ci: final runner label fix - correct config format
Runner Diagnose / diag (push) Successful in 0s
Runner Diagnose / diagnose (push) Successful in 0s

This commit is contained in:
灵应
2026-07-12 02:01:07 +08:00
parent 8978819fcc
commit 35ce5fe26d
+61 -70
View File
@@ -1,14 +1,14 @@
name: Runner Fix - Unify Labels
name: Runner Fix - Final
on:
push:
branches:
- ci/runner-fix
jobs:
fix-runners:
runs-on: saas
fix-all:
runs-on: runtime-builder
steps:
- name: 重新注册主runner和runner-3
- name: 统一所有runner标签
run: |
set -e
@@ -16,33 +16,31 @@ jobs:
ACT_BIN="$CI_DIR/act_runner"
GITEA_URL="https://git.xiaoxiajianji.com"
LABELS="saas,runtime-builder,host,ubuntu-latest"
TOKEN="1f8058d097e3942a9ed31c44382baf7f08311272"
echo "=== 获取 registration token ==="
REG_TOKEN=$(curl -sk -X POST \
-H "Authorization: token 1f8058d097e3942a9ed31c44382baf7f08311272" \
-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: ${REG_TOKEN:0:10}..."
if [ -z "$REG_TOKEN" ]; then
echo "ERROR: 无法获取 registration token"
exit 1
fi
echo "Token 获取成功: ${REG_TOKEN:0:8}..."
# 修复主runner(它有runtime-builder标签,当前没在跑任务)
# 修复 runner-2
echo ""
echo "=== 修复runner ==="
systemctl stop act_runner.service
echo "=== 修复 runner-2 ==="
systemctl stop act_runner-2.service
sleep 2
cd "$CI_DIR"
cd "$CI_DIR/runner-2"
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,启动服务..."
systemctl start act_runner.service
echo "主runner已启动"
--name "xiaoxia-ci-runner-2" --labels "$LABELS" --no-interactive 2>&1
# 确保systemd服务用 .runner 文件
sed -i 's|daemon.*|daemon -c /var/lib/xiaoxia-ci/runner-2/.runner|' /etc/systemd/system/act_runner-2.service
systemctl daemon-reload
systemctl start act_runner-2.service
echo "runner-2 done"
# 修复runner-3(它有saas标签,可能没在跑任务)
# 修复 runner-3
echo ""
echo "=== 修复 runner-3 ==="
systemctl stop act_runner-3.service
@@ -50,65 +48,58 @@ jobs:
cd "$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
echo "注册完成,启动服务..."
systemctl start act_runner-3.service
echo "runner-3已启动"
# runner-2正在跑这个job,最后处理:用systemd drop-in让它下次启动时自动重新注册
echo ""
echo "=== 设置 runner-2 自动重新注册 ==="
cat > /tmp/reregister_runner2.sh << 'INNEREOF'
#!/bin/bash
set -e
CI_DIR="/var/lib/xiaoxia-ci"
ACT_BIN="$CI_DIR/act_runner"
GITEA_URL="https://git.xiaoxiajianji.com"
LABELS="saas,runtime-builder,host,ubuntu-latest"
REG_TOKEN=$(curl -sk -X POST \
-H "Authorization: token 1f8058d097e3942a9ed31c44382baf7f08311272" \
"$GITEA_URL/api/v1/repos/xiaoxia/xiaoxia-saas/actions/runners/registration-token" \
| python3 -c "import sys,json; print(json.load(sys.stdin).get('token',''))")
if [ -n "$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 successfully"
fi
INNEREOF
chmod +x /tmp/reregister_runner2.sh
mkdir -p /etc/systemd/system/act_runner-2.service.d/
cat > /etc/systemd/system/act_runner-2.service.d/override.conf << 'OVERRIDE'
[Service]
ExecStartPre=/tmp/reregister_runner2.sh
OVERRIDE
--name "xiaoxia-ci-runner-3" --labels "$LABELS" --no-interactive 2>&1
sed -i 's|daemon.*|daemon -c /var/lib/xiaoxia-ci/runner-3/.runner|' /etc/systemd/system/act_runner-3.service
systemctl daemon-reload
echo "runner-2已设置启动时自动重新注册"
systemctl start act_runner-3.service
echo "runner-3 done"
# 等runner连上
sleep 20
# 验证
# 修复主runner(当前正在跑这个job,最后处理:先改配置再用nohup重启自己)
echo ""
echo "=== 验证结果 ==="
curl -sk -H "Authorization: token 1f8058d097e3942a9ed31c44382baf7f08311272" \
echo "=== 修复主runner(当前runner ==="
cd "$CI_DIR"
# 先注册生成 .runner 文件
rm -f .runner
$ACT_BIN register --instance "$GITEA_URL" --token "$REG_TOKEN" \
--name "xiaoxia-ci-runner" --labels "$LABELS" --no-interactive 2>&1
# 修改systemd服务:从 .runner-config.yaml 改成 .runner
cat > /etc/systemd/system/act_runner.service << 'SVC'
[Unit]
Description=Gitea Actions Runner
After=network.target
[Service]
WorkingDirectory=/var/lib/xiaoxia-ci
ExecStart=/var/lib/xiaoxia-ci/act_runner daemon -c /var/lib/xiaoxia-ci/.runner
Restart=always
RestartSec=5
Type=simple
[Install]
WantedBy=multi-user.target
SVC
systemctl daemon-reload
echo "主runner配置已更新,准备重启(当前job会被断开)..."
# 等runner-2和runner-3先连上,再重启自己
sleep 25
# 后台重启自己
nohup bash -c "sleep 3 && systemctl restart act_runner.service" > /dev/null 2>&1 &
echo "重启指令已发出"
# 验证当前状态
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)} 个')
print(f'当前在线: {len(online)} 个')
for r in online:
labels = [l['name'] for l in r['labels']]
print(f' {r[\"name\"]}: {labels}')
"
# 最后:用nohup重启runner-2(这会断开当前job,但标签已经在另外两个上生效了)
echo ""
echo "重启 runner-2 以应用新标签(当前job会被断开)..."
nohup bash -c "sleep 5 && systemctl restart act_runner-2.service" > /dev/null 2>&1 &
echo "重启指令已发出"