feat(governance): 新增服務健康缺口矩陣

apps/api/src/services/service_health_gap_matrix.py

@@ -0,0 +1,282 @@
+"""
+Service health gap matrix snapshot.
+
+Loads the latest committed, read-only service health / stale endpoint gap
+matrix. This module never restarts services, changes endpoints, probes live
+systems, reads secrets, sends notifications, triggers workflows, deploys, or
+executes runtime actions.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+from typing import Any
+
+from src.services.snapshot_paths import default_evaluations_dir
+
+_DEFAULT_EVALUATIONS_DIR = default_evaluations_dir(Path(__file__))
+_SNAPSHOT_PATTERN = "service_health_gap_matrix_*.json"
+_SCHEMA_VERSION = "service_health_gap_matrix_v1"
+
+
+def load_latest_service_health_gap_matrix(
+    evaluations_dir: Path | None = None,
+) -> dict[str, Any]:
+    """Load the newest committed service health gap matrix snapshot."""
+    directory = evaluations_dir or _DEFAULT_EVALUATIONS_DIR
+    candidates = sorted(directory.glob(_SNAPSHOT_PATTERN))
+    if not candidates:
+        raise FileNotFoundError(f"no service health gap matrix snapshots found in {directory}")
+
+    latest = candidates[-1]
+    with latest.open(encoding="utf-8") as handle:
+        payload = json.load(handle)
+
+    if not isinstance(payload, dict):
+        raise ValueError(f"{latest}: expected JSON object")
+    _require_schema(payload, _SCHEMA_VERSION, str(latest))
+    _require_read_only_boundaries(payload, str(latest))
+    _require_operation_boundaries(payload, str(latest))
+    _require_rollup_consistency(payload, str(latest))
+    _require_target_evidence(payload, str(latest))
+    _require_gap_evidence(payload, str(latest))
+    _require_operator_denials(payload, str(latest))
+    _require_no_plaintext_secret_payload_keys(payload, str(latest))
+    return payload
+
+
+def _require_schema(payload: dict[str, Any], expected: str, label: str) -> None:
+    actual = payload.get("schema_version")
+    if actual != expected:
+        raise ValueError(f"{label}: expected schema_version={expected}, got {actual!r}")
+
+
+def _require_read_only_boundaries(payload: dict[str, Any], label: str) -> None:
+    program_status = payload.get("program_status") or {}
+    if program_status.get("read_only_mode") is not True:
+        raise ValueError(f"{label}: program_status.read_only_mode must be true")
+
+    approval_boundaries = payload.get("approval_boundaries") or {}
+    allowed = sorted(key for key, value in approval_boundaries.items() if value is not False)
+    if allowed:
+        raise ValueError(f"{label}: approval boundaries must remain false: {allowed}")
+
+
+def _require_operation_boundaries(payload: dict[str, Any], label: str) -> None:
+    boundaries = payload.get("operation_boundaries") or {}
+    if boundaries.get("read_only_api_allowed") is not True:
+        raise ValueError(f"{label}: read_only_api_allowed must be true")
+
+    blocked_flags = {
+        "service_restart_allowed",
+        "pod_restart_allowed",
+        "host_restart_allowed",
+        "rollout_restart_allowed",
+        "endpoint_change_allowed",
+        "configmap_patch_allowed",
+        "active_probe_allowed",
+        "external_health_probe_allowed",
+        "live_benchmark_allowed",
+        "provider_switch_allowed",
+        "paid_api_call_allowed",
+        "secret_read_allowed",
+        "secret_plaintext_allowed",
+        "notification_send_allowed",
+        "workflow_trigger_allowed",
+        "deploy_trigger_allowed",
+        "reload_trigger_allowed",
+        "runtime_execution_allowed",
+    }
+    allowed = sorted(flag for flag in blocked_flags if boundaries.get(flag) is not False)
+    if allowed:
+        raise ValueError(f"{label}: operation boundaries must remain false: {allowed}")
+
+
+def _require_rollup_consistency(payload: dict[str, Any], label: str) -> None:
+    targets = payload.get("service_health_targets") or []
+    gaps = payload.get("health_gaps") or []
+    stale_endpoints = payload.get("stale_endpoints") or []
+    rollups = payload.get("rollups") or {}
+
+    if rollups.get("total_targets") != len(targets):
+        raise ValueError(f"{label}: rollups.total_targets must match service_health_targets")
+    if rollups.get("by_kind") != _count_by(targets, "kind"):
+        raise ValueError(f"{label}: rollups.by_kind must match service_health_targets")
+    if rollups.get("by_status") != _count_by(targets, "status"):
+        raise ValueError(f"{label}: rollups.by_status must match service_health_targets")
+    if rollups.get("by_freshness_status") != _count_by(targets, "freshness_status"):
+        raise ValueError(f"{label}: rollups.by_freshness_status must match service_health_targets")
+
+    requiring_action = sorted(
+        target.get("target_id")
+        for target in targets
+        if target.get("status") == "action_required"
+    )
+    if sorted(rollups.get("target_ids_requiring_action") or []) != requiring_action:
+        raise ValueError(f"{label}: rollups.target_ids_requiring_action must match targets")
+
+    if sorted(rollups.get("health_gap_ids") or []) != sorted(gap.get("gap_id") for gap in gaps):
+        raise ValueError(f"{label}: rollups.health_gap_ids must match health_gaps")
+
+    if sorted(rollups.get("stale_endpoint_ids") or []) != sorted(
+        endpoint.get("endpoint_id") for endpoint in stale_endpoints
+    ):
+        raise ValueError(f"{label}: rollups.stale_endpoint_ids must match stale_endpoints")
+
+    critical_targets = sorted(
+        target.get("target_id")
+        for target in targets
+        if target.get("risk_level") == "critical"
+    )
+    if sorted(rollups.get("critical_target_ids") or []) != critical_targets:
+        raise ValueError(f"{label}: rollups.critical_target_ids must match service_health_targets")
+
+    zero_count_fields = {
+        "service_restart_allowed_count",
+        "endpoint_change_allowed_count",
+        "active_probe_allowed_count",
+        "notification_send_allowed_count",
+        "runtime_execution_allowed_count",
+    }
+    non_zero = sorted(field for field in zero_count_fields if rollups.get(field) != 0)
+    if non_zero:
+        raise ValueError(f"{label}: operation permission rollup counts must remain 0: {non_zero}")
+
+
+def _require_target_evidence(payload: dict[str, Any], label: str) -> None:
+    targets = payload.get("service_health_targets") or []
+    missing = sorted(
+        target.get("target_id")
+        for target in targets
+        if not target.get("health_contract")
+        or not target.get("endpoint_contract")
+        or not target.get("evidence_refs")
+        or not target.get("next_action")
+    )
+    if missing:
+        raise ValueError(
+            f"{label}: service_health_targets must include health, endpoint, evidence, next_action: {missing}"
+        )
+
+    required_target_ids = {
+        "production_api_health_public",
+        "awoooi_web_health_manifest",
+        "ollama_three_layer_health_contract",
+        "openclaw_health_endpoint_contract",
+        "prometheus_alertmanager_endpoint_reference",
+        "gitea_workflow_runner_health_contract",
+        "kali_scanner_health_reference",
+    }
+    present = {target.get("target_id") for target in targets}
+    missing_required = sorted(required_target_ids - present)
+    if missing_required:
+        raise ValueError(f"{label}: missing required service health targets: {missing_required}")
+
+
+def _require_gap_evidence(payload: dict[str, Any], label: str) -> None:
+    gaps = payload.get("health_gaps") or []
+    endpoints = payload.get("stale_endpoints") or []
+    required_gap_ids = {
+        "endpoint_reference_stale_hosts",
+        "gitea_runner_attestation_health_gap",
+        "health_check_script_not_authoritative",
+        "openclaw_nemo_rca_health_review",
+        "security_scanner_health_evidence_gap",
+    }
+    present_gaps = {gap.get("gap_id") for gap in gaps}
+    missing_gaps = sorted(required_gap_ids - present_gaps)
+    if missing_gaps:
+        raise ValueError(f"{label}: missing required health gaps: {missing_gaps}")
+
+    missing_gap_fields = sorted(
+        gap.get("gap_id")
+        for gap in gaps
+        if not gap.get("target_ids") or not gap.get("evidence_refs") or not gap.get("next_action")
+    )
+    if missing_gap_fields:
+        raise ValueError(f"{label}: health gaps must include targets, evidence, next_action: {missing_gap_fields}")
+
+    required_endpoint_ids = {
+        "legacy_188_ollama_provider_endpoint",
+        "prometheus_alertmanager_110_188_split",
+        "openclaw_8088_comment_vs_8089_contract",
+    }
+    present_endpoints = {endpoint.get("endpoint_id") for endpoint in endpoints}
+    missing_endpoints = sorted(required_endpoint_ids - present_endpoints)
+    if missing_endpoints:
+        raise ValueError(f"{label}: missing required stale endpoints: {missing_endpoints}")
+
+    missing_endpoint_fields = sorted(
+        endpoint.get("endpoint_id")
+        for endpoint in endpoints
+        if not endpoint.get("stale_ref")
+        or not endpoint.get("current_truth")
+        or not endpoint.get("evidence_refs")
+        or not endpoint.get("next_action")
+    )
+    if missing_endpoint_fields:
+        raise ValueError(
+            f"{label}: stale endpoints must include stale_ref, current_truth, evidence, next_action: {missing_endpoint_fields}"
+        )
+
+
+def _require_operator_denials(payload: dict[str, Any], label: str) -> None:
+    contract = payload.get("operator_contract") or {}
+    must_not_interpret_as = set(contract.get("must_not_interpret_as") or [])
+    required_denials = {
+        "服務重啟批准",
+        "endpoint 修改批准",
+        "active probe 批准",
+        "live health check 已執行",
+        "Secret payload 已讀取或可輸出",
+        "Telegram 成功通知批准",
+        "workflow / deploy / reload 觸發批准",
+        "provider 切換批准",
+        "runtime execution 授權",
+        "OpenClaw 取代或降級批准",
+    }
+    if not required_denials.issubset(must_not_interpret_as):
+        raise ValueError(f"{label}: operator_contract.must_not_interpret_as is missing required denials")
+
+    restart_policy = str(contract.get("restart_policy") or "")
+    if "P1-005" not in restart_policy or "只能列缺口" not in restart_policy:
+        raise ValueError(f"{label}: restart_policy must preserve P1-005 read-only boundary")
+
+    endpoint_policy = str(contract.get("endpoint_policy") or "")
+    if "P1-005" not in endpoint_policy or "不修改端點" not in endpoint_policy:
+        raise ValueError(f"{label}: endpoint_policy must preserve P1-005 endpoint boundary")
+
+    notification_policy = str(contract.get("notification_policy") or "")
+    if "P1-007" not in notification_policy or "成功 smoke 不通知" not in notification_policy:
+        raise ValueError(f"{label}: notification_policy must preserve failure-only notification boundary")
+
+
+def _require_no_plaintext_secret_payload_keys(value: Any, label: str, path: str = "$") -> None:
+    if isinstance(value, dict):
+        forbidden_key_fragments = {
+            "secret_value",
+            "token_value",
+            "authorization_header",
+            "private_key",
+            "webhook_secret",
+            "runner_token",
+            "api_key_value",
+            "password_value",
+        }
+        for key, nested in value.items():
+            lowered = str(key).lower()
+            if any(fragment in lowered for fragment in forbidden_key_fragments):
+                raise ValueError(f"{label}: forbidden secret payload key at {path}.{key}")
+            _require_no_plaintext_secret_payload_keys(nested, label, f"{path}.{key}")
+    elif isinstance(value, list):
+        for index, nested in enumerate(value):
+            _require_no_plaintext_secret_payload_keys(nested, label, f"{path}[{index}]")
+
+
+def _count_by(items: list[dict[str, Any]], key: str) -> dict[str, int]:
+    counts: dict[str, int] = {}
+    for item in items:
+        value = item.get(key)
+        counts[value] = counts.get(value, 0) + 1
+    return counts