ci: unify all runner labels via saas runner
Runner Diagnose / diagnose (push) Successful in 0s
Runner Fix & Scale / fix-all-runners (push) Successful in 0s

This commit is contained in:
灵应
2026-07-12 01:46:22 +08:00
parent 1ceeb191ea
commit 809b2d6f71
+113
View File
@@ -0,0 +1,113 @@
name: Unify All Runner Labels
on:
push:
branches:
- ci/runner-fix
jobs:
unify-labels:
runs-on: saas
steps:
- name: 后台执行所有runner标签统一
run: |
cat > /tmp/unify_labels.sh << 'EOF'
#!/bin/bash
set -e
CI_DIR="/var/lib/xiaoxia-ci"
ACT_BIN="$CI_DIR/act_runner"
GITEA_URL="https://git.xiaoxiajianji.com"
LABELS="saas,runtime-builder,host,ubuntu-latest"
GITEA_TOKEN="1f8058d097e3942a9ed31c44382baf7f08311272"
# 获取registration token
REG_TOKEN=$(curl -sk -X POST \
-H "Authorization: token $GITEA_TOKEN" \
"$GITEA_URL/api/v1/repos/xiaoxia/xiaoxia-saas/actions/runners/registration-token" \
| python3 -c "import sys,json; print(json.load(sys.stdin)['token'])")
echo "Registration token obtained" >> /tmp/unify_progress.log
# 修复主runner
echo "=== Fixing main runner ===" >> /tmp/unify_progress.log
systemctl stop act_runner.service 2>&1 | tee -a /tmp/unify_progress.log
sleep 3
cd "$CI_DIR"
rm -f .runner .runner-config.yaml
$ACT_BIN register --instance "$GITEA_URL" --token "$REG_TOKEN" \
--name "xiaoxia-ci-runner" --labels "$LABELS" --no-interactive 2>&1 | tee -a /tmp/unify_progress.log
# 修改systemd服务,改用.runner配置(原来用的是.runner-config.yaml
sed -i 's|--config .runner-config.yaml|daemon -c .runner|' /etc/systemd/system/act_runner.service
sed -i 's|ExecStart=/var/lib/xiaoxia-ci/act_runner daemon -c .runner|ExecStart=/var/lib/xiaoxia-ci/act_runner daemon -c /var/lib/xiaoxia-ci/.runner|' /etc/systemd/system/act_runner.service
systemctl daemon-reload
systemctl start act_runner.service 2>&1 | tee -a /tmp/unify_progress.log
echo "Main runner restarted" >> /tmp/unify_progress.log
# 修复runner-3runner-2正在跑这个job,最后处理)
echo "=== Fixing runner-3 ===" >> /tmp/unify_progress.log
systemctl stop act_runner-3.service 2>&1 | tee -a /tmp/unify_progress.log
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 2>&1 | tee -a /tmp/unify_progress.log
systemctl start act_runner-3.service 2>&1 | tee -a /tmp/unify_progress.log
echo "Runner-3 restarted" >> /tmp/unify_progress.log
# 等其他runner都上线了再处理runner-2自己
sleep 20
# 最后处理runner-2:把重新注册命令放到systemd的ExecStartPre里,让服务重启时自己注册
echo "=== Fixing runner-2 (self) ===" >> /tmp/unify_progress.log
# 写一个重启时执行的脚本
cat > /tmp/reregister_runner2.sh << 'INNER'
#!/bin/bash
set -e
CI_DIR="/var/lib/xiaoxia-ci"
ACT_BIN="$CI_DIR/act_runner"
cd "$CI_DIR/runner-2"
rm -f .runner
$ACT_BIN register --instance "https://git.xiaoxiajianji.com" \
--token "$REG_TOKEN" \
--name "xiaoxia-ci-runner-2" \
--labels "saas,runtime-builder,host,ubuntu-latest" \
--no-interactive
INNER
chmod +x /tmp/reregister_runner2.sh
# 让runner-2服务在启动时重新注册(用override)
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
# 重启runner-2服务(这会导致当前job断开,但标签已经在其他runner上生效了)
echo "Restarting runner-2 (will disconnect this job)..." >> /tmp/unify_progress.log
nohup systemctl restart act_runner-2.service > /dev/null 2>&1 &
echo "All done initiated" >> /tmp/unify_progress.log
EOF
chmod +x /tmp/unify_labels.sh
nohup /tmp/unify_labels.sh > /tmp/unify_full.log 2>&1 &
echo "后台脚本已启动 PID: $!"
echo "等25秒看主runner和runner-3的修复结果..."
sleep 25
echo ""
echo "=== 当前runner状态 ==="
curl -sk -H "Authorization: token 1f8058d097e3942a9ed31c44382baf7f08311272" \
"https://git.xiaoxiajianji.com/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: {len(online)}')
for r in online:
labels = [l['name'] for l in r['labels']]
print(f' {r[\"name\"]}: {labels}')
"