fix(telemetry): recover worker agent and Tailscale transport automatically

tests/test_compute_contour_api.py

@@ -144,3 +144,17 @@ def test_contour_router_returns_404_for_unknown_contour(tmp_path: Path) -> None:
     with pytest.raises(HTTPException) as error:
         install("missing")
     assert error.value.status_code == 404
+
+
+def test_broker_apply_uses_explicit_installation_root(tmp_path, monkeypatch):
+    from k1link.web import compute_contour_api as api
+    root = tmp_path / 'prepared-stack'
+    monkeypatch.setenv('MISSIONCORE_TELEMETRY_PLANE_ROOT', str(root))
+    seen = []
+    def apply(target, installation):
+        seen.append(installation)
+        return {'ready': True}
+    monkeypatch.setattr(api, 'apply_broker_network', apply)
+    router = api.build_compute_contour_router(root_provider=lambda: tmp_path / 'system')
+    _endpoint(router, '/api/v1/system/contours/{contour_id}/network/broker', 'POST')('worker-006')
+    assert seen == [root]

tests/test_telemetry_startup.py

@@ -0,0 +1,146 @@
+from __future__ import annotations
+
+import importlib.util
+import json
+import plistlib
+import subprocess
+import sys
+from pathlib import Path
+
+import pytest
+
+SCRIPTS = Path(__file__).parents[1] / "scripts"
+sys.path.insert(0, str(SCRIPTS))
+try:
+    spec = importlib.util.spec_from_file_location(
+        "telemetry_startup", SCRIPTS / "manage_telemetry_startup.py"
+    )
+    manager = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(manager)
+finally:
+    sys.path.pop(0)
+
+
+def setup(tmp_path, monkeypatch):
+    stack = tmp_path / "state"
+    repo = tmp_path / "release"
+    agents = tmp_path / "agents"
+    for p in (stack, repo, agents):
+        p.mkdir()
+    for name in (
+        "compose.yaml",
+        "runtime/agents.json",
+        "runtime/mosquitto/acl",
+        "runtime/mosquitto/passwords",
+    ):
+        p = stack / name
+        p.parent.mkdir(parents=True, exist_ok=True)
+        p.write_text("fixture")
+    (stack / ".env").write_text("MISSIONCORE_MQTT_BIND_ADDRESS=192.168.1.5\nSECRET=preserve-me\n")
+    (repo / ".venv/bin").mkdir(parents=True)
+    (repo / ".venv/bin/python").write_text("fixture")
+    docker = tmp_path / "docker"
+    docker.write_text("fixture")
+    monkeypatch.setattr(manager, "DOCKER", str(docker))
+    (agents / (manager.CORE + ".plist")).write_bytes(
+        plistlib.dumps(
+            {
+                "Label": manager.CORE,
+                "WorkingDirectory": str(repo),
+                "EnvironmentVariables": {"MISSIONCORE_DATA_DIR": "/existing-data"},
+            }
+        )
+    )
+    monkeypatch.setattr(
+        manager,
+        "command",
+        lambda *args: subprocess.CompletedProcess(
+            [], 0, b"hostname worker.example.ts.net\nstricthostkeychecking true\n", b""
+        ),
+    )
+    return stack, repo, agents
+
+
+def test_plan_preserves_credentials_and_pins_loopback_transport(tmp_path, monkeypatch):
+    stack, repo, agents = setup(tmp_path, monkeypatch)
+    before = (stack / ".env").read_bytes()
+    doc, files = manager.plan(stack, "worker-test", repo, agents)
+    assert (stack / ".env").read_bytes() == before
+    assert files[stack / ".env"] == b"MISSIONCORE_MQTT_BIND_ADDRESS=127.0.0.1\nSECRET=preserve-me\n"
+    assert "preserve-me" not in json.dumps(doc)
+    startup = plistlib.loads(files[agents / (manager.STARTUP + ".plist")])
+    assert startup["RunAtLoad"] and startup["StartInterval"] == 30
+    tunnel = plistlib.loads(files[agents / (manager.TUNNEL + ".plist")])
+    assert "127.0.0.1:1883:127.0.0.1:1883" in tunnel["ProgramArguments"]
+    assert "StrictHostKeyChecking=yes" in tunnel["ProgramArguments"]
+    assert tunnel["KeepAlive"] and tunnel["ThrottleInterval"] >= 5
+    core = plistlib.loads(files[agents / (manager.CORE + ".plist")])
+    assert core["EnvironmentVariables"]["MISSIONCORE_DATA_DIR"] == "/existing-data"
+    assert core["EnvironmentVariables"]["MISSIONCORE_TELEMETRY_PLANE_ROOT"] == str(stack)
+
+
+def test_plan_refuses_non_tailscale_and_changed_input(tmp_path, monkeypatch):
+    stack, repo, agents = setup(tmp_path, monkeypatch)
+    before, _ = manager.plan(stack, "worker-test", repo, agents)
+    with (stack / ".env").open("a") as f:
+        f.write("NEW=setting\n")
+    after, _ = manager.plan(stack, "worker-test", repo, agents)
+    assert before["sha256"] != after["sha256"]
+    monkeypatch.setattr(
+        manager,
+        "command",
+        lambda *args: subprocess.CompletedProcess(
+            [], 0, b"hostname worker.local\nstricthostkeychecking true\n", b""
+        ),
+    )
+    with pytest.raises(ValueError, match="Tailscale"):
+        manager.plan(stack, "worker-test", repo, agents)
+
+
+def test_delayed_docker_retries_without_compose_or_workload_launch(monkeypatch, tmp_path):
+    calls = []
+    monkeypatch.setattr(manager, "receiver_ready", lambda: False)
+
+    def run(args, timeout):
+        calls.append(args)
+        return subprocess.CompletedProcess(args, 1 if "info" in args else 0)
+
+    monkeypatch.setattr(manager, "command", run)
+    assert manager.reconcile(tmp_path) == "waiting-for-docker"
+    assert len(calls) == 2 and calls[1][0] == "/usr/bin/open"
+    assert not any("compose" in c for c in calls)
+
+
+def test_ready_receiver_is_not_restarted(monkeypatch, tmp_path):
+    monkeypatch.setattr(manager, "receiver_ready", lambda: True)
+    monkeypatch.setattr(manager, "broker_loopback", lambda: True)
+    monkeypatch.setattr(
+        manager, "command", lambda *args: pytest.fail("Must not restart healthy services")
+    )
+    assert manager.reconcile(tmp_path) == "ready"
+
+
+def test_old_lan_binding_is_reconciled_even_when_receiver_healthy(monkeypatch, tmp_path):
+    calls = []
+    bindings = iter((False, True))
+    monkeypatch.setattr(manager, "receiver_ready", lambda: True)
+    monkeypatch.setattr(manager, "broker_loopback", lambda: next(bindings))
+
+    def run(args, timeout):
+        calls.append(args)
+        return subprocess.CompletedProcess(args, 0)
+
+    monkeypatch.setattr(manager, "command", run)
+    assert manager.reconcile(tmp_path) == "ready"
+    assert calls[-1][-3:] == ["broker", "timescale", "normalizer"]
+    assert "--no-build" in calls[-1] and "--pull" in calls[-1] and "never" in calls[-1]
+
+
+def test_failed_compose_reports_waiting_without_ready(monkeypatch, tmp_path):
+    monkeypatch.setattr(manager, "receiver_ready", lambda: False)
+    monkeypatch.setattr(
+        manager,
+        "command",
+        lambda args, timeout: subprocess.CompletedProcess(args, int("up" in args)),
+    )
+    assert manager.reconcile(tmp_path) == "waiting-for-receiver"