[Enrique Saurez (esaurez)]

Follow-up to #754.

A process that a concurrent test spawns holds the test process's descriptors from its fork until it executes its program, and with them the claim of a launch's log that a test closed. The store waits up to CLAIM_RELEASE_WAIT, one second (#656), for such a claim to be released before it reports a launch as indeterminate, which covers that moment unless the machine is loaded. #754 let its staging test wait for a later recovery in that case (c32b50e). Three other tests still expect a launch to end at once:

• an_abandoned_launch_that_spawned_nothing_has_exited (nvx-rs: add the sandbox store, launch journal, and runtime identity #628, nvx-rs/src/host/store/tests.rs): its recovery after a refused commit, and its abandon.
• a_recorded_vmm_that_cannot_be_held_has_exited_once_its_log_is_released (nvx-rs: add the sandbox store, launch journal, and runtime identity #628, same file): its checks of a recorded VMM that cannot be held, once the test releases the log.
• unrecorded_launches_end_once_their_log_is_released (nvx-rs: move host::instance onto the store and supervisor #721, aci_edge_sandboxes/tests/openvmm_fake.rs): its start, once the test releases the log.

Changes

• The two store tests accept an indeterminate abandon, and wait with released() for a later recovery, or a later check, to end the launch, as the staging test does.
• The aci_edge_sandboxes test retries a start that fails with BackendError for up to 10 seconds, as a caller retries an indeterminate recovery, and fails on any other error.

Only tests change.

Reproduction

On Linux, temporary tests, which this PR does not add, spawned from another thread a child that sleeps three seconds between its fork and its exec, while a launch's log was open:

• abandon() returned Indeterminate("a process holds its log, and no process serves its control endpoint yet"), where an_abandoned_launch_that_spawned_nothing_has_exited expected Exited. released() saw the next recovery end the launch 1.7 seconds later.
• end_identified() returned Indeterminate("a process still holds its log, and no process matches its VMM process 42"), where a_recorded_vmm_that_cannot_be_held_has_exited_once_its_log_is_released expected Exited. released() saw it end the launch 1.7 seconds later.

Validation

At this branch's head, on dev's 6b6e1215:

• Windows 11 (x64, MSVC):
  • nvx-rs, with NVX_BUNDLE=skip: cargo fmt --all --check; cargo clippy --locked --all-targets --all-features -- -D warnings and cargo clippy --locked --lib --no-default-features -- -D warnings; cargo test --locked --all-features, then cargo test --locked; cargo doc --locked --no-deps --all-features with RUSTDOCFLAGS=-D warnings; cargo +1.89 check --locked --all-targets --all-features.
  • aci_edge_sandboxes: the same commands with --workspace, and python scripts/nvx.py test-agent-bundling --toolchain 1.95.0.
  • tests/runtime-adapter: the protocol-only and host-only configurations, as check-nvx-rs runs them.
• Linux (x86_64): the same nvx-rs and aci_edge_sandboxes commands, and both macOS cargo check --target aarch64-apple-darwin runs.

[Copilot]

🟢 Approval recommended

The focused test changes accommodate temporary log ownership while preserving the required recovery and lifecycle outcomes.

No open issues 🎉

What changed in this PR

This test-only follow-up to #754 reduces launch-recovery flakiness when children spawned by concurrent tests temporarily retain a log claim. Production behavior is unchanged.

Changes:

• Store tests use released() to wait for recovery or exit checks after the log is closed.
• The fake OpenVMM test retries starts returning BackendError for up to 10 seconds, while preserving cleanup and workload assertions.

File	Description
nvx-rs/​src/​host/​store/​tests.rs	Allows delayed log-claim release in abandon and recorded-VMM tests.
aci_edge_sandboxes/​tests/​openvmm_fake.rs	Adds bounded start retries after releasing an unrecorded launch’s log.

_{Review effort: Balanced}