pilotprotocol/.github/workflows/architecture.yml at main · pilot-protocol/pilotprotocol · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
name: architecture-gates

on: [pull_request, push]

permissions:
  contents: read   # PILOT-113: least-privilege default

jobs:
  gates:
    name: Architecture gates
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v7

      - uses: actions/setup-go@v5
        with:
          go-version-file: go.mod

      # check-layers extracted to pilot-protocol/check-layers; the
      # layer-import + test-import gates that used to live here now
      # run from that sibling repo against any consumer that imports
      # web4 packages.

      # plugin-matrix removed: every plugin (updater, skillinject,
      # trustedagents, nameserver, dataexchange, eventstream, tasks,
      # gateway, policy, webhook) extracted to its own sibling repo,
      # and the no_<plugin> build tags they exercised left with them.

      - name: Race detector — all library packages (-short)
        # Broad `-race` over the whole pkg/cmd/internal tree so a data race in
        # any library — keyexchange frame parsing, envelope codecs, routing,
        # transport — fails CI, not just the daemon lock graph below. -short
        # skips the known-slow stress tests (run un-short in dedicated steps);
        # -parallel 4 is the project convention to avoid port/socket exhaustion
        # under the race detector's higher resource use. This step is also what
        # keeps the rotate-key/sign regression (TestConcurrentRotateKeyAndSign
        # in pkg/daemon) and the new keyexchange/IPC fuzz seed corpora under the
        # race detector on every PR.
        run: go test -race -short -parallel 4 -count=1 -timeout 12m ./pkg/... ./cmd/... ./internal/...

      - name: Lock-graph race detection (sec 4.8)
        # pkg/registry/ and pkg/secure/ extracted to pilot-protocol/{rendezvous,common}
        # in PR #155; lock-graph coverage for those layers now runs from
        # their own repos. pkg/daemon stays — that's where the daemon-side
        # lock graph (Store.mu, ReplayMu, SalvageMu, tm.mu, …) actually lives.
        # Kept distinct from the broad step above because it runs WITHOUT -short
        # so the pkg/daemon stress tests are race-checked too.
        run: go test -race -timeout 5m ./pkg/daemon/...

      - name: Lock-graph stress harness (sec 4.8) — TestConcurrentDialEncryptDecrypt
        # 1000 goroutines × 30 s × 3 reps = ~90 s wall time. Race-clean,
        # no panic, no goroutine leak. Skipped by -short; run here without.
        run: go test -race -count=1 -timeout 10m -run TestConcurrentDialEncryptDecrypt ./tests/...

      - name: P4/P9 - goroutine reaping
        run: go test ./tests -run "TestDaemonShutdown|TestStreamLifecycleGoroutines|TestPluginShutdown"