feat(driver): implement sandbox RPCs (Create/Get/List/Stop/Delete/Validate + Watch)

[kadinsayani]

Summary

Wires up the ComputeDriver gRPC service against LXD: every RPC beyond GetCapabilities was previously Unimplemented.

• ValidateSandboxCreate — required-field validation, label-key format checks (shared with create_sandbox's own validation so the two can't drift), and GPU count sanity (rejects 0, None already means "one GPU, default assignment")
• GetSandbox / ListSandboxes — read from LXD instance config; resolvable by sandbox_name or sandbox_id (LXD has no by-id lookup, so sandbox_id-only requests fall back to scanning for a config match)
• CreateSandbox — create stopped, push token file, start; rolls back (deletes) the instance if the token push fails, so a failed create doesn't leave an orphaned container; always uses the configured default image in v1 (template.image is accepted but not yet consulted)
• StopSandbox — graceful stop; idempotent on not-running (handles both LXD's synchronous 400 and its asynchronous OperationFailed shape via a shared is_already_stopped helper)
• DeleteSandbox — force-stop + delete; idempotent on 404 at every step (including a concurrent-deletion race on the force-stop step itself); broadcasts a WatchSandboxes Deleted event on success
• WatchSandboxes — deletion broadcast channel so the gateway evicts a deleted sandbox immediately rather than waiting on its 60s reconcile loop; a lagged receiver gets a data_loss stream error (rather than silently missing events) so it knows to reconnect and re-list
• Fail-fast startup check: exits with a clear error if the configured default image alias isn't present in LXD, instead of surfacing an opaque 404 on the first create_sandbox call
• README: adds a v1-alpha status line, a Quickstart (the real sequence verified end-to-end this cycle), and Security/Known-limitations sections

End-to-end verification

Ran a full smoke test against a real OpenShell gateway + this driver (plaintext gateway, --bind-address 0.0.0.0, LXD on lxdbr0):

• Manual CLI lifecycle: sandbox create reached Ready with a fully-populated policy (filesystem_policy, landlock, process) — exercises token delivery, container DHCP bring-up, and supervisor policy fetch in one pass. sandbox exec returned a working shell as the sandbox user. sandbox list immediately after sandbox delete did not show the deleted sandbox, confirming WatchSandboxes's broadcast path (not just the ~60s reconcile fallback).
• The real e2e Rust test suite from OpenShell's e2e/rust/tests/ (smoke, sandbox_lifecycle ×2, sandbox_labels) — 4/4 passing, run unmodified against this driver.

Test plan

• Unit tests (cargo test -p openshell-driver-lxd --lib) — 7 passing, including new ValidateSandboxCreate coverage for label-key and GPU-count validation.
• Integration tests (tests/sandbox_lifecycle.rs) — full create → get → list → stop → delete lifecycle + sandbox_id-only resolution + 404/invalid-argument safety, against real LXD + the openshell-sandbox image — 5/5 passing.
• Manual e2e smoke test against a real gateway (see above).
• cargo build --workspace, clippy -D warnings, fmt --check, shellcheck all clean.

[Copilot]

Pull request overview

Implements the remaining ComputeDriver sandbox lifecycle RPCs for openshell-driver-lxd by wiring them to LXD instance operations/config, adds deletion-driven WatchSandboxes events, and updates the sandbox image build + CI to support end-to-end integration testing against a real LXD daemon.

Changes:

• Implement ValidateSandboxCreate, Create/Get/List/Stop/DeleteSandbox, and WatchSandboxes in the gRPC service + core driver logic (including LXD config/device/profile mapping).
• Add real-LXD integration tests covering the full sandbox lifecycle via the in-process gRPC service.
• Update image/build/CI scripts (rockcraft-based openshell-sandbox build, deterministic lxc publish compression, and leaner apt installs).

Reviewed changes

Copilot reviewed 16 out of 17 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
scripts/setup-lxd-test-env.sh	Adjusts LXD test env image-alias existence check before staging the test image.
scripts/openshell-container-init.sh	Simplifies /sandbox ownership setup inside the container init wrapper.
scripts/build-sandbox-image.sh	Makes published image compression deterministic (--compression none).
rockcraft.yaml	Builds openshell-sandbox from source via the Rust plugin and pins the version/commit.
crates/openshell-driver-lxd/tests/sandbox_lifecycle.rs	Adds full sandbox lifecycle integration tests (create/get/list/stop/delete + error cases).
crates/openshell-driver-lxd/tests/get_capabilities.rs	Updates test wiring for the new LxdComputeDriver::new(config, lxd) constructor.
crates/openshell-driver-lxd/src/mapping.rs	Adds mapping helpers to translate between proto sandbox shapes and LXD instance config/devices/profiles.
crates/openshell-driver-lxd/src/main.rs	Adds LXD endpoint selection (Unix socket vs HTTPS) and constructs LxdClient for the driver.
crates/openshell-driver-lxd/src/lib.rs	Exposes the new internal mapping module.
crates/openshell-driver-lxd/src/grpc.rs	Implements sandbox RPC handlers and adds a broadcast-backed WatchSandboxes deletion stream.
crates/openshell-driver-lxd/src/driver.rs	Implements the core sandbox lifecycle against LXD (create/push token/start, get/list, stop, delete, id lookup).
crates/openshell-driver-lxd/src/config.rs	Adds --gateway-grpc-port used to construct OPENSHELL_ENDPOINT for sandboxes.
crates/lxd-client/src/operations.rs	Logs malformed websocket event frames at WARN instead of silently skipping.
Cargo.toml	Enables tokio-stream’s sync feature for BroadcastStream.
Cargo.lock	Updates lockfile for dependency feature changes.
.github/workflows/tests.yaml	Uses apt-get install --no-install-recommends in test workflows.
.github/workflows/static-checks.yaml	Uses apt-get install --no-install-recommends in static-check workflows.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Pull request overview

Copilot reviewed 31 out of 32 changed files in this pull request and generated 2 comments.

[Copilot]

Pull request overview

Copilot reviewed 14 out of 15 changed files in this pull request and generated 3 comments.

Comments suppressed due to low confidence (1)

crates/openshell-driver-lxd/src/grpc.rs:171

• BroadcastStreamRecvError also has a Closed variant; the current match is non-exhaustive and won’t compile (or will require a wildcard arm). Handle Closed explicitly and map it to a gRPC status (e.g. Unavailable).

            Err(BroadcastStreamRecvError::Lagged(n)) => Err(Status::data_loss(format!(
                "WatchSandboxes receiver lagged and missed {n} event(s); reconnect and re-list to resync"
            ))),
        });

[Copilot]

Pull request overview

Copilot reviewed 14 out of 15 changed files in this pull request and generated 3 comments.

[morphis]

LGTM