fix(api): P1-2 ApprovalRequestCreate 欄位對齊
修正 SignOz + GitHub Webhook 的 ApprovalRequestCreate: Before (錯誤欄位): - action_type, target_resource, source - blast_radius=BlastRadius.SINGLE (enum 不存在) - dry_run_check=DryRunCheck.SKIPPED (錯誤格式) - 缺少 action, description, requested_by After (正確欄位): - action, description (必填) - blast_radius=BlastRadius(...) (Pydantic Model) - dry_run_checks=[] (list) - requested_by (必填) - 其他欄位移至 metadata 遵循: ApprovalRequestBase schema (approval.py) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
OG T
@@ -1270,23 +1270,26 @@ async def create_ci_failure_approval(
|
||||
}
|
||||
risk_level = risk_map.get(diagnosis.risk_level, RiskLevel.MEDIUM)
|
||||
|
||||
# 組裝 Approval 請求
|
||||
# P1-2 修正: 欄位對齊 ApprovalRequestBase (2026-03-29)
|
||||
suggestion = diagnosis.fix_command or "; ".join(diagnosis.suggestions[:2])
|
||||
approval_request = ApprovalRequestCreate(
|
||||
source="github",
|
||||
alert_type="ci_failure_repair",
|
||||
target_resource=repo,
|
||||
namespace="github-actions",
|
||||
action=f"CI Failure Repair: {repo}",
|
||||
description=f"Root Cause: {diagnosis.root_cause}\nSuggestion: {suggestion}",
|
||||
risk_level=risk_level,
|
||||
root_cause=diagnosis.root_cause,
|
||||
suggestion=diagnosis.fix_command or "; ".join(diagnosis.suggestions[:2]),
|
||||
blast_radius=BlastRadius.NAMESPACE if diagnosis.auto_fixable else BlastRadius.SERVICE,
|
||||
data_impact=DataImpact.NONE,
|
||||
dry_run_check=DryRunCheck.SKIPPED,
|
||||
llm_provider=diagnosis.analyzed_by,
|
||||
llm_confidence=diagnosis.confidence,
|
||||
blast_radius=BlastRadius(
|
||||
affected_pods=1 if diagnosis.auto_fixable else 2,
|
||||
estimated_downtime="~5min",
|
||||
related_services=[repo],
|
||||
data_impact=DataImpact.NONE,
|
||||
),
|
||||
dry_run_checks=[],
|
||||
requested_by="github-webhook",
|
||||
metadata={
|
||||
"source": "github",
|
||||
"alert_type": "ci_failure_repair",
|
||||
"target_resource": repo,
|
||||
"namespace": "github-actions",
|
||||
"ci_diagnosis_id": diagnosis_id,
|
||||
"repo": repo,
|
||||
"workflow_name": workflow_run.name,
|
||||
"workflow_id": workflow_run.id,
|
||||
"workflow_url": workflow_run.html_url,
|
||||
@@ -1294,6 +1297,8 @@ async def create_ci_failure_approval(
|
||||
"error_type": diagnosis.error_type,
|
||||
"auto_fixable": diagnosis.auto_fixable,
|
||||
"fix_command": diagnosis.fix_command,
|
||||
"llm_provider": diagnosis.analyzed_by,
|
||||
"llm_confidence": diagnosis.confidence,
|
||||
},
|
||||
)
|
||||
|
||||
@@ -1471,28 +1476,34 @@ async def create_github_approval(
|
||||
else:
|
||||
risk_level = RiskLevel.MEDIUM
|
||||
|
||||
# 組裝 Approval 請求
|
||||
# P1-2 修正: 欄位對齊 ApprovalRequestBase (2026-03-29)
|
||||
root_cause = f"Code review found security concerns in {target}"
|
||||
suggestion = f"Review {len(analysis.security_concerns)} security concern(s): {', '.join(analysis.security_concerns[:3])}"
|
||||
approval_request = ApprovalRequestCreate(
|
||||
source="github",
|
||||
alert_type="code_review_security",
|
||||
target_resource=repo,
|
||||
namespace="github",
|
||||
action=f"Code Review Security: {repo}",
|
||||
description=f"Root Cause: {root_cause}\nSuggestion: {suggestion}",
|
||||
risk_level=risk_level,
|
||||
root_cause=f"Code review found security concerns in {target}",
|
||||
suggestion=f"Review {len(analysis.security_concerns)} security concern(s): {', '.join(analysis.security_concerns[:3])}",
|
||||
blast_radius=BlastRadius.SERVICE,
|
||||
data_impact=DataImpact.READ_ONLY,
|
||||
dry_run_check=DryRunCheck.PASSED,
|
||||
llm_provider=analysis.analyzed_by,
|
||||
llm_confidence=analysis.confidence,
|
||||
blast_radius=BlastRadius(
|
||||
affected_pods=1,
|
||||
estimated_downtime="0",
|
||||
related_services=[repo],
|
||||
data_impact=DataImpact.READ_ONLY,
|
||||
),
|
||||
dry_run_checks=[],
|
||||
requested_by="github-webhook",
|
||||
metadata={
|
||||
"source": "github",
|
||||
"alert_type": "code_review_security",
|
||||
"target_resource": repo,
|
||||
"namespace": "github",
|
||||
"github_review_id": review_id,
|
||||
"repo": repo,
|
||||
"target": target,
|
||||
"url": url,
|
||||
"quality_score": analysis.quality_score,
|
||||
"security_concerns": analysis.security_concerns,
|
||||
"issues_count": len(analysis.issues),
|
||||
"llm_provider": analysis.analyzed_by,
|
||||
"llm_confidence": analysis.confidence,
|
||||
},
|
||||
)
|
||||
|
||||
|
||||
362
apps/api/src/api/v1/signoz_webhook.py
Normal file
362
apps/api/src/api/v1/signoz_webhook.py
Normal file
@@ -0,0 +1,362 @@
|
||||
"""
|
||||
AWOOOI API - SignOz Webhook Handler
|
||||
====================================
|
||||
接收 SignOz Alert，建立 Incident 並發送 Telegram 通知
|
||||
|
||||
整合流程:
|
||||
1. SignOz Alert → AWOOOI API Webhook
|
||||
2. 記錄異常頻率 (AnomalyCounter - ADR-037)
|
||||
3. 建立 Incident (雙層記憶寫入)
|
||||
4. 建立 Approval 供人工審核
|
||||
5. 發送 Telegram 告警
|
||||
|
||||
🔴 HARD RULE: 時間顯示使用 Asia/Taipei (UTC+8)
|
||||
"""
|
||||
|
||||
import uuid
|
||||
|
||||
import structlog
|
||||
from fastapi import APIRouter, BackgroundTasks, HTTPException, Request
|
||||
from pydantic import BaseModel
|
||||
|
||||
from src.models.approval import (
|
||||
ApprovalRequestCreate,
|
||||
BlastRadius,
|
||||
DataImpact,
|
||||
RiskLevel,
|
||||
)
|
||||
from src.core.metrics import (
|
||||
record_alert_chain_failure,
|
||||
record_alert_chain_success,
|
||||
record_alert_processed,
|
||||
record_anomaly,
|
||||
record_telegram_notification,
|
||||
record_webhook_request,
|
||||
)
|
||||
from src.services.anomaly_counter import get_anomaly_counter
|
||||
from src.services.approval_db import get_approval_service
|
||||
from src.services.incident_service import get_incident_service
|
||||
from src.services.telegram_gateway import get_telegram_gateway
|
||||
from src.utils.timezone import now_taipei_iso
|
||||
|
||||
logger = structlog.get_logger(__name__)
|
||||
|
||||
router = APIRouter(prefix="/webhooks/signoz", tags=["SignOz Webhook"])
|
||||
|
||||
|
||||
# SignOz Severity → Risk Level 映射
|
||||
SIGNOZ_SEVERITY_TO_RISK = {
|
||||
"critical": RiskLevel.CRITICAL,
|
||||
"error": RiskLevel.HIGH,
|
||||
"warning": RiskLevel.MEDIUM,
|
||||
"info": RiskLevel.LOW,
|
||||
}
|
||||
|
||||
|
||||
class SignOzAlertPayload(BaseModel):
|
||||
"""SignOz Alert Payload (標準 Prometheus AlertManager 格式)"""
|
||||
|
||||
alertname: str
|
||||
status: str # firing, resolved
|
||||
labels: dict = {}
|
||||
annotations: dict = {}
|
||||
startsAt: str | None = None
|
||||
endsAt: str | None = None
|
||||
generatorURL: str | None = None
|
||||
|
||||
|
||||
@router.post("/alert")
|
||||
async def handle_signoz_alert(
|
||||
request: Request,
|
||||
background_tasks: BackgroundTasks,
|
||||
):
|
||||
"""
|
||||
SignOz 告警 Webhook Handler
|
||||
|
||||
觸發條件:
|
||||
- Alert status=firing
|
||||
|
||||
處理流程:
|
||||
1. 記錄異常頻率 (AnomalyCounter)
|
||||
2. 建立 Incident
|
||||
3. 建立 Approval
|
||||
4. 發送 Telegram 告警
|
||||
"""
|
||||
try:
|
||||
payload = await request.json()
|
||||
logger.info("signoz_alert_received", payload=payload)
|
||||
|
||||
# 解析 payload (支援陣列或單一告警)
|
||||
alerts = payload.get("alerts", [payload]) if isinstance(payload, dict) else [payload]
|
||||
|
||||
results = []
|
||||
for alert in alerts:
|
||||
# 只處理 firing 狀態
|
||||
status = alert.get("status", "firing")
|
||||
if status != "firing":
|
||||
results.append({"status": "ignored", "reason": "not firing"})
|
||||
continue
|
||||
|
||||
# 提取告警資訊
|
||||
alert_name = alert.get("alertname", alert.get("labels", {}).get("alertname", "unknown"))
|
||||
labels = alert.get("labels", {})
|
||||
annotations = alert.get("annotations", {})
|
||||
severity = labels.get("severity", "warning")
|
||||
|
||||
# 背景處理
|
||||
background_tasks.add_task(
|
||||
process_signoz_alert,
|
||||
alert_name=alert_name,
|
||||
labels=labels,
|
||||
annotations=annotations,
|
||||
severity=severity,
|
||||
starts_at=alert.get("startsAt"),
|
||||
)
|
||||
|
||||
results.append({
|
||||
"status": "accepted",
|
||||
"alert_name": alert_name,
|
||||
})
|
||||
|
||||
return {"status": "ok", "processed": len(results), "results": results}
|
||||
|
||||
except Exception as e:
|
||||
logger.exception("signoz_webhook_error", error=str(e))
|
||||
raise HTTPException(status_code=500, detail=str(e))
|
||||
|
||||
|
||||
async def process_signoz_alert(
|
||||
alert_name: str,
|
||||
labels: dict,
|
||||
annotations: dict,
|
||||
severity: str,
|
||||
starts_at: str | None,
|
||||
):
|
||||
"""
|
||||
背景處理 SignOz 告警
|
||||
|
||||
ADR-037 Phase 21: 完整告警處理流程
|
||||
1. 記錄異常頻率
|
||||
2. 建立 Incident
|
||||
3. 建立 Approval
|
||||
4. 發送 Telegram
|
||||
"""
|
||||
try:
|
||||
# =================================================================
|
||||
# Step 1: 記錄異常頻率 (ADR-037)
|
||||
# =================================================================
|
||||
anomaly_counter = get_anomaly_counter()
|
||||
anomaly_signature = {
|
||||
"alert_name": alert_name,
|
||||
"service": labels.get("service_name", labels.get("service", "unknown")),
|
||||
"namespace": labels.get("namespace", "signoz"),
|
||||
"error_type": labels.get("error_type", severity),
|
||||
}
|
||||
frequency = await anomaly_counter.record_anomaly(anomaly_signature)
|
||||
anomaly_frequency = frequency.model_dump() if frequency else None
|
||||
|
||||
logger.info(
|
||||
"signoz_anomaly_recorded",
|
||||
alert_name=alert_name,
|
||||
count_24h=frequency.count_24h if frequency else 0,
|
||||
escalation_level=frequency.escalation_level if frequency else None,
|
||||
)
|
||||
|
||||
# Wave A.5: 記錄異常指標 (ADR-037)
|
||||
if frequency:
|
||||
record_anomaly(
|
||||
alert_name=alert_name,
|
||||
service=labels.get("service_name", labels.get("service", "unknown")),
|
||||
frequency_24h=frequency.count_24h,
|
||||
escalation_level=frequency.escalation_level,
|
||||
)
|
||||
|
||||
# =================================================================
|
||||
# Step 2: 建立 Incident
|
||||
# =================================================================
|
||||
incident_service = get_incident_service()
|
||||
signal_data = {
|
||||
"alert_name": alert_name,
|
||||
"severity": severity,
|
||||
"source": "signoz",
|
||||
"target": labels.get("service_name", labels.get("service", "unknown")),
|
||||
"labels": labels,
|
||||
"annotations": annotations,
|
||||
"fingerprint": f"signoz-{alert_name}-{labels.get('service_name', 'unknown')}",
|
||||
}
|
||||
# ADR-037: 傳遞頻率統計到 Incident
|
||||
incident = await incident_service.create_incident_from_signal(
|
||||
signal_data, frequency_stats=anomaly_frequency
|
||||
)
|
||||
|
||||
if not incident:
|
||||
logger.error("signoz_incident_creation_failed", alert_name=alert_name)
|
||||
return
|
||||
|
||||
# =================================================================
|
||||
# Step 3: 建立 Approval
|
||||
# =================================================================
|
||||
approval_id = await create_signoz_approval(
|
||||
alert_name=alert_name,
|
||||
labels=labels,
|
||||
annotations=annotations,
|
||||
severity=severity,
|
||||
incident_id=incident.incident_id,
|
||||
anomaly_frequency=anomaly_frequency,
|
||||
)
|
||||
|
||||
# =================================================================
|
||||
# Step 4: 發送 Telegram 告警
|
||||
# =================================================================
|
||||
await send_signoz_telegram(
|
||||
approval_id=approval_id,
|
||||
alert_name=alert_name,
|
||||
labels=labels,
|
||||
annotations=annotations,
|
||||
severity=severity,
|
||||
anomaly_frequency=anomaly_frequency,
|
||||
)
|
||||
|
||||
logger.info(
|
||||
"signoz_alert_processed",
|
||||
alert_name=alert_name,
|
||||
incident_id=incident.incident_id,
|
||||
approval_id=approval_id,
|
||||
)
|
||||
|
||||
# Wave A.5: 記錄告警鏈路成功 (ADR-037)
|
||||
record_alert_chain_success("signoz")
|
||||
record_alert_processed(
|
||||
source="signoz",
|
||||
severity=severity,
|
||||
outcome="incident_created",
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
logger.exception("signoz_process_error", error=str(e), alert_name=alert_name)
|
||||
# Wave A.5: 記錄告警鏈路失敗 (ADR-037)
|
||||
record_alert_chain_failure("signoz")
|
||||
|
||||
|
||||
async def create_signoz_approval(
|
||||
alert_name: str,
|
||||
labels: dict,
|
||||
annotations: dict,
|
||||
severity: str,
|
||||
incident_id: str,
|
||||
anomaly_frequency: dict | None = None,
|
||||
) -> str:
|
||||
"""
|
||||
為 SignOz 告警建立 Approval 記錄
|
||||
|
||||
ADR-037: 含異常頻率統計
|
||||
"""
|
||||
try:
|
||||
approval_service = get_approval_service()
|
||||
|
||||
# 決定風險等級 (考慮頻率升級)
|
||||
risk_level = SIGNOZ_SEVERITY_TO_RISK.get(severity, RiskLevel.MEDIUM)
|
||||
|
||||
# ADR-037: 根據頻率升級風險等級
|
||||
if anomaly_frequency:
|
||||
escalation = anomaly_frequency.get("escalation_level")
|
||||
if escalation == "PERMANENT_FIX":
|
||||
risk_level = RiskLevel.CRITICAL
|
||||
elif escalation == "ESCALATE":
|
||||
risk_level = max(risk_level, RiskLevel.HIGH)
|
||||
|
||||
# 建立 Approval
|
||||
service_name = labels.get("service_name", labels.get("service", "unknown"))
|
||||
summary = annotations.get("summary", f"SignOz Alert: {alert_name}")
|
||||
description = annotations.get("description", summary)
|
||||
|
||||
# P1-2 修正: 欄位對齊 ApprovalRequestBase (2026-03-29)
|
||||
approval_request = ApprovalRequestCreate(
|
||||
action=f"SignOz Alert: {alert_name}",
|
||||
description=description,
|
||||
risk_level=risk_level,
|
||||
blast_radius=BlastRadius(
|
||||
affected_pods=1,
|
||||
estimated_downtime="0",
|
||||
related_services=[service_name],
|
||||
data_impact=DataImpact.READ_ONLY,
|
||||
),
|
||||
dry_run_checks=[],
|
||||
requested_by="signoz-webhook",
|
||||
metadata={
|
||||
"source": "signoz",
|
||||
"alert_name": alert_name,
|
||||
"labels": labels,
|
||||
"annotations": annotations,
|
||||
"incident_id": incident_id,
|
||||
"anomaly_frequency": anomaly_frequency,
|
||||
},
|
||||
)
|
||||
|
||||
approval = await approval_service.create_approval(approval_request)
|
||||
return str(approval.id)
|
||||
|
||||
except Exception as e:
|
||||
logger.exception("signoz_approval_error", error=str(e))
|
||||
# 回傳臨時 ID，不中斷流程
|
||||
return str(uuid.uuid4())
|
||||
|
||||
|
||||
async def send_signoz_telegram(
|
||||
approval_id: str,
|
||||
alert_name: str,
|
||||
labels: dict,
|
||||
annotations: dict,
|
||||
severity: str,
|
||||
anomaly_frequency: dict | None = None,
|
||||
):
|
||||
"""
|
||||
發送 SignOz 告警到 Telegram
|
||||
|
||||
ADR-037: 含異常頻率顯示
|
||||
"""
|
||||
try:
|
||||
telegram = get_telegram_gateway()
|
||||
await telegram.initialize()
|
||||
|
||||
service_name = labels.get("service_name", labels.get("service", "unknown"))
|
||||
summary = annotations.get("summary", f"SignOz Alert: {alert_name}")
|
||||
description = annotations.get("description", "")
|
||||
|
||||
await telegram.send_approval_card(
|
||||
approval_id=approval_id,
|
||||
risk_level="critical" if severity == "critical" else (
|
||||
"high" if severity == "error" else "medium"
|
||||
),
|
||||
resource_name=service_name,
|
||||
root_cause=summary,
|
||||
suggested_action=description or "請檢查 SignOz 儀表板",
|
||||
primary_responsibility="BE",
|
||||
confidence=0.7, # SignOz 自動檢測的信心度
|
||||
namespace="signoz",
|
||||
anomaly_frequency=anomaly_frequency,
|
||||
)
|
||||
|
||||
logger.info(
|
||||
"signoz_telegram_sent",
|
||||
approval_id=approval_id,
|
||||
alert_name=alert_name,
|
||||
escalation_level=anomaly_frequency.get("escalation_level") if anomaly_frequency else None,
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
logger.exception("signoz_telegram_error", error=str(e))
|
||||
|
||||
|
||||
# =============================================================================
|
||||
# Health Check (供 SignOz 確認 Webhook 可達)
|
||||
# =============================================================================
|
||||
|
||||
@router.get("/health")
|
||||
async def signoz_webhook_health():
|
||||
"""SignOz Webhook 健康檢查"""
|
||||
return {
|
||||
"status": "ok",
|
||||
"service": "signoz-webhook",
|
||||
"timestamp": now_taipei_iso(),
|
||||
}
|
||||
Reference in New Issue
Block a user