Files
awoooi/apps/api/src/services/drift_detector.py
OG T 53e1ae7ad7
Some checks failed
CD Pipeline / Deploy Prometheus Alert Rules (push) Has been cancelled
CD Pipeline / build-and-deploy (push) Has been cancelled
fix(phase25): I2 NIM system prompt + I4 field_path 正則匹配修正
I2: nemotron.analyze() 補上 system role (NIM 標準 message format)
    - 舊: messages=[{role:user, ...}]
    - 新: messages=[{role:system, ...}, {role:user, ...}]
    - 效果: K8s operator 角色定義,改善 tool calling 品質

I4: drift_detector._is_allowlisted/_is_critical 用正則取代 strip
    - 舊: replace('[*]','') 後 startswith/in → 無法匹配 containers[0]
    - 新: [*] → \[\d+\] 正則,正確匹配所有索引
    - 修復: containers[*].image 現在能匹配 containers[0].image
2026-04-05 12:11:05 +08:00

341 lines
11 KiB
Python
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
"""
Drift Detector - Phase 25 P2 Config Drift Detection
=====================================================
職責:比對 Git YAML vs K8s 實際狀態,輸出結構化 DriftItem 列表
不判斷嚴重性,不解釋意圖,只做事實比對
版本: v1.0
建立: 2026-04-04 (台北時區)
建立者: ogt (首席架構師設計) + Claude Code (實作)
"""
from __future__ import annotations
import asyncio
import subprocess
import uuid
from pathlib import Path
from typing import Any
import structlog
import yaml
from src.models.drift import DriftItem, DriftLevel, DriftReport
logger = structlog.get_logger(__name__)
# 白名單欄位(靜默記錄,不告警)
_DEFAULT_ALLOWLIST_FIELDS = frozenset([
"spec.replicas",
"spec.template.spec.containers[*].resources.requests",
"spec.template.spec.containers[*].resources.limits",
"metadata.annotations",
"metadata.labels.pod-template-hash",
"metadata.resourceVersion",
"metadata.generation",
"metadata.uid",
"status",
])
# 關鍵欄位(必須立即告警)
_DEFAULT_CRITICAL_FIELDS = frozenset([
"spec.template.spec.containers[*].image",
"spec.template.spec.containers[*].env",
"spec.template.spec.containers[*].ports",
"spec.template.spec.volumes",
"spec.template.spec.serviceAccountName",
])
class GitStateReader:
"""從 Git HEAD 讀取 K8s YAML 狀態"""
def __init__(self, k8s_dir: str = "k8s"):
self._k8s_dir = Path(k8s_dir)
async def read(self, namespace: str) -> dict[str, Any]:
"""
讀取 Git HEAD 中指定 namespace 的所有 K8s YAML
Returns:
{resource_key: parsed_yaml_dict}
resource_key 格式: "{kind}/{name}"
"""
try:
result = await asyncio.get_event_loop().run_in_executor(
None, self._read_sync, namespace
)
return result
except Exception as e:
logger.warning("git_state_read_failed", namespace=namespace, error=str(e))
return {}
def _read_sync(self, namespace: str) -> dict[str, Any]:
resources: dict[str, Any] = {}
if not self._k8s_dir.exists():
logger.warning("k8s_dir_not_found", path=str(self._k8s_dir))
return resources
for yaml_file in self._k8s_dir.rglob("*.yaml"):
try:
with open(yaml_file) as f:
docs = list(yaml.safe_load_all(f))
for doc in docs:
if not doc or not isinstance(doc, dict):
continue
metadata = doc.get("metadata", {})
ns = metadata.get("namespace", "")
if ns and ns != namespace:
continue
kind = doc.get("kind", "")
name = metadata.get("name", "")
if kind and name:
key = f"{kind}/{name}"
resources[key] = doc
except Exception as e:
logger.debug("yaml_parse_failed", file=str(yaml_file), error=str(e))
return resources
class K8sStateReader:
"""從 kubectl 讀取 K8s 實際狀態"""
async def read(self, namespace: str) -> dict[str, Any]:
"""
透過 kubectl 取得指定 namespace 的實際狀態
Returns:
{resource_key: actual_resource_dict}
"""
try:
result = await asyncio.get_event_loop().run_in_executor(
None, self._read_sync, namespace
)
return result
except Exception as e:
logger.warning("k8s_state_read_failed", namespace=namespace, error=str(e))
return {}
def _read_sync(self, namespace: str) -> dict[str, Any]:
resources: dict[str, Any] = {}
resource_types = ["deployment", "service", "configmap", "ingress"]
for rtype in resource_types:
try:
proc = subprocess.run(
["kubectl", "get", rtype, "-n", namespace, "-o", "yaml"],
capture_output=True,
text=True,
timeout=30,
)
if proc.returncode != 0:
logger.debug("kubectl_failed", type=rtype, stderr=proc.stderr[:200])
continue
data = yaml.safe_load(proc.stdout)
if not data or data.get("kind") != "List":
continue
for item in data.get("items", []):
kind = item.get("kind", rtype.capitalize())
name = item.get("metadata", {}).get("name", "")
if name:
key = f"{kind}/{name}"
resources[key] = item
except subprocess.TimeoutExpired:
logger.warning("kubectl_timeout", type=rtype, namespace=namespace)
except Exception as e:
logger.warning("kubectl_error", type=rtype, error=str(e))
return resources
class DriftDetector:
"""
比對 Git vs K8s 實際狀態,輸出 DriftItem 列表
職責邊界:只做事實比對,不判斷嚴重性,不解釋意圖
"""
def __init__(
self,
k8s_dir: str = "k8s",
allowlist_fields: frozenset | None = None,
critical_fields: frozenset | None = None,
):
self._git_reader = GitStateReader(k8s_dir)
self._k8s_reader = K8sStateReader()
self._allowlist = allowlist_fields or _DEFAULT_ALLOWLIST_FIELDS
self._critical_fields = critical_fields or _DEFAULT_CRITICAL_FIELDS
async def scan(self, namespace: str, triggered_by: str = "cron") -> DriftReport:
"""
掃描指定 namespace 的漂移
Args:
namespace: K8s namespace
triggered_by: 觸發來源cron / webhook / api
Returns:
DriftReport含 DriftItem 列表,尚未分析 intent
"""
report_id = str(uuid.uuid4())[:8]
logger.info("drift_scan_start", namespace=namespace, report_id=report_id)
git_state, k8s_state = await asyncio.gather(
self._git_reader.read(namespace),
self._k8s_reader.read(namespace),
)
items: list[DriftItem] = []
# 比對 Git 中有的資源
for resource_key, git_resource in git_state.items():
actual_resource = k8s_state.get(resource_key)
if actual_resource is None:
# 資源在 Git 中存在但 K8s 中不存在(可能尚未部署)
logger.debug("resource_missing_in_k8s", resource=resource_key)
continue
kind, name = resource_key.split("/", 1)
diffs = self._diff_resources(git_resource, actual_resource, kind, name, namespace)
items.extend(diffs)
high_count = sum(1 for i in items if i.drift_level == DriftLevel.HIGH)
medium_count = sum(1 for i in items if i.drift_level == DriftLevel.MEDIUM)
info_count = sum(1 for i in items if i.drift_level == DriftLevel.INFO)
logger.info(
"drift_scan_done",
namespace=namespace,
report_id=report_id,
high=high_count,
medium=medium_count,
info=info_count,
)
return DriftReport(
report_id=report_id,
namespace=namespace,
items=items,
high_count=high_count,
medium_count=medium_count,
info_count=info_count,
triggered_by=triggered_by,
)
def _diff_resources(
self,
git_res: dict,
actual_res: dict,
kind: str,
name: str,
namespace: str,
) -> list[DriftItem]:
"""逐欄位比對兩個資源,回傳差異列表"""
items: list[DriftItem] = []
# 只比對 spec 層metadata 的動態欄位太多)
git_spec = git_res.get("spec", {})
actual_spec = actual_res.get("spec", {})
diffs = self._flatten_diff("spec", git_spec, actual_spec)
for field_path, (git_val, actual_val) in diffs.items():
is_allowlisted = self._is_allowlisted(field_path)
if is_allowlisted:
level = DriftLevel.INFO
elif self._is_critical(field_path):
level = DriftLevel.HIGH
else:
level = DriftLevel.MEDIUM
items.append(DriftItem(
resource_kind=kind,
resource_name=name,
namespace=namespace,
field_path=field_path,
git_value=git_val,
actual_value=actual_val,
drift_level=level,
is_allowlisted=is_allowlisted,
))
return items
def _flatten_diff(
self,
prefix: str,
git_dict: Any,
actual_dict: Any,
) -> dict[str, tuple[Any, Any]]:
"""遞迴展開並比對兩個 dict回傳 {field_path: (git_val, actual_val)}"""
diffs: dict[str, tuple[Any, Any]] = {}
if not isinstance(git_dict, dict) or not isinstance(actual_dict, dict):
if git_dict != actual_dict:
diffs[prefix] = (git_dict, actual_dict)
return diffs
all_keys = set(git_dict.keys()) | set(actual_dict.keys())
for key in all_keys:
path = f"{prefix}.{key}"
git_val = git_dict.get(key)
actual_val = actual_dict.get(key)
if git_val == actual_val:
continue
if isinstance(git_val, dict) and isinstance(actual_val, dict):
diffs.update(self._flatten_diff(path, git_val, actual_val))
else:
diffs[path] = (git_val, actual_val)
return diffs
@staticmethod
def _pattern_matches(pattern: str, field_path: str) -> bool:
"""
匹配 field_path 是否符合 pattern。
支援兩種萬用字元:
- [*] → 任意索引 (e.g. containers[*] 匹配 containers[0], containers[1])
- * → 任意字串段
2026-04-05 Claude Code: I4 修正 — 舊邏輯直接 strip [*] 導致
containers[*].image 無法匹配 containers[0].image (首席架構師 Review I4)
"""
import re as _re
# 將 pattern 轉為正則:[*] → \[\d+\]* → [^.]+
regex = _re.escape(pattern)
regex = regex.replace(r"\[\*\]", r"\[\d+\]")
regex = regex.replace(r"\*", r"[^.]+")
# 允許 pattern 是前綴field_path 可能更深,. 或 [ 或字串結尾均可)
return bool(_re.match(f"^{regex}(\\.|\\[|$)", field_path))
def _is_allowlisted(self, field_path: str) -> bool:
"""判斷欄位是否在白名單(靜默記錄不告警)"""
return any(self._pattern_matches(p, field_path) for p in self._allowlist)
def _is_critical(self, field_path: str) -> bool:
"""判斷欄位是否為關鍵欄位HIGH 等級)"""
return any(self._pattern_matches(p, field_path) for p in self._critical_fields)
# =============================================================================
# Singleton
# =============================================================================
_detector: DriftDetector | None = None
def get_drift_detector() -> DriftDetector:
global _detector
if _detector is None:
_detector = DriftDetector()
return _detector