test(grpc): TypeScript wire-level contract suite + codecov integration coverage

[markovejnovic]

Summary

A TypeScript integration suite (test/grpc/) that exercises the public gRPC surface (hyper.grpc.v0.Hyper) over the real wire, from outside the BEAM — driven by an ExUnit :integration wrapper so the server-side Elixir coverage it produces flows to Codecov.

• TS contract suite (@grpc/grpc-js + @grpc/proto-loader, vitest): loads the committed proto/hyper/grpc/v0/hyper.proto directly; generated TS types are gitignored and rebuilt by npm scripts (repo codegen rule).
  • errors.test.ts — 7 tests pinning the proto's documented status-code contracts (NOT_FOUND ×3, INVALID_ARGUMENT ×4).
  • lifecycle.test.ts — LoadImage → RESOURCE_EXHAUSTED probe (TERA) → CreateVm → GetVm → ListVms → GetVmUsage → StopVm → polls GetVm to NOT_FOUND.
• ExUnit wrapper test/e2e/grpc_contract_test.exs starts the gRPC server on port 50061 under the booted app tree and shells out to npm test.
• Two real server bugs found and fixed by the suite:
  1. fix(grpc): unrecognised enum ints (e.g. instance_type: 999 — only producible by an off-BEAM client) raised FunctionClauseError → INTERNAL; now INVALID_ARGUMENT per the proto contract.
  2. fix(grpc): StopVm never tore the VM down — it flipped the gen_statem to the stub :stopping state and nothing terminated the supervisor, leaving the guest running and GetVm resolving forever. New Hyper.stop_vm/1 resolves the VM's supervisor via the routing registry and terminates it on the owning node.
• Codecov: the integration job previously uploaded no coverage; it now runs mix coveralls.json and uploads under a new integration flag, plus vitest JUnit → Test Analytics under grpc-ts.

Test plan

• Live run on a provisioned KVM host: errors 7/7 + lifecycle 1/1 over the wire (real VM boot/stop), post-stop NOT_FOUND observed, no leaked firecracker
• Unit gates: format / compile -Werror / credo --strict / dialyzer / 316 tests
• CI integration job green with both Codecov uploads accepted (this PR run)

[codecov]

❌ 1 Tests Failed:

Tests completed	Failed	Passed	Skipped
155	1	154	316

View the full list of 1 ❄️ flaky test(s)

Elixir.Hyper.E2e.VmLifecycleTest::test load -> create_vm -> exec -> stop reclaims the VM's dm volume

Flake rate in main: 33.33% (Passed 2 times, Failed 1 times)

Stack Traces | 61.7s run time

1) test load -> create_vm -> exec -> stop reclaims the VM's dm volume (Hyper.E2e.VmLifecycleTest)
     test/e2e/vm_lifecycle_test.exs:33
     no usage row after stop_image_vm — the teardown flush never landed
     code: assert poll_until(fn -> Usage.total(vm_id) != nil end, :timer.seconds(30)),
     stacktrace:
       test/e2e/vm_lifecycle_test.exs:66: (test)

To view more test analytics, go to the Test Analytics Dashboard
_{📋 Got 3 mins? Take this short survey to help us improve Test Analytics.}

[github-actions]

Test Results

  4 files  + 1  163 suites  +4   2m 4s ⏱️ -28s
471 tests +11  470 ✅ +10    0 💤 ±0  1 ❌ +1 
798 runs  +14  470 ✅ +10  327 💤 +3  1 ❌ +1 

For more details on these failures, see this check.

Results for commit 53864ed. ± Comparison against base commit ef7243a.

♻️ This comment has been updated with latest results.

[markovejnovic]

Status: branch is review-complete and ready to merge on its own scope. The integration job's remaining red is main's new metering assertion (vm_lifecycle_test.exs:66): a pre-existing Meter blind spot — a VM living under the 1s sample interval never writes a usage row because the first Accumulator observation only sets the baseline. This suite exposes it by warming the alpine image cache (the VM's whole lifecycle drops to ~1s). Being fixed separately per Marko. The integration coverage upload step (flag integration) is gated on a green job and will be exercised once that fix lands; the grpc-ts Test Analytics upload already works (verified in run 28905343125).