bug(e2e): macOS script stdin EOF detaches lifecycle test sessions

Aperta Adatta ai principianti
#5,638 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 1 giorno

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
2/5
Tempo stimato
1-3 ore
Idoneità per principianti
78/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
kubernetes, rust

Direzione di ricerca

Inizia dai test E2E di sandbox_lifecycle e trova i quattro test nominati che invocano macOS script tramite la CLI. Leggi come questi test raccolgono l’output del comando, quindi esegui i test interessati su macOS usando il comando di riproduzione nell’issue. Mantieni aperta la pipe stdin di script durante la raccolta dell’output; il lavoro è completato quando l’output trasmesso viene acquisito e i codici di uscita previsti, inclusi 7 e 17, vengono mantenuti nelle esecuzioni ripetute.

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

Descrizione

state:triage-needed
User Story

I have been investigating OpenShell's E2E tests on Kubernetes. I encountered intermittent failures when running the sandbox lifecycle suite from macOS. This prevents macOS from being a trustworthy platform for validating OpenShell lifecycle behaviour.

Problem Statement

Four sandbox_lifecycle E2E tests fail intermittently or consistently when run from macOS:

  • canonical_main_nonzero_exit_preserves_status
  • canonical_main_exit_zero_completes_persistent_sandbox
  • sandbox_create_with_no_keep_preserves_failure_then_cleans_up
  • sandbox_create_with_no_keep_cleans_up_after_tty_command

The tests launch the CLI through macOS script to provide a TTY. When the test runner's stdin reaches EOF, macOS script can write the terminal's VEOF value — Ctrl-D, byte 0x04 — to the PTY. Ctrl-D detach is intentional for an interactive canonical-main attachment (#3744), but here it is injected by the harness rather than typed by a user. The supervisor interprets the byte as a detach request. The tests then observe missing streamed output and an exit status of 0 instead of the command's actual status (7 or 17 in the failing tests).

Impact / Why This Matters

These are false test failures: the sandbox command can succeed or fail with its real status, while the test sees a detached session with no output and status 0. This blocks macOS validation, forces reruns, and can hide real output-forwarding or exit-status regressions. Rerunning is not a reliable workaround because the behavior is timing-dependent.

On 2026-11-17, this reproduced on upstream main 0be63c5b0de27692df03ba0e4b3bc212312d3b3c, with matching upstream source, CLI, chart and published images on macOS + OpenShift. All four unmodified tests failed. Holding the script stdin pipe open until output collection completed made all four pass against the same revision and images.

Earlier macOS + kind runs also failed without the guard and passed with it, using digest-pinned upstream images.

Acceptance Criteria
  • The four affected lifecycle tests pass consistently when run repeatedly from macOS.
  • Each test captures the command's expected streamed output.
  • Commands with nonzero exit statuses preserve their expected exit codes (7 and 17 in the affected tests).
  • The test harness keeps script stdin open until the command completes and output collection finishes.
Reproduction Steps
  1. On macOS, check out upstream OpenShell at 0be63c5b0de27692df03ba0e4b3bc212312d3b3c. For OpenShift, apply the wrapper fixes from eviehoward/OpenShell#10. Leave the lifecycle tests unchanged for the initial run.
  2. Build the CLI:
    cargo build -p openshell-cli
    
  3. With an existing cluster context, run the four affected tests using matching upstream images:
    export OPENSHELL_E2E_KUBE_CONTEXT="$(kubectl config current-context)"
    export OPENSHELL_BIN="$PWD/target/debug/openshell"
    export OPENSHELL_E2E_KUBE_BUILD_IMAGES=0
    export OPENSHELL_REGISTRY=ghcr.io/nvidia/openshell
    export IMAGE_TAG=0be63c5b0de27692df03ba0e4b3bc212312d3b3c
    
    e2e/with-kube-gateway.sh cargo nextest run \
      --manifest-path e2e/rust/Cargo.toml --target-dir e2e/rust/target \
      --features e2e-kubernetes --test sandbox_lifecycle \
      --config-file .config/nextest.toml --profile e2e-kubernetes \
      --no-tests fail --test-threads 1 \
      -E 'test(=canonical_main_exit_zero_completes_persistent_sandbox) | test(=canonical_main_nonzero_exit_preserves_status) | test(=sandbox_create_with_no_keep_cleans_up_after_tty_command) | test(=sandbox_create_with_no_keep_preserves_failure_then_cleans_up)'
    
  4. Observe missing streamed output in the zero-exit tests and status 0 instead of 7 or 17 in the nonzero-exit tests. The recorded unmodified run failed all four tests in 134.003 seconds.
  5. In each of the four affected tests, replace cmd.output() with a spawned child whose piped stdin handle stays alive until wait_with_output() completes:
    cmd.stdin(Stdio::piped());
    let mut child = cmd.spawn().expect("spawn openshell sandbox create");
    let _stdin_guard = child.stdin.take().expect("retain script stdin");
    let output = child.wait_with_output().await.expect("wait for sandbox create");
    
  6. Repeat the same command. The recorded guarded run passed all four tests in 167.815 seconds.
Suggested UX (if applicable)

No response

Environment
  • OpenShell: upstream main 0be63c5b0de27692df03ba0e4b3bc212312d3b3c; CLI built from that revision.
  • OS: macOS, Darwin 27.2.0, ARM64, using /usr/bin/script.
  • Deployment: upstream Kubernetes E2E harness on OpenShift, two AMD64 nodes running Kubernetes v1.35.9; agent-sandbox v1.0.3.
  • Images: ghcr.io/nvidia/openshell/{gateway,supervisor,sandbox}:0be63c5b0de27692df03ba0e4b3bc212312d3b3c.
  • Resolved registry digests with Linux AMD64 selection: gateway sha256:c3788be7a214115001741b385170734169bcda9ad410420efde33d1742177fad; supervisor sha256:4b3fab6470e9fb7012d931e0b52fc0fbb101fbab27759c83790930feead448b0; sandbox sha256:c46b4539845b589ca0b7717f3baee57003c796ab65e4817b79f74f6d849f156c.
Logs
Current-upstream results (2026-11-17):


Original tests: Summary [134.003s] 4 tests run: 0 passed, 4 failed, 5 skipped
Stdin guard:   Summary [167.815s] 4 tests run: 4 passed, 5 skipped


Minimal failure output:


thread 'canonical_main_nonzero_exit_preserves_status' panicked:
assertion `left == right` failed: unexpected result
  left: Some(0)
 right: Some(7)
...
thread 'sandbox_create_with_no_keep_cleans_up_after_tty_command' panicked:
main output was not streamed
...
thread 'sandbox_create_with_no_keep_preserves_failure_then_cleans_up' panicked:
ephemeral main should preserve its failure code
  left: Some(0)
 right: Some(17)


Historical diagnostic supervisor logging captured the explicit detach decision (not recaptured in the current-main run):


detach=true ctrl_d=true input_bytes=1 input_error=false
Lingua principale
Rust
Stelle
21.5k
Fork
2.1k
Merge medio
2g 20h
PR unite (30g)
246

Preparare l'ambiente

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 NVIDIA/OpenShell

Tutte le issue di NVIDIA/OpenShell

Issue simili

Altre issue su Rust

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.