Repository navigation
test: preserve timed diagnostics for CLI file-transfer acceptance #109
Description
Problem
The CLI acceptance fixture in e2e/provider_live.py::cli_express cannot diagnose intermittent file-transfer failures: it discards upload/download stderr and records no per-step elapsed time or HTTP status. This blocks reliable classification of a failed production-provider check.
Observed while validating #108 (candidate client/server 0.4.0a1 and built CLI 0.5.0):
- Configured full CLI case: PASS, 376.63 s, including create/info/list/exec/stream/upload/download/delete/delete-again.
- Additional run with E2B credential variable names explicitly mapped to the ABX_SANDBOX Express configuration: FAIL, 355.17 s. Lifecycle/list/exec/stream passed; upload exited 2, download exited 4, downloaded content was absent. Both deletion calls exited 0; the second confirmed the sandbox was absent.
- Creation explicitly requests a 300 s sandbox timeout. Both total runtimes exceeded that duration, but the passing run was longer, and no step timings survive. Expiration is not confirmed.
- Exit 2 and 4 cover multiple error categories; stderr and HTTP status are absent. No product root cause, Provider defect, or network-audit attribution can be established from these observations.
- The separate new unified-template startup suite is 5/5 PASS, including two services started together. The failure above is not evidence of template startup failure.
Requested narrow follow-up
Add per-step monotonic elapsed time, sanitized stderr / available HTTP status classification, and explicit effective Provider configuration identity (never credential values) to this SDK-owned E2E fixture. Keep original pass/failure evidence separate. Diagnose a reproduced failure before opening a specific CLI/Provider product fix.
Acceptance
- Failed file transfers preserve useful, credential-safe diagnostics.
- Sandbox timeout and the elapsed time at each operation can be compared.
- Owned-resource deletion remains guaranteed and records its result.
- No behavior is weakened to turn an unexplained failure into a pass.
The fixture improvement belongs in a separate change; the unified template contract remains scoped in #107 / #108.
Focused follow-up evidence
A new owned sandbox with the same explicit Express environment mapping passed file upload/download with exact byte comparison. Create took 42.601 s, upload 29.551 s, download 30.885 s; download completed 103.038 s after the diagnostic began. Delete and repeated delete exited 0; the latter confirmed the sandbox absent. Total 180.946 s with the same 300 s create timeout. No retries, fixture changes or product changes were used.
This confirms the candidate file functionality worked in the focused experiment; it does not explain or erase the earlier full acceptance failure. The diagnostic improvement requested by this Issue remains open.
Activity