feat(ci): P2-2 ChatOps集成 - 飞书机器人对接Gitea Actions #533

Merged
auto-approve-bot merged 1 commits from ci/chatops-integration into develop 2026-07-18 18:42:36 +08:00
7 changed files with 1618 additions and 0 deletions
+19
View File
@@ -0,0 +1,19 @@
"""CI ChatOps 工具包 - 飞书机器人对接 Gitea Actions
模块:
config - 配置管理(环境变量)
gitea_client - Gitea API 客户端封装
feishu_notify - 飞书通知(失败/恢复/E2E摘要)
ci_query - CI 状态查询
ci_trigger - CI 重跑触发
webhook_server - Gitea webhook 接收服务(FastAPI
"""
__all__ = [
"config",
"gitea_client",
"feishu_notify",
"ci_query",
"ci_trigger",
"webhook_server",
]
+296
View File
@@ -0,0 +1,296 @@
#!/usr/bin/env python3
"""
CI 状态查询模块 - 查询 run 列表、某分支/某 PR 的 CI 状态、失败详情
支持查询类型:
- list_runs: 列出最近的 workflow runs
- branch_status: 某分支最新 CI 状态
- pr_status: 某 PR 的 CI 状态
- failure_detail: 某次 run 的失败详情
用法:
python3 scripts/ci/chatops/ci_query.py --branch develop
python3 scripts/ci/chatops/ci_query.py --pr 123
python3 scripts/ci/chatops/ci_query.py --run-id 456 --detail
设计:
- 与飞书机器人 /ci status 命令对接
- 返回结构化数据,上层负责格式化输出
"""
import argparse
import sys
from . import config
from .gitea_client import GiteaClient
class CIQuery:
"""CI 状态查询器"""
def __init__(self, gitea_client=None):
self.gitea = gitea_client or GiteaClient()
# ── 查询方法 ──────────────────────────────────────
def get_branch_status(self, branch, limit=5):
"""获取指定分支最新的 CI 状态
Returns:
dict: {branch, latest_run, recent_runs, overall_status}
"""
runs, total = self.gitea.list_runs(branch=branch, limit=limit)
if not runs:
return {
"branch": branch,
"latest_run": None,
"recent_runs": [],
"overall_status": "no_runs",
"total_count": total,
}
latest = runs[0]
overall = self._derive_overall_status(runs)
return {
"branch": branch,
"latest_run": latest,
"recent_runs": runs,
"overall_status": overall,
"total_count": total,
}
def get_pr_status(self, pr_number):
"""获取指定 PR 的 CI 状态
Returns:
dict: {pr_number, pr_title, runs, overall_status}
"""
pr = self.gitea.get_pr(pr_number)
if not pr:
return {
"pr_number": pr_number,
"pr_title": "未知",
"runs": [],
"overall_status": "pr_not_found",
}
pr_title = pr.get("title", "")
runs = self.gitea.get_pr_ci_runs(pr_number, limit=10)
overall = self._derive_overall_status(runs) if runs else "no_runs"
return {
"pr_number": pr_number,
"pr_title": pr_title,
"runs": runs,
"overall_status": overall,
"head_sha": pr.get("head", {}).get("sha", ""),
}
def get_failure_detail(self, run_id):
"""获取某次 run 的失败详情
Returns:
dict: {run_info, failed_jobs, summary}
"""
run = self.gitea.get_run(run_id)
if not run:
return {"run_info": None, "failed_jobs": [], "summary": "Run not found"}
failed_jobs = self.gitea.get_failed_jobs_summary(run_id, max_lines_per_job=30)
summary_parts = []
for job in failed_jobs:
step = f"(步骤: {job['failed_step']}" if job["failed_step"] else ""
summary_parts.append(f"{job['name']}{step}")
summary = "\n".join(summary_parts) if summary_parts else "无失败 job(可能还在运行中)"
return {
"run_info": run,
"failed_jobs": failed_jobs,
"summary": summary,
"total_jobs": len(self.gitea.get_run_jobs(run_id)),
}
def list_recent_runs(self, status=None, branch=None, limit=10):
"""列出最近的 runs"""
runs, total = self.gitea.list_runs(status=status, branch=branch, limit=limit)
return {"runs": runs, "total_count": total}
# ── 辅助方法 ────────────────────────────────────
@staticmethod
def _derive_overall_status(runs):
"""根据最近 runs 推导整体状态
Returns:
success: 最近一次成功
failing: 最近一次失败(连续失败)
flaky: 有失败有成功(最近一次失败
running: 有正在运行的
unknown: 未知
"""
if not runs:
return "no_runs"
# 检查是否有运行中的
running = [r for r in runs if r.get("status") != "completed"]
if running:
return "running"
# 看最近一次
latest = runs[0]
latest_conclusion = latest.get("conclusion", "unknown")
if latest_conclusion == "success":
return "success"
if latest_conclusion == "failure":
# 检查是否连续失败
consecutive_failures = 0
for r in runs:
if r.get("conclusion") == "failure":
consecutive_failures += 1
else:
break
# 看之前有没有成功
has_success = any(r.get("conclusion") == "success" for r in runs)
if has_success:
return "flaky"
return "failing"
return "unknown"
# ── 格式化输出 ────────────────────────────────────
@staticmethod
def format_branch_status(status_data):
"""格式化分支状态为人类可读文本"""
branch = status_data["branch"]
latest = status_data["latest_run"]
overall = status_data["overall_status"]
status_emoji = {
"success": "",
"failing": "🔴",
"flaky": "🟡",
"running": "🔄",
"no_runs": "",
"unknown": "",
}.get(overall, "")
lines = [f"**CI 状态:{branch} 分支**", f"整体状态: {status_emoji} {overall}"]
if latest:
name = latest.get("name", "Unknown")
conclusion = latest.get("conclusion", latest.get("status", "unknown"))
run_id = latest.get("id", "")
created = latest.get("created_at", "")[:16].replace("T", " ")
run_url = f"{config.GITEA_URL}/{config.GITEA_REPO}/actions/runs/{run_id}"
lines.append(f"最新: [{name} #{run_id}]({run_url}) - {conclusion} ({created})")
recent = status_data["recent_runs"]
if len(recent) > 1:
lines.append(f"\n最近 {len(recent)} 次:")
for r in recent[:5]:
c = r.get("conclusion", r.get("status", "?"))
emoji = {"success": "", "failure": "", "skipped": "⏭️"}.get(c, "🔄")
lines.append(f" {emoji} #{r.get('id', '?')} {r.get('name', '?')[:30]} - {c}")
return "\n".join(lines)
@staticmethod
def format_pr_status(status_data):
"""格式化 PR 状态为人类可读文本"""
pr_num = status_data["pr_number"]
pr_title = status_data["pr_title"]
overall = status_data["overall_status"]
status_emoji = {
"success": "",
"failing": "🔴",
"flaky": "🟡",
"running": "🔄",
"no_runs": "",
"pr_not_found": "",
"unknown": "",
}.get(overall, "")
pr_url = f"{config.GITEA_URL}/{config.GITEA_REPO}/pulls/{pr_num}"
lines = [
f"**CI 状态:PR #{pr_num}**",
f"标题: [{pr_title}]({pr_url})",
f"状态: {status_emoji} {overall}",
]
runs = status_data["runs"]
if runs:
lines.append(f"\nCI Runs ({len(runs)}):")
for r in runs[:5]:
c = r.get("conclusion", r.get("status", "?"))
emoji = {"success": "", "failure": "", "skipped": "⏭️"}.get(c, "🔄")
run_id = r.get("id", "?")
run_url = f"{config.GITEA_URL}/{config.GITEA_REPO}/actions/runs/{run_id}"
lines.append(f" {emoji} [{r.get('name', '?')[:30]} #{run_id}]({run_url}) - {c}")
return "\n".join(lines)
# ── CLI 入口 ──────────────────────────────────────────
def main():
parser = argparse.ArgumentParser(description="CI 状态查询")
parser.add_argument("--branch", help="查询指定分支的 CI 状态")
parser.add_argument("--pr", type=int, help="查询指定 PR 的 CI 状态")
parser.add_argument("--run-id", help="查询指定 run 的详情")
parser.add_argument("--detail", action="store_true", help="显示失败详情")
parser.add_argument("--limit", type=int, default=5, help="返回数量限制")
parser.add_argument("--status", help="按状态过滤: success/failure/running")
args = parser.parse_args()
query = CIQuery()
if args.run_id:
if args.detail:
result = query.get_failure_detail(args.run_id)
print(f"Run #{args.run_id} 失败详情:")
print(result["summary"])
if result["failed_jobs"]:
print("\n详细日志尾部:")
for job in result["failed_jobs"]:
print(f"\n--- {job['name']} ---")
print(job["log_tail"][:500] if job["log_tail"] else "无日志")
else:
run = query.gitea.get_run(args.run_id)
if run:
print(f"Run #{args.run_id}: {run.get('name')} - {run.get('conclusion', run.get('status'))}")
print(f"分支: {run.get('head_branch', '?')}")
print(f"触发: {run.get('event', '?')}")
else:
print(f"Run {args.run_id} 不存在")
elif args.pr:
result = query.get_pr_status(args.pr)
print(CIQuery.format_pr_status(result))
elif args.branch:
result = query.get_branch_status(args.branch, limit=args.limit)
print(CIQuery.format_branch_status(result))
elif args.status:
result = query.list_recent_runs(status=args.status, limit=args.limit)
for r in result["runs"]:
print(
f"#{r.get('id')} {r.get('name')[:40]} - {r.get('conclusion', r.get('status'))} ({r.get('head_branch', '?')})"
)
else:
parser.print_help()
return 1
return 0
if __name__ == "__main__":
sys.exit(main())
+169
View File
@@ -0,0 +1,169 @@
#!/usr/bin/env python3
"""
CI 触发模块 - 重新运行失败 job、重跑整个 workflow、取消 run
支持操作:
- rerun_failed: 重跑失败的 jobs
- rerun_all: 重跑整个 workflow
- cancel: 取消运行中的 run
用法:
python3 scripts/ci/chatops/ci_trigger.py --run-id 456 --action rerun_failed
python3 scripts/ci/chatops/ci_trigger.py --run-id 456 --action rerun_all
python3 scripts/ci/chatops/ci_trigger.py --run-id 456 --action cancel
设计:
- 与飞书机器人 /ci rerun 命令对接
- 操作前自动校验 run 状态,避免无效操作
"""
import argparse
import sys
from . import config
from .gitea_client import GiteaClient
class CITrigger:
"""CI 操作触发器"""
def __init__(self, gitea_client=None):
self.gitea = gitea_client or GiteaClient()
# ── 触发操作 ─────────────────────────────────────
def rerun_failed(self, run_id):
"""重跑失败的 jobs
Returns:
dict: {success, message, new_run_id?}
"""
run = self.gitea.get_run(run_id)
if not run:
return {"success": False, "message": f"Run {run_id} 不存在"}
status = run.get("status", "")
if status != "completed":
return {
"success": False,
"message": f"Run {run_id} 当前状态为 {status},仅 completed 状态才能重跑",
}
result = self.gitea.rerun_failed_jobs(run_id)
if result is None:
return {"success": False, "message": "重跑请求失败"}
# Gitea rerun 后返回的 run id 通常不变(复用原 run)
return {
"success": True,
"message": f"已触发重跑失败 jobs: Run #{run_id}",
"run_id": run_id,
"run_url": f"{config.GITEA_URL}/{config.GITEA_REPO}/actions/runs/{run_id}",
}
def rerun_all(self, run_id):
"""重跑整个 workflow run
Returns:
dict: {success, message, run_id, run_url}
"""
run = self.gitea.get_run(run_id)
if not run:
return {"success": False, "message": f"Run {run_id} 不存在"}
status = run.get("status", "")
if status == "running" or status == "pending":
return {
"success": False,
"message": f"Run {run_id} 正在运行中,无需重跑",
}
result = self.gitea.rerun_run(run_id)
if result is None:
return {"success": False, "message": "重跑请求失败"}
return {
"success": True,
"message": f"已触发完整重跑: Run #{run_id}",
"run_id": run_id,
"run_url": f"{config.GITEA_URL}/{config.GITEA_REPO}/actions/runs/{run_id}",
}
def cancel_run(self, run_id):
"""取消运行中的 run
Returns:
dict: {success, message}
"""
run = self.gitea.get_run(run_id)
if not run:
return {"success": False, "message": f"Run {run_id} 不存在"}
status = run.get("status", "")
if status == "completed":
return {
"success": False,
"message": f"Run {run_id} 已完成,无需取消",
}
result = self.gitea.cancel_run(run_id)
if result is None:
return {"success": False, "message": "取消请求失败"}
return {
"success": True,
"message": f"已取消 Run #{run_id}",
"run_id": run_id,
}
def rerun_latest_failed(self, branch="develop", workflow_id=None):
"""重跑指定分支最近一次失败的 run
用于快速恢复场景,不需要先查 run_id
"""
runs, _ = self.gitea.list_runs(branch=branch, workflow_id=workflow_id, status="failure", limit=5)
if not runs:
return {"success": False, "message": f"{branch} 分支没有失败的 run"}
latest = runs[0]
run_id = latest.get("id")
return self.rerun_failed(run_id)
# ── CLI 入口 ──────────────────────────────────────────
def main():
parser = argparse.ArgumentParser(description="CI 触发操作")
parser.add_argument("--run-id", required=True, help="Workflow Run ID")
parser.add_argument(
"--action",
required=True,
choices=["rerun_failed", "rerun_all", "cancel"],
help="操作类型",
)
args = parser.parse_args()
trigger = CITrigger()
if args.action == "rerun_failed":
result = trigger.rerun_failed(args.run_id)
elif args.action == "rerun_all":
result = trigger.rerun_all(args.run_id)
elif args.action == "cancel":
result = trigger.cancel_run(args.run_id)
else:
print(f"未知操作: {args.action}")
return 1
status = "" if result["success"] else ""
print(f"{status} {result['message']}")
if result.get("run_url"):
print(f" {result['run_url']}")
return 0 if result["success"] else 1
if __name__ == "__main__":
sys.exit(main())
+55
View File
@@ -0,0 +1,55 @@
#!/usr/bin/env python3
"""
ChatOps 配置管理 - 统一从环境变量读取配置,不硬编码任何敏感信息
环境变量:
GITEA_URL Gitea 地址 (默认 https://git.xiaoxiajianji.com)
GITEA_REPO 仓库路径 (默认 xiaoxia/xiaoxia-saas)
GITEA_TOKEN Gitea API Token (优先使用)
GITEA_USERNAME Gitea 用户名 (密码认证时)
GITEA_PASSWORD Gitea 密码 (密码认证时)
FEISHU_WEBHOOK_URL 飞书自定义机器人 webhook 地址
FEISHU_APP_ID 飞书应用 App ID (应用机器人模式,预留)
FEISHU_APP_SECRET 飞书应用 App Secret (应用机器人模式,预留)
CHATOPS_NOTIFY_BRANCHES 触发通知的分支,逗号分隔 (默认 main,develop)
CHATOPS_WEBHOOK_PORT webhook 服务监听端口 (默认 8090)
CHATOPS_WEBHOOK_SECRET Gitea webhook 密钥 (校验签名,可选)
"""
import os
# ── Gitea 配置 ────────────────────────────────────────
GITEA_URL = os.environ.get("GITEA_URL", "https://git.xiaoxiajianji.com").rstrip("/")
GITEA_REPO = os.environ.get("GITEA_REPO", "xiaoxia/xiaoxia-saas")
GITEA_TOKEN = os.environ.get("GITEA_TOKEN", "")
GITEA_USERNAME = os.environ.get("GITEA_USERNAME", "")
GITEA_PASSWORD = os.environ.get("GITEA_PASSWORD", "")
# ── 飞书配置 ──────────────────────────────────────────
FEISHU_WEBHOOK_URL = os.environ.get("FEISHU_WEBHOOK_URL", "")
FEISHU_APP_ID = os.environ.get("FEISHU_APP_ID", "")
FEISHU_APP_SECRET = os.environ.get("FEISHU_APP_SECRET", "")
# ── 通知配置 ──────────────────────────────────────────
NOTIFY_BRANCHES = [b.strip() for b in os.environ.get("CHATOPS_NOTIFY_BRANCHES", "main,develop").split(",") if b.strip()]
# ── Webhook 服务配置 ──────────────────────────────────
WEBHOOK_PORT = int(os.environ.get("CHATOPS_WEBHOOK_PORT", "8090"))
WEBHOOK_SECRET = os.environ.get("CHATOPS_WEBHOOK_SECRET", "")
# ── 常量 ──────────────────────────────────────────────
PAGE_LIMIT = 50 # Gitea API 每页最大数量
def has_gitea_auth() -> bool:
"""检查是否配置了 Gitea 认证信息"""
if GITEA_TOKEN:
return True
if GITEA_USERNAME and GITEA_PASSWORD:
return True
return False
def has_feishu_webhook() -> bool:
"""检查是否配置了飞书 webhook"""
return bool(FEISHU_WEBHOOK_URL)
+390
View File
@@ -0,0 +1,390 @@
#!/usr/bin/env python3
"""
飞书通知模块 - CI 关键事件推送
支持通知类型:
- branch_failure: main/develop 分支 CI 失败
- branch_recovery: main/develop 分支 CI 从失败恢复(绿色恢复)
- e2e_failure: E2E 测试失败摘要
- pr_failure: PR CI 失败(可选)
用法:
# 命令行直接调用(供 CI workflow 使用)
python3 -m scripts.ci.chatops.feishu_notify --mode failure --run-id 12345
python3 scripts/ci/chatops/feishu_notify.py --mode recovery --run-id 12345
# Python 模块调用
from scripts.ci.chatops.feishu_notify import FeishuNotifier
notifier = FeishuNotifier()
notifier.notify_branch_failure(run_id=12345, branch="develop")
设计原则:
1. 通知失败永远不阻断主流程(返回 0)
2. 卡片信息丰富,一键跳转 Gitea 详情页
3. 失败通知包含错误摘要,不用点进去就能判断严重程度
"""
import argparse
import json
import sys
import urllib.request
from datetime import datetime
from . import config
from .gitea_client import GiteaClient
class FeishuNotifier:
"""飞书通知发送器"""
def __init__(self, webhook_url=None, gitea_client=None):
self.webhook_url = webhook_url or config.FEISHU_WEBHOOK_URL
self.gitea = gitea_client or GiteaClient()
def _send_card(self, card_payload):
"""发送飞书卡片消息
Returns:
True 表示发送成功(飞书返回 code=0)
"""
if not self.webhook_url:
print("[INFO] 未配置 FEISHU_WEBHOOK_URL,跳过飞书通知")
return False
payload = {"msg_type": "interactive", "card": card_payload}
data = json.dumps(payload).encode("utf-8")
req = urllib.request.Request(
self.webhook_url,
data=data,
headers={"Content-Type": "application/json"},
method="POST",
)
try:
with urllib.request.urlopen(req, timeout=10) as resp:
resp_body = resp.read().decode("utf-8")
result = json.loads(resp_body)
if result.get("code", 0) != 0:
print(
f"[WARN] 飞书通知返回错误: {result.get('msg', resp_body)}",
file=sys.stderr,
)
return False
return True
except Exception as e:
print(f"[WARN] 飞书通知发送失败: {e}", file=sys.stderr)
return False
# ── 通知模板 ──────────────────────────────────────
def _run_url(self, run_id):
return f"{config.GITEA_URL}/{config.GITEA_REPO}/actions/runs/{run_id}"
def _pr_url(self, pr_number):
return f"{config.GITEA_URL}/{config.GITEA_REPO}/pulls/{pr_number}"
def notify_branch_failure(self, run_id, branch, run_data=None):
"""main/develop 分支 CI 失败通知
包含: 失败 job 列表、错误摘要、一键重跑链接
"""
run = run_data or self.gitea.get_run(run_id)
if not run:
print(f"[WARN] 无法获取 run {run_id} 详情", file=sys.stderr)
return False
workflow_name = run.get("name", "Unknown Workflow")
commit_msg = run.get("head_commit", {}).get("message", "未知").splitlines()[0][:60]
commit_sha = run.get("head_sha", "")[:8]
actor = (
run.get("trigger_event", {}).get("actor", {}).get("login", "unknown")
if isinstance(run.get("trigger_event"), dict)
else run.get("actor", "unknown")
)
run_url = self._run_url(run_id)
# 获取失败 job 摘要
failed_jobs = self.gitea.get_failed_jobs_summary(run_id, max_lines_per_job=15)
# 构建失败摘要
failure_summary = ""
if failed_jobs:
job_lines = []
for job in failed_jobs[:3]: # 最多显示 3 个
step_info = f"{job['failed_step']}" if job["failed_step"] else ""
job_lines.append(f"• **{job['name']}**{step_info}")
if job["log_tail"]:
# 取最后 3 行日志
tail_lines = job["log_tail"].strip().splitlines()[-3:]
for line in tail_lines:
clean_line = line.strip()[:100]
if clean_line:
job_lines.append(f" `{clean_line}`")
failure_summary = "\n".join(job_lines)
else:
failure_summary = "(获取失败详情中,点击查看日志)"
# 字段
fields = [
{"is_short": True, "text": {"tag": "lark_md", "content": f"**分支**\n{branch}"}},
{"is_short": True, "text": {"tag": "lark_md", "content": f"**Workflow**\n{workflow_name}"}},
{"is_short": True, "text": {"tag": "lark_md", "content": f"**提交**\n`{commit_sha}`"}},
{"is_short": True, "text": {"tag": "lark_md", "content": f"**触发者**\n{actor}"}},
{"is_short": False, "text": {"tag": "lark_md", "content": f"**提交信息**\n{commit_msg}"}},
{"is_short": False, "text": {"tag": "lark_md", "content": f"**失败详情**\n{failure_summary}"}},
]
card = {
"header": {
"title": {
"tag": "plain_text",
"content": f"❌ CI告警:{branch} 分支构建失败",
},
"status": "red",
},
"elements": [
{"tag": "div", "fields": fields},
{
"tag": "action",
"actions": [
{
"tag": "button",
"text": {"tag": "plain_text", "content": "查看失败日志"},
"url": run_url,
"type": "danger",
},
{
"tag": "button",
"text": {"tag": "plain_text", "content": "重跑失败Job"},
"url": f"{run_url}/rerun-failed-jobs",
"type": "default",
},
],
},
],
}
result = self._send_card(card)
print(f"[INFO] 分支失败通知已发送: {branch} run={run_id}")
return result
def notify_branch_recovery(self, run_id, branch, previous_failure_run_id=None):
"""分支 CI 恢复通知(从失败变成功)"""
run = self.gitea.get_run(run_id)
if not run:
print(f"[WARN] 无法获取 run {run_id} 详情", file=sys.stderr)
return False
workflow_name = run.get("name", "Unknown Workflow")
commit_sha = run.get("head_sha", "")[:8]
run_url = self._run_url(run_id)
# 计算恢复耗时
duration_text = "已恢复"
if previous_failure_run_id:
prev_run = self.gitea.get_run(previous_failure_run_id)
if prev_run:
# 简单计算两个 run 的时间差
prev_time = prev_run.get("created_at", "")
cur_time = run.get("created_at", "")
if prev_time and cur_time:
try:
t1 = datetime.fromisoformat(prev_time.replace("Z", "+00:00"))
t2 = datetime.fromisoformat(cur_time.replace("Z", "+00:00"))
diff = (t2 - t1).total_seconds() / 60
duration_text = f"故障时长约 {diff:.0f} 分钟"
except Exception:
pass
fields = [
{"is_short": True, "text": {"tag": "lark_md", "content": f"**分支**\n{branch}"}},
{"is_short": True, "text": {"tag": "lark_md", "content": f"**Workflow**\n{workflow_name}"}},
{"is_short": True, "text": {"tag": "lark_md", "content": f"**提交**\n`{commit_sha}`"}},
{"is_short": True, "text": {"tag": "lark_md", "content": "**状态**\n✅ 已恢复"}},
{"is_short": False, "text": {"tag": "lark_md", "content": f"**说明**\n{duration_text}"}},
]
card = {
"header": {
"title": {
"tag": "plain_text",
"content": f"✅ CI通知:{branch} 分支构建已恢复",
},
"status": "green",
},
"elements": [
{"tag": "div", "fields": fields},
{
"tag": "action",
"actions": [
{
"tag": "button",
"text": {"tag": "plain_text", "content": "查看详情"},
"url": run_url,
"type": "primary",
}
],
},
],
}
result = self._send_card(card)
print(f"[INFO] 分支恢复通知已发送: {branch} run={run_id}")
return result
def notify_e2e_failure(self, run_id, branch="develop", pr_number=None):
"""E2E 测试失败摘要通知"""
failed_jobs = self.gitea.get_failed_jobs_summary(run_id, max_lines_per_job=50)
e2e_jobs = [j for j in failed_jobs if "e2e" in j["name"].lower() or "test" in j["name"].lower()]
if not e2e_jobs:
# 没有明确的 e2e job,取所有失败的
e2e_jobs = failed_jobs
run_url = self._run_url(run_id)
title_suffix = f"PR #{pr_number}" if pr_number else f"{branch} 分支"
# 构建失败用例摘要
case_summary = ""
for job in e2e_jobs[:3]:
case_summary += f"**{job['name']}**\n"
if job["log_tail"]:
# 尝试提取 FAIL 行
fail_lines = [
line.strip()
for line in job["log_tail"].splitlines()
if "FAIL" in line or "fail" in line.lower() or "" in line or "" in line
][:5]
if fail_lines:
for line in fail_lines:
case_summary += f"{line[:120]}\n"
else:
tail = job["log_tail"].strip().splitlines()[-5:]
for line in tail:
case_summary += f" `{line.strip()[:100]}`\n"
case_summary += "\n"
if not case_summary:
case_summary = "(点击查看完整测试报告)"
fields = [
{"is_short": True, "text": {"tag": "lark_md", "content": f"**来源**\n{title_suffix}"}},
{"is_short": True, "text": {"tag": "lark_md", "content": f"**失败Job数**\n{len(e2e_jobs)}"}},
{"is_short": False, "text": {"tag": "lark_md", "content": f"**失败摘要**\n{case_summary}"}},
]
card = {
"header": {
"title": {
"tag": "plain_text",
"content": f"🧪 CI告警:E2E 测试失败 - {title_suffix}",
},
"status": "orange",
},
"elements": [
{"tag": "div", "fields": fields},
{
"tag": "action",
"actions": [
{
"tag": "button",
"text": {"tag": "plain_text", "content": "查看完整报告"},
"url": run_url,
"type": "danger",
}
],
},
],
}
result = self._send_card(card)
print(f"[INFO] E2E失败通知已发送: run={run_id}")
return result
def notify_pr_failure(self, run_id, pr_number, pr_title=""):
"""PR CI 失败通知(轻量版,可选开启)"""
run_url = self._run_url(run_id)
pr_url = self._pr_url(pr_number)
failed_jobs = self.gitea.get_failed_jobs_summary(run_id, max_lines_per_job=10)
failure_names = [j["name"] for j in failed_jobs[:3]]
failure_text = "".join(failure_names) if failure_names else "未知"
fields = [
{
"is_short": False,
"text": {"tag": "lark_md", "content": f"**PR**\n[#{pr_number} {pr_title[:50]}]({pr_url})"},
},
{"is_short": False, "text": {"tag": "lark_md", "content": f"**失败任务**\n{failure_text}"}},
]
card = {
"header": {
"title": {
"tag": "plain_text",
"content": f"⚠️ CI通知:PR #{pr_number} 构建失败",
},
"status": "yellow",
},
"elements": [
{"tag": "div", "fields": fields},
{
"tag": "action",
"actions": [
{
"tag": "button",
"text": {"tag": "plain_text", "content": "查看日志"},
"url": run_url,
"type": "default",
},
{
"tag": "button",
"text": {"tag": "plain_text", "content": "查看PR"},
"url": pr_url,
"type": "default",
},
],
},
],
}
result = self._send_card(card)
print(f"[INFO] PR失败通知已发送: PR #{pr_number} run={run_id}")
return result
# ── CLI 入口 ──────────────────────────────────────────
def main():
parser = argparse.ArgumentParser(description="飞书 CI 通知")
parser.add_argument(
"--mode",
required=True,
choices=["failure", "recovery", "e2e_failure", "pr_failure"],
help="通知模式",
)
parser.add_argument("--run-id", required=True, help="Workflow Run ID")
parser.add_argument("--branch", default="develop", help="分支名")
parser.add_argument("--pr-number", type=int, help="PR 编号")
parser.add_argument("--pr-title", default="", help="PR 标题")
parser.add_argument("--prev-run-id", help="上一个失败的 run ID(恢复通知用)")
parser.add_argument("--webhook", help="飞书 webhook URL(覆盖环境变量)")
args = parser.parse_args()
notifier = FeishuNotifier(webhook_url=args.webhook)
if args.mode == "failure":
notifier.notify_branch_failure(args.run_id, args.branch)
elif args.mode == "recovery":
notifier.notify_branch_recovery(args.run_id, args.branch, previous_failure_run_id=args.prev_run_id)
elif args.mode == "e2e_failure":
notifier.notify_e2e_failure(args.run_id, branch=args.branch, pr_number=args.pr_number)
elif args.mode == "pr_failure":
notifier.notify_pr_failure(args.run_id, args.pr_number, pr_title=args.pr_title)
return 0
if __name__ == "__main__":
sys.exit(main())
+243
View File
@@ -0,0 +1,243 @@
#!/usr/bin/env python3
"""
Gitea API 客户端封装 - Actions + PR + Webhook 相关接口
基于 urllib 实现,无第三方依赖,与 ci_dashboard.py 风格一致。
支持 token 和 basic auth 两种认证方式。
"""
import base64
import json
import sys
import urllib.error
import urllib.request
from . import config
class GiteaClient:
"""Gitea API 客户端"""
def __init__(
self,
base_url=None,
repo=None,
token=None,
username=None,
password=None,
):
self.base_url = (base_url or config.GITEA_URL).rstrip("/")
self.repo = repo or config.GITEA_REPO
self.token = token or config.GITEA_TOKEN
self.username = username or config.GITEA_USERNAME
self.password = password or config.GITEA_PASSWORD
self.api_base = f"{self.base_url}/api/v1/repos/{self.repo}"
def _request(self, path, method="GET", data=None):
"""通用 HTTP 请求
Args:
path: API 路径(相对于 /api/v1/repos/{repo}/
method: HTTP 方法
data: 请求体(dict 或 bytes
Returns:
解析后的 JSON 数据,失败返回 None
"""
url = f"{self.api_base}/{path}"
body = None
if data is not None:
if isinstance(data, (dict, list)):
body = json.dumps(data).encode("utf-8")
else:
body = data if isinstance(data, bytes) else str(data).encode()
req = urllib.request.Request(url, data=body, method=method)
req.add_header("Content-Type", "application/json")
if self.token:
req.add_header("Authorization", f"token {self.token}")
elif self.username and self.password:
auth = base64.b64encode(f"{self.username}:{self.password}".encode()).decode()
req.add_header("Authorization", f"Basic {auth}")
try:
with urllib.request.urlopen(req, timeout=30) as resp:
resp_body = resp.read().decode()
if not resp_body:
return {}
return json.loads(resp_body)
except urllib.error.HTTPError as e:
err_body = ""
try:
err_body = e.read().decode()
except Exception:
pass
print(
f"[WARN] HTTP {e.code}: {url} - {err_body[:200]}",
file=sys.stderr,
)
return None
except Exception as e:
print(f"[WARN] 请求失败 {url}: {e}", file=sys.stderr)
return None
# ── Actions: Workflow Runs ────────────────────────
def list_runs(
self,
status=None,
branch=None,
event=None,
workflow_id=None,
page=1,
limit=config.PAGE_LIMIT,
):
"""获取 workflow runs 列表
Returns:
(runs列表, 总数)
"""
params = []
if status:
params.append(f"status={status}")
if branch:
params.append(f"branch={branch}")
if event:
params.append(f"event={event}")
if workflow_id:
params.append(f"workflow_id={workflow_id}")
params.append(f"page={page}")
params.append(f"limit={limit}")
path = f"actions/runs?{'&'.join(params)}"
data = self._request(path)
if not data:
return [], 0
runs = data.get("workflow_runs", [])
total = data.get("total_count", 0)
return runs, total
def get_run(self, run_id):
"""获取单个 run 详情"""
return self._request(f"actions/runs/{run_id}")
def get_run_jobs(self, run_id):
"""获取 run 的 jobs 列表"""
data = self._request(f"actions/runs/{run_id}/jobs")
if not data:
return []
return data.get("jobs", [])
def get_job_log(self, run_id, job_id):
"""获取 job 日志(纯文本)"""
url = f"{self.api_base}/actions/runs/{run_id}/jobs/{job_id}/logs"
req = urllib.request.Request(url)
if self.token:
req.add_header("Authorization", f"token {self.token}")
elif self.username and self.password:
auth = base64.b64encode(f"{self.username}:{self.password}".encode()).decode()
req.add_header("Authorization", f"Basic {auth}")
try:
with urllib.request.urlopen(req, timeout=30) as resp:
return resp.read().decode("utf-8", errors="replace")
except Exception as e:
print(f"[WARN] 获取日志失败 job={job_id}: {e}", file=sys.stderr)
return ""
def rerun_run(self, run_id):
"""重新运行整个 workflow run"""
return self._request(f"actions/runs/{run_id}/rerun", method="POST")
def rerun_failed_jobs(self, run_id):
"""重新运行失败的 jobs"""
return self._request(f"actions/runs/{run_id}/rerun-failed-jobs", method="POST")
def cancel_run(self, run_id):
"""取消 run"""
return self._request(f"actions/runs/{run_id}/cancel", method="POST")
# ── Actions: Workflows ────────────────────────────
def list_workflows(self):
"""获取 workflow 列表"""
data = self._request("actions/workflows")
if not data:
return []
return data.get("workflows", [])
def get_workflow(self, workflow_id):
"""获取单个 workflow 详情"""
return self._request(f"actions/workflows/{workflow_id}")
# ── Pull Requests ─────────────────────────────────
def get_pr(self, pr_number):
"""获取 PR 详情"""
return self._request(f"pulls/{pr_number}")
def get_pr_ci_runs(self, pr_number, limit=20):
"""获取 PR 关联的 CI runs(通过 head_sha 查询)"""
pr = self.get_pr(pr_number)
if not pr:
return []
head_sha = pr.get("head", {}).get("sha", "")
if not head_sha:
return []
# 用 head_sha 过滤 runs
runs, _ = self.list_runs(limit=limit)
return [r for r in runs if r.get("head_sha", "") == head_sha]
# ── 便捷方法 ──────────────────────────────────────
def get_latest_run(self, branch, workflow_id=None, status=None):
"""获取指定分支最新的 run"""
runs, _ = self.list_runs(branch=branch, workflow_id=workflow_id, status=status, limit=5)
return runs[0] if runs else None
def get_failed_jobs_summary(self, run_id, max_lines_per_job=30):
"""获取失败 job 的摘要信息(用于通知)
Returns:
list[dict]: 每个失败 job 的 {name, conclusion, failed_step, log_tail}
"""
jobs = self.get_run_jobs(run_id)
if not jobs:
return []
failed = [j for j in jobs if j.get("status") == "completed" and j.get("conclusion") == "failure"]
if not failed:
# 运行中的也返回,方便定位
failed = [j for j in jobs if j.get("status") != "completed"]
result = []
for job in failed[:5]: # 最多取 5 个失败 job
job_id = job.get("id", "")
name = job.get("name", "Unknown")
conclusion = job.get("conclusion", job.get("status", "unknown"))
# 找失败的 step
failed_step = ""
steps = job.get("steps", [])
for step in steps:
if step.get("conclusion") == "failure":
failed_step = step.get("name", "")
break
# 取日志尾部
log_tail = ""
if job_id:
log = self.get_job_log(run_id, job_id)
if log:
lines = log.strip().splitlines()
log_tail = "\n".join(lines[-max_lines_per_job:])
result.append(
{
"name": name,
"conclusion": conclusion,
"failed_step": failed_step,
"log_tail": log_tail,
"job_id": job_id,
}
)
return result
+446
View File
@@ -0,0 +1,446 @@
#!/usr/bin/env python3
"""
Gitea Webhook 接收服务 - FastAPI 实现
功能:
1. 接收 Gitea Actions webhook 事件,触发飞书通知
2. 接收飞书机器人回调消息,处理 /ci 交互命令
3. 维护简单的状态缓存,检测分支恢复等状态变化
部署:
部署到构建服务器,监听 8090 端口(可配置)
Gitea webhook 指向: http://<server>:8090/webhook/gitea
飞书消息回调指向: http://<server>:8090/webhook/feishu
依赖:
fastapi + uvicorn(可选,未安装时仅模块可用,服务不可启动)
注意:
本文件为第一版骨架,通知逻辑已实现,飞书交互命令待后续完善。
"""
import hashlib
import hmac
import json
import sys
import threading
import time
from typing import Optional
from . import config
from .gitea_client import GiteaClient
# FastAPI 是可选依赖,未安装时仅导出类不启动服务
try:
from fastapi import FastAPI, Header, HTTPException, Request
from fastapi.responses import JSONResponse
FASTAPI_AVAILABLE = True
except ImportError:
FASTAPI_AVAILABLE = False
FastAPI = None # type: ignore
# ── 状态缓存 ──────────────────────────────────────────
class StateCache:
"""简单的内存状态缓存,用于检测状态变化
记录每个分支最后一次 run 的状态,用于判断:
- 是否从失败变成功(恢复通知)
- 是否连续失败(避免重复告警)
"""
def __init__(self, max_entries=100):
self._cache = {} # {branch: {last_status, last_run_id, last_notified_failure}}
self._lock = threading.Lock()
self._max = max_entries
def get(self, key):
with self._lock:
return self._cache.get(key)
def set(self, key, value):
with self._lock:
self._cache[key] = value
# 简单的淘汰策略
if len(self._cache) > self._max:
oldest_key = next(iter(self._cache))
del self._cache[oldest_key]
def check_and_update(self, branch, run_id, conclusion):
"""检查状态变化并更新缓存
Returns:
dict: {is_new_failure, is_recovery, previous_status, previous_run_id}
"""
prev = self.get(branch) or {}
prev_status = prev.get("last_status", "unknown")
prev_run_id = prev.get("last_run_id")
is_new_failure = False
is_recovery = False
if conclusion == "failure" and prev_status != "failure":
is_new_failure = True
if conclusion == "success" and prev_status == "failure":
is_recovery = True
self.set(
branch,
{
"last_status": conclusion,
"last_run_id": run_id,
"last_updated": time.time(),
"last_notified_failure": run_id if is_new_failure else prev.get("last_notified_failure"),
},
)
return {
"is_new_failure": is_new_failure,
"is_recovery": is_recovery,
"previous_status": prev_status,
"previous_run_id": prev_run_id,
}
state_cache = StateCache()
# ── Gitea Webhook 处理 ────────────────────────────────
def verify_gitea_signature(payload: bytes, signature: str) -> bool:
"""校验 Gitea webhook 签名(X-Gitea-Signature
Gitea 使用 HMAC-SHA256 签名,格式: sha256=xxx
"""
if not config.WEBHOOK_SECRET:
return True # 未配置密钥则跳过校验
if not signature:
return False
try:
algo, sig_hex = signature.split("=", 1)
if algo != "sha256":
return False
expected = hmac.new(config.WEBHOOK_SECRET.encode(), payload, hashlib.sha256).hexdigest()
return hmac.compare_digest(expected, sig_hex)
except Exception:
return False
def handle_gitea_webhook(payload: dict, event_type: str) -> dict:
"""处理 Gitea webhook 事件
Args:
payload: webhook 请求体
event_type: X-Gitea-Event 头
Returns:
dict: {handled, notifications_sent, message}
"""
if event_type != "create" and event_type != "push":
# 我们主要关心 push 和 actions 事件
# Gitea Actions 的 webhook 事件类型可能是 "push" 或专门的 actions 事件
pass
# 尝试提取 run 信息
run_info = _extract_run_info(payload)
if not run_info:
return {"handled": False, "notifications_sent": 0, "message": "非 CI 事件,跳过"}
branch = run_info["branch"]
run_id = run_info["run_id"]
status = run_info["status"]
conclusion = run_info.get("conclusion", "")
# 只处理已完成的 run
if status != "completed":
return {"handled": True, "notifications_sent": 0, "message": f"Run {run_id} 仍在运行中 ({status})"}
# 检查是否在通知分支列表中
if branch not in config.NOTIFY_BRANCHES:
return {
"handled": True,
"notifications_sent": 0,
"message": f"分支 {branch} 不在通知列表中",
}
# 检查状态变化
change_info = state_cache.check_and_update(branch, run_id, conclusion)
notifications = 0
# 延迟导入,避免循环依赖
from .feishu_notify import FeishuNotifier
notifier = FeishuNotifier()
if conclusion == "failure" and change_info["is_new_failure"]:
# 新失败 → 发失败通知
notifier.notify_branch_failure(run_id, branch)
notifications += 1
# 检查是否是 E2E 失败
gitea = GiteaClient()
failed_jobs = gitea.get_failed_jobs_summary(run_id)
has_e2e = any("e2e" in j["name"].lower() for j in failed_jobs)
if has_e2e:
notifier.notify_e2e_failure(run_id, branch=branch)
notifications += 1
elif conclusion == "success" and change_info["is_recovery"]:
# 从失败恢复 → 发恢复通知
prev_run_id = change_info.get("previous_run_id")
notifier.notify_branch_recovery(run_id, branch, previous_failure_run_id=prev_run_id)
notifications += 1
return {
"handled": True,
"notifications_sent": notifications,
"message": f"分支 {branch} run {run_id} {conclusion}",
}
def _extract_run_info(payload: dict) -> Optional[dict]:
"""从 webhook payload 中提取 run 信息
Gitea Actions webhook 的 payload 结构可能不同,这里做兼容处理。
如果 payload 不是 run 事件,返回 None。
"""
# 尝试多种可能的结构
if "workflow_run" in payload:
wr = payload["workflow_run"]
return {
"run_id": wr.get("id"),
"branch": wr.get("head_branch", ""),
"status": wr.get("status", ""),
"conclusion": wr.get("conclusion", ""),
"name": wr.get("name", ""),
}
if "action" in payload and "pull_request" in payload:
# PR 事件,暂不处理
return None
if "ref" in payload and "head_commit" in payload:
# push 事件,不是 run 事件
return None
return None
# ── 飞书消息处理 ──────────────────────────────────────
def handle_feishu_message(payload: dict) -> dict:
"""处理飞书机器人回调消息
支持命令:
/ci status [branch] - 查询分支 CI 状态
/ci rerun <run-id> - 重跑失败的 jobs
/ci help - 帮助
注意: 第一版骨架,仅解析命令,实际执行逻辑待完善。
"""
# 飞书消息回调格式
header = payload.get("header", {})
event_type = header.get("event_type", "")
if event_type == "url_verification":
# 飞书 URL 验证
return {"challenge": payload.get("challenge", "")}
if event_type != "im.message.receive_v1":
return {"handled": False, "message": f"非消息事件: {event_type}"}
event = payload.get("event", {})
message = event.get("message", {})
content_str = message.get("content", "{}")
try:
content = json.loads(content_str)
except json.JSONDecodeError:
content = {}
text = content.get("text", "")
if not text:
return {"handled": False, "message": "空消息"}
# 解析命令
text = text.strip()
if not text.startswith("/ci"):
return {"handled": False, "message": "非 CI 命令"}
parts = text.split()
if len(parts) < 2:
return _help_response()
cmd = parts[1].lower()
if cmd == "status":
branch = parts[2] if len(parts) > 2 else "develop"
return _handle_status_command(branch)
elif cmd == "rerun":
if len(parts) < 3:
return {"text": "用法: /ci rerun <run-id> 或 /ci rerun latest [branch]"}
arg = parts[2]
if arg == "latest":
branch = parts[3] if len(parts) > 3 else "develop"
return _handle_rerun_latest(branch)
return _handle_rerun_command(arg)
elif cmd == "help":
return _help_response()
else:
return {"text": f"未知命令: {cmd}\n输入 /ci help 查看帮助"}
def _handle_status_command(branch: str) -> dict:
"""处理 /ci status 命令"""
from .ci_query import CIQuery
query = CIQuery()
result = query.get_branch_status(branch)
reply = CIQuery.format_branch_status(result)
return {"text": reply}
def _handle_rerun_command(run_id: str) -> dict:
"""处理 /ci rerun 命令"""
from .ci_trigger import CITrigger
trigger = CITrigger()
try:
result = trigger.rerun_failed(int(run_id))
except ValueError:
return {"text": f"无效的 run id: {run_id}"}
if result["success"]:
return {"text": f"{result['message']}\n{result.get('run_url', '')}"}
return {"text": f"{result['message']}"}
def _handle_rerun_latest(branch: str) -> dict:
"""处理 /ci rerun latest 命令"""
from .ci_trigger import CITrigger
trigger = CITrigger()
result = trigger.rerun_latest_failed(branch=branch)
if result["success"]:
return {"text": f"{result['message']}\n{result.get('run_url', '')}"}
return {"text": f"{result['message']}"}
def _help_response() -> dict:
"""返回帮助信息"""
help_text = """**CI ChatOps 命令帮助**
`/ci status [branch]` 查询分支 CI 状态(默认 develop)
`/ci rerun <run-id>` 重跑指定 run 的失败 jobs
`/ci rerun latest [branch]` 重跑分支最近一次失败的 run
`/ci help` 显示此帮助
**环境变量配置:**
`GITEA_TOKEN` / `GITEA_USERNAME + GITEA_PASSWORD`
`FEISHU_WEBHOOK_URL`
`CHATOPS_NOTIFY_BRANCHES=main,develop`
"""
return {"text": help_text}
# ── FastAPI 应用 ──────────────────────────────────────
def create_app():
"""创建 FastAPI 应用
如果 FastAPI 未安装,返回 None
"""
if not FASTAPI_AVAILABLE:
print(
"[WARN] FastAPI 未安装,无法启动 webhook 服务。" " 请运行: pip install fastapi uvicorn",
file=sys.stderr,
)
return None
app = FastAPI(title="CI ChatOps Webhook", version="0.1.0")
@app.post("/webhook/gitea")
async def gitea_webhook(
request: Request,
x_gitea_event: str = Header(default=""),
x_gitea_signature: str = Header(default=""),
):
body = await request.body()
# 签名校验
if not verify_gitea_signature(body, x_gitea_signature):
raise HTTPException(status_code=401, detail="Invalid signature")
try:
payload = json.loads(body.decode())
except json.JSONDecodeError as e:
raise HTTPException(status_code=400, detail="Invalid JSON") from e
result = handle_gitea_webhook(payload, x_gitea_event)
return JSONResponse(content=result)
@app.post("/webhook/feishu")
async def feishu_webhook(request: Request):
body = await request.body()
try:
payload = json.loads(body.decode())
except json.JSONDecodeError as e:
raise HTTPException(status_code=400, detail="Invalid JSON") from e
result = handle_feishu_message(payload)
return JSONResponse(content=result)
@app.get("/health")
async def health():
return {"status": "ok", "service": "ci-chatops"}
return app
# ── CLI 入口 ──────────────────────────────────────────
def main():
"""启动 webhook 服务"""
import argparse
parser = argparse.ArgumentParser(description="CI ChatOps Webhook 服务")
parser.add_argument("--port", type=int, default=config.WEBHOOK_PORT, help="监听端口")
parser.add_argument("--host", default="0.0.0.0", help="监听地址")
args = parser.parse_args()
app = create_app()
if not app:
print("[ERROR] FastAPI 不可用,请先安装: pip install fastapi uvicorn")
return 1
try:
import uvicorn
except ImportError:
print("[ERROR] uvicorn 未安装,请先安装: pip install uvicorn")
return 1
print(f"[INFO] CI ChatOps Webhook 服务启动: http://{args.host}:{args.port}")
print("[INFO] Gitea webhook: POST /webhook/gitea")
print("[INFO] 飞书 webhook: POST /webhook/feishu")
print("[INFO] 健康检查: GET /health")
print(f"[INFO] 通知分支: {', '.join(config.NOTIFY_BRANCHES)}")
uvicorn.run(app, host=args.host, port=args.port)
return 0
if __name__ == "__main__":
sys.exit(main())