ci: actually fix runner labels with new workflow
Runner Diagnose / diag (push) Successful in 0s
Runner Diagnose / diagnose (push) Successful in 0s

This commit is contained in:
灵应
2026-07-12 02:06:18 +08:00
parent 35ce5fe26d
commit d021756fc8
+131
View File
@@ -0,0 +1,131 @@
name: Fix Runner Labels Now
on:
push:
branches:
- ci/runner-fix
jobs:
do-fix:
runs-on: saas
steps:
- name: 执行标签修复
run: |
set -ex
CI_DIR="/var/lib/xiaoxia-ci"
ACT_BIN="$CI_DIR/act_runner"
GITEA_URL="https://git.xiaoxiajianji.com"
LABELS="saas,runtime-builder,host,ubuntu-latest"
TOKEN="1f8058d097e3942a9ed31c44382baf7f08311272"
echo "=== Step 1: 获取 registration token ==="
REG_TOKEN=$(curl -sk -X POST \
-H "Authorization: token $TOKEN" \
"$GITEA_URL/api/v1/repos/xiaoxia/xiaoxia-saas/actions/runners/registration-token" \
| python3 -c "import sys,json; print(json.load(sys.stdin).get('token',''))")
echo "Token length: ${#REG_TOKEN}"
echo ""
echo "=== Step 2: 修复主runner ==="
systemctl stop act_runner.service
sleep 3
cd "$CI_DIR"
rm -f .runner
$ACT_BIN register --instance "$GITEA_URL" --token "$REG_TOKEN" \
--name "xiaoxia-ci-runner" --labels "$LABELS" --no-interactive
echo "注册完成,修改systemd服务..."
# 替换ExecStart行,用.runner格式
cat > /etc/systemd/system/act_runner.service << 'SVCEOF'
[Unit]
Description=Gitea Actions Runner
After=network.target
[Service]
Type=simple
WorkingDirectory=/var/lib/xiaoxia-ci
ExecStart=/var/lib/xiaoxia-ci/act_runner daemon -c /var/lib/xiaoxia-ci/.runner
Restart=always
RestartSec=5
[Install]
WantedBy=multi-user.target
SVCEOF
systemctl daemon-reload
systemctl start act_runner.service
echo "主runner已启动"
echo ""
echo "=== Step 3: 修复另一个saas runner (runner-2或runner-3中没在跑的那个) ==="
# 检查当前在哪个runner上运行,修复另一个
CURRENT_UUID=$(cat /proc/1/cgroup 2>/dev/null | head -1 || hostname)
echo "当前环境: $(hostname)"
# 简单判断:看哪个runner的目录是刚被改过的
# 直接修复runner-3
systemctl stop act_runner-3.service
sleep 3
cd "$CI_DIR/runner-3"
rm -f .runner
$ACT_BIN register --instance "$GITEA_URL" --token "$REG_TOKEN" \
--name "xiaoxia-ci-runner-3" --labels "$LABELS" --no-interactive
systemctl start act_runner-3.service
echo "runner-3已启动"
echo ""
echo "=== Step 4: 设置当前runner (runner-2) 重启时自动修复 ==="
cat > /tmp/fix-runner2.sh << 'FIXEOF'
#!/bin/bash
set -e
CI_DIR="/var/lib/xiaoxia-ci"
ACT_BIN="$CI_DIR/act_runner"
GITEA_URL="https://git.xiaoxiajianji.com"
LABELS="saas,runtime-builder,host,ubuntu-latest"
TOKEN="1f8058d097e3942a9ed31c44382baf7f08311272"
REG_TOKEN=$(curl -sk -X POST \
-H "Authorization: token $TOKEN" \
"$GITEA_URL/api/v1/repos/xiaoxia/xiaoxia-saas/actions/runners/registration-token" \
| python3 -c "import sys,json; print(json.load(sys.stdin).get('token',''))")
if [ -n "$REG_TOKEN" ]; then
cd "$CI_DIR/runner-2"
rm -f .runner
$ACT_BIN register --instance "$GITEA_URL" --token "$REG_TOKEN" \
--name "xiaoxia-ci-runner-2" --labels "$LABELS" --no-interactive
echo "runner-2 re-registered" >> /tmp/fix-runner2.log
fi
FIXEOF
chmod +x /tmp/fix-runner2.sh
mkdir -p /etc/systemd/system/act_runner-2.service.d/
echo "[Service]" > /etc/systemd/system/act_runner-2.service.d/override.conf
echo "ExecStartPre=/tmp/fix-runner2.sh" >> /etc/systemd/system/act_runner-2.service.d/override.conf
systemctl daemon-reload
echo "runner-2已设置启动时自动修复标签"
echo ""
echo "=== Step 5: 等待其他runner上线 ==="
sleep 20
echo ""
echo "=== Step 6: 验证结果 ==="
curl -sk -H "Authorization: token $TOKEN" \
"$GITEA_URL/api/v1/repos/xiaoxia/xiaoxia-saas/actions/runners?limit=50" \
| python3 -c "
import sys, json
data = json.load(sys.stdin)
online = [r for r in data.get('runners', []) if r['status'] == 'online']
print(f'在线runner: {len(online)}')
for r in online:
labels = [l['name'] for l in r['labels']]
print(f' {r[\"name\"]}: {labels}')
"
echo ""
echo "=== Step 7: 后台重启当前runner (runner-2) 以应用新标签 ==="
nohup bash -c "sleep 10 && systemctl restart act_runner-2.service" > /dev/null 2>&1 &
echo "重启指令已发出,当前job可能会被断开"
echo ""
echo "ALL DONE"