ci: direct runner label fix - foreground execution
Runner Diagnose / diagnose (push) Successful in 0s

This commit is contained in:
灵应
2026-07-12 01:50:17 +08:00
parent 809b2d6f71
commit e697cc5769
+101 -31
View File
@@ -1,44 +1,114 @@
name: Runner Fix & Scale
name: Runner Fix - Unify Labels
on:
push:
branches:
- ci/runner-fix
jobs:
fix-all-runners:
runs-on: host
fix-runners:
runs-on: saas
steps:
- name: 一次性修复所有runner标签
- name: 重新注册主runner和runner-3
run: |
set -e
CI_DIR="/var/lib/xiaoxia-ci"
ACT_BIN="$CI_DIR/act_runner"
GITEA_URL="https://git.xiaoxiajianji.com"
TOKEN="token123" # 会在下面替换
echo "========== 1. 查当前3个runner的配置 =========="
for d in "$CI_DIR" "$CI_DIR/runner-2" "$CI_DIR/runner-3"; do
echo "--- $d ---"
if [ -f "$d/.runner" ]; then
cat "$d/.runner" | python3 -c "import sys,json; d=json.load(sys.stdin); print(f'name={d.get(\"name\",\"?\")} labels={d.get(\"labels\",[])} uuid={d.get(\"uuid\",\"?\")[:8]}')" 2>/dev/null || cat "$d/.runner"
elif [ -f "$d/.runner-config.yaml" ]; then
head -30 "$d/.runner-config.yaml"
else
echo "no config found"
ls -la "$d/" | head -20
fi
echo ""
done
echo "========== 2. 查当前systemd服务 =========="
for s in act_runner.service act_runner-2.service act_runner-3.service; do
echo "--- $s ---"
systemctl cat "$s" 2>/dev/null | head -20
echo ""
done
echo "========== 3. 查Gitea API的runner列表 =========="
LABELS="saas,runtime-builder,host,ubuntu-latest"
echo "=== 获取 registration token ==="
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 [ -z "$REG_TOKEN" ]; then
echo "ERROR: 无法获取 registration token"
exit 1
fi
echo "Token 获取成功: ${REG_TOKEN:0:8}..."
# 修复主runner(它有runtime-builder标签,当前没在跑任务)
echo ""
echo "=== 修复主runner ==="
systemctl stop act_runner.service
sleep 2
cd "$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,启动服务..."
systemctl start act_runner.service
echo "主runner已启动"
# 修复runner-3(它有saas标签,可能没在跑任务)
echo ""
echo "=== 修复 runner-3 ==="
systemctl stop act_runner-3.service
sleep 2
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
systemctl daemon-reload
echo "runner-2已设置启动时自动重新注册"
# 等runner连上
sleep 20
# 验证
echo ""
echo "=== 验证结果 ==="
curl -sk -H "Authorization: token 1f8058d097e3942a9ed31c44382baf7f08311272" \
"$GITEA_URL/api/v1/repos/xiaoxia/xiaoxia-saas/actions/runners?limit=20" 2>/dev/null | python3 -m json.tool 2>/dev/null || echo "API call failed"
echo "========== 完成 =========="
"$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')
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"