will continue to run."
pnpm pnpm run --no-bail "Continue running the remaining matched scripts even if one of them fails. The command still exits with a non-zero exit code if any script failed."
Nx --nxBail "Stop command execution after the first failed task."
Cargo cargo test --no-fail-fast, cargo build --keep-going "Run all tests regardless of failure"; "Do not abort the build as soon as there is an error"
Proposal
vp run --continue [TASK_SPECIFIER]
  • A task that fails does not cancel the run.
  • A task whose dependencies all succeeded still runs, and its result is cached as in any other run.
  • A task that depends on a failed task, directly or through others, does not run. It is reported as skipped, naming the failure it waited on. This is Turborepo's dependencies-successful.
  • The run exits non-zero if any task failed, after every runnable task has finished.
  • Ctrl-C still cancels everything.

With the example above it would print:

$ vp run --cache --continue -r check
vp run: 0/4 cache hit (0%), 1 failed.

$ vp run --cache --continue -r check
vp run: 3/4 cache hit (75%), 1 failed.

A config key for the default would let a workspace choose it for CI. The flag alone is enough to start.

Non-goals
  • Changing the default.
  • Running the dependents of a failed task (Turborepo's always).
  • Retrying a failed task.
How we hit it

A monorepo of about 80 packages that runs vp run -r test in CI. One test timed out by 10 ms, and three healthy suites that were still running were killed with exit 137. None of the three stored a result, so the retry ran all four again, one of them with 1,282 tests. We now run one process per package to keep each verdict, at the cost described above.

Lingua principale
Rust
Stelle
473
Fork
44
Merge medio
1g 19m
PR unite (30g)
55

Preparare l'ambiente

Apri in Codespaces

Avvia il container di sviluppo del progetto nel browser, con il tuo account GitHub.

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di voidzero-dev/vite-task

Tutte le issue di voidzero-dev/vite-task

Issue simili

Altre issue su Rust

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.