fix(telemetry): restore worker agent delivery and truthful system status

tests/test_compute_contour_network.py

@@ -63,6 +63,21 @@ def test_broker_apply_preserves_private_environment_and_changes_only_bind(
     assert result["ready"] is True
 
 
+def test_broker_apply_repairs_missing_listener_without_changing_secrets(tmp_path, monkeypatch):
+    before = "MISSIONCORE_MQTT_BIND_ADDRESS=192.168.68.56\nSECRET=keep\n"
+    (tmp_path / ".env").write_text(before)
+    (tmp_path / "compose.yaml").write_text("services: {}\n")
+    reachable = iter((False, True))
+    calls = []
+    monkeypatch.setattr(network, "_address_belongs_to_host", lambda _: True)
+    monkeypatch.setattr(network, "_tcp_reachable", lambda *a, **k: next(reachable))
+    monkeypatch.setattr(network, "_run_compose_broker", lambda path: calls.append(path))
+    result = network.apply_broker_network(_target(), tmp_path)
+    assert not result["changed"] and result["repaired"] and result["ready"]
+    assert calls == [tmp_path]
+    assert (tmp_path / ".env").read_text() == before
+
+
 def test_broker_apply_rolls_back_environment_when_listener_does_not_open(
     tmp_path: Path,
     monkeypatch: pytest.MonkeyPatch,

tests/test_system_telemetry_api.py

@@ -447,6 +447,36 @@ def test_compute_contour_maps_to_worker_identity_without_singleton_defaults() ->
     assert profile.address == "192.0.2.25"
 
 
+def test_gpu_clocks_and_explicit_container_status_are_not_invented() -> None:
+    raw = _agent_raw_document({"samples": [
+        {"measurement": "nvidia_smi", "payload": {"fields": {
+            "clocks_current_sm": 210, "clocks_current_memory": 405,
+        }}},
+        {"measurement": "docker_container_status", "payload": {
+            "tags": {"container_name": "sentinel-frigate", "container_status": "exited"},
+            "fields": {},
+        }},
+    ]})
+    assert raw["gpu"]["sm_clock_mhz"] == 210
+    assert raw["gpu"]["memory_clock_mhz"] == 405
+    assert raw["container_states"]["sentinel-frigate"]["state"]["Status"] == "exited"
+
+
+def test_unavailable_pipeline_is_not_ready_and_profile_wait_survives(tmp_path) -> None:
+    probe = _probe()
+    probe["raw"]["perception"] = {"state": "unavailable", "collector_state": "unavailable"}
+    service = WorkerTelemetryService(WorkerProfileStore(tmp_path), lambda _: probe, cache_seconds=0)
+    assert all(s["state"] == "unavailable" for s in service.snapshot(1)["pipeline"]["stages"])
+    probe["raw"]["perception"] = {
+        "state": "waiting", "profile_name": "K1 DDRNet", "input_pauses": 1,
+        "live_children": 4, "buffer_bytes": 1024,
+    }
+    profile = service.snapshot(1)["pipeline"]
+    assert profile["service_state"] == "waiting"
+    assert profile["live_children"] == 4 and profile["buffer_bytes"] == 1024
+    assert all(s["state"] == "waiting" for s in profile["stages"])
+
+
 def test_profile_apply_fails_closed_on_wrong_node_and_keeps_old_profile(
     tmp_path: Path,
 ) -> None:

tests/test_telemetry_normalizer_contract.py

@@ -126,6 +126,22 @@ def test_normalizer_enforces_oss_compatible_bounded_retention() -> None:
     assert "DELETE FROM contour_telemetry_samples" in NORMALIZER_SOURCE
     assert "INTERVAL '30 days'" in NORMALIZER_SOURCE
     assert "RETENTION_INTERVAL_SECONDS" in NORMALIZER_SOURCE
+    assert "ORDER BY observed_at LIMIT %s" in NORMALIZER_SOURCE
+    assert "statement_timeout = '2s'" in NORMALIZER_SOURCE
+    assert "samples.tableoid = expired.tableoid" in NORMALIZER_SOURCE
+    assert "target=_retention_loop" in NORMALIZER_SOURCE
+    callback = NORMALIZER_SOURCE.split("    def on_message(", 1)[1]
+    assert "DELETE FROM" not in callback
+
+
+def test_telegraf_network_recovery_and_database_memory_are_bounded() -> None:
+    compose = yaml.safe_load(COMPOSE_PATH.read_text())
+    assert "shared_buffers=128MB" in compose["services"]["timescale"]["command"]
+    for platform, outputs in (("windows", 2), ("linux", 1)):
+        template = (REPOSITORY_ROOT / "deploy/telemetry-plane/telegraf" /
+                    f"mission-core-{platform}.conf.tmpl").read_text()
+        assert template.count('startup_error_behavior = "retry"') == outputs
+        assert "metric_buffer_limit = 2000" in template
 
 
 def test_normalizer_rejects_unschematized_pipeline_payload() -> None: