基于TencentOS的PVMCubeSandBox模板构建失败

模板构建失败。

该模板构建失败，无法查看详情
Job ID:
ceaa39fc-aa7b-43d4-974a-1ef430a074f7
ceaa39fc-aa7b-43d4-974a-1ef430a074f7
grep -R "ceaa39fc-aa7b-43d4-974a-1ef430a074f7" /var/log 2>/dev/null
journalctl -xe | grep ceaa39fc
docker ps | grep cube
docker logs <cube-container> | grep ceaa39fc
systemctl list-units | grep cube
cube-sandbox.service
cube-server.service
cube-agent.service
journalctl -u cube-sandbox -n 200 --no-pager
journalctl -u cube-sandbox -n 500 | egrep "ERROR|failed|ceaa39fc"
镜像:
harbor-intranet.allintechinc.com/llm-planning/llm-planning-sandb...
Harbor 镜像
      |
      v
CubeSandbox Builder
      |
      v
生成 CFS/MinIO 快照
      |
      v
Template
docker login harbor-intranet.allintechinc.com
docker pull harbor-intranet.allintechinc.com/llm-planning/llm-planning-sandbox:xxx
unauthorized: authentication required
x509: certificate signed by unknown authority
/etc/docker/certs.d/harbor-intranet.allintechinc.com/ca.crt
CUBE_S3_ENDPOINT=http://192.168.0.6:9000
bucket=cube-volumes
mc ls cube/cube-volumes
df -h
/var
/tmp
/mnt
PVM host detected
kvm_pvm loaded
lsmod | grep kvm
kvm_pvm
kvm
/dev/kvm
ls -l /dev/kvm
kubectl get template -A
docker exec -it <cube-container> sh
grep -R "tpl-464f1b5538764bd7afa240ca" /
grep -R "ceaa39fc-aa7b-43d4-974a-1ef430a074f7" /var/log 2>/dev/null
systemctl list-units | grep cube
journalctl -xe -n 200 | grep -i cube
df -h
SELECT * FROM `t_cube_template_definition`
WHERE template_id = 'tpl-464f1b5538764bd7afa240ca'
...
record not found
SELECT * FROM `t_cube_snapshot`
WHERE snapshot_id = 'tpl-464f1b5538764bd7afa240ca'
...
record not found
模板构建失败
无法查看详情
cube-sandbox-cubelet.service
cube-sandbox-cubemaster.service
cube-sandbox-cube-templatecenter.service
cube-sandbox-cube-api.service
cube-sandbox-cubeops.service
journalctl -u cube-sandbox-cubelet -n 500 --no-pager
journalctl -u cube-sandbox-cubelet --no-pager | \
grep -Ei "template|snapshot|build|error|failed|harbor|image"
journalctl -u cube-sandbox-cube-lifecycle-manager -n 500 --no-pager
journalctl -u cube-sandbox-cube-lifecycle-manager --no-pager | \
grep -Ei "template|snapshot|error|failed"
journalctl -u cube-sandbox-cube-templatecenter -n 500 --no-pager
459a081ffa37 cube-lifecycle-manager
docker logs 459a081ffa37 --tail 500
docker logs 459a081ffa37 2>&1 | \
grep -Ei "template|snapshot|error|failed"
harbor-intranet.allintechinc.com/llm-planning/...
docker ps
CUBE_PVM_ENABLE=1
lsmod | grep kvm
ls -l /dev/kvm
docker exec -it cube-sandbox-minio sh
mc alias ls
mc ls <alias>/cube-volumes
tpl-464f1b5538764bd7afa240ca
journalctl -u cube-sandbox-cubelet -n 200 --no-pager

journalctl -u cube-sandbox-cube-lifecycle-manager -n 200 --no-pager

docker logs 459a081ffa37 --tail 200
journalctl -u cube-sandbox-cubelet -n 500 --no-pager
template
snapshot
pull
image
qemu
kvm
mount
error
failed
journalctl -u cube-sandbox-cubelet --no-pager | \
grep -Ei "template|snapshot|image|pull|qemu|kvm|error|failed"
journalctl -u cube-sandbox-cube-lifecycle-manager -n 500 --no-pager
journalctl -u cube-sandbox-cube-lifecycle-manager --no-pager | \
grep -Ei "template|snapshot|build|error|failed"
journalctl -u cube-sandbox-cube-templatecenter -n 300 --no-pager
docker ps -a | grep -Ei "cube|template|builder|qemu"
CUBE_PVM_ENABLE=1
lsmod | grep kvm
ls -l /dev/kvm
/dev/kvm
harbor-intranet.allintechinc.com/llm-planning/...
curl -k https://harbor-intranet.allintechinc.com/v2/
{}
401 Unauthorized
创建模板
 ↓
CubeSandbox
 ↓
pull Harbor image
 ↓
失败
 ↓
删除 template/snapshot记录
CUBE_PVM_ENABLE=1
cubelet initial pid=1710

/usr/local/services/cubetoolbox/scripts/systemd/cubelet-start.sh: line 57:
1710 Killed

(exec "${CUBELET_BIN}" --config ... )
cubelet stable pid=1949
systemd Started cube-sandbox-cubelet.service
Consumed 10.594s CPU time, 237.9M memory peak
dmesg -T | grep -Ei "killed process|out of memory|oom"
journalctl -k | grep -Ei "oom|killed"
Out of memory: Killed process 1710 (cubelet)
Killed process 1710 (cubelet) total-vm...
free -h
cat /proc/meminfo | head
systemctl cat cube-sandbox-cubelet
MemoryLimit=
MemoryMax=
MemoryMax=256M
systemctl show cube-sandbox-cubelet | grep -i memory
"DisableHostCgroup":true,
"DisableVmCgroup":false
cat /sys/fs/cgroup/memory.events
oom 1
oom_kill 1
grep CUBELET_BIN \
/usr/local/services/cubetoolbox/scripts/systemd/cubelet-start.sh
CUBELET_BIN=/usr/local/services/cubetoolbox/Cubelet/cubelet
/usr/local/services/cubetoolbox/Cubelet/cubelet \
--config /xxx/conf.yaml \
--dynamic-conf-path /xxx/conf.yaml
journalctl -k -n 200
segfault
apparmor
selinux
denied
cube-lifecycle-manager starting

redis_addr:
127.0.0.1:6379

cubemaster_url:
http://192.168.0.6:8089

listen:
0.0.0.0:8083
bootstrap complete entries=0

http server listening

proxy joined
             cube-webui
                |
           cube-proxy
                |
        cube-lifecycle-manager
                |
             redis
                |
            cubelet  ❌
                |
          qemu/container runtime
cubelet -> sandbox runtime
dmesg -T | grep -Ei "oom|killed"
free -h
df -h
systemctl cat cube-sandbox-cubelet
ps aux | grep cubelet
journalctl -k -n 100 --no-pager
Mem: 125Gi
used: 9.7Gi
available:115Gi
Swap:0
dmesg | grep -Ei "oom|killed"
ps aux | grep cubelet
root 1949
/usr/local/services/cubetoolbox/Cubelet/bin/cubelet
Sl
cubelet initial pid=1710
1710 Killed
cubelet stable pid=1949
cubelet-start.sh
start cubelet &
wait

if alive
  use it
else
  restart
fi
sed -n '1,150p' \
/usr/local/services/cubetoolbox/scripts/systemd/cubelet-start.sh
cubelet-start.sh: line 57:
1710 Killed
nl -ba \
/usr/local/services/cubetoolbox/scripts/systemd/cubelet-start.sh \
| sed -n '45,70p'
bootstrap process
       |
       |
       +--> real cubelet
1710
exec/reload
1949
listen file[
/usr/local/services/cubetoolbox/Cubelet/dynamicconf/conf.yaml
]
cubelet
 |
加载config
 |
读取dynamicconf
 |
重新fork
 |
旧pid退出
 |
新pid接管
Killed
ss -lntp | grep cubelet
netstat -lntp | grep 1949
find /usr/local/services/cubetoolbox/Cubelet \
-name "*.log" \
-o -name "*.out"
ls -lh /usr/local/services/cubetoolbox/Cubelet/log/
qemu
builder
template
cube-proxy
cube-egress
cube-webui
cube-lifecycle-manager
redis
mysql
minio
coredns
docker ps -a | grep -Ei "qemu|sandbox|builder|template"
bootstrap complete entries=0

proxy joined

replay done
cube-master
cube-proxy
lifecycle-manager
redis
nl -ba \
/usr/local/services/cubetoolbox/scripts/systemd/cubelet-start.sh \
| sed -n '45,70p'
find /usr/local/services/cubetoolbox/Cubelet \
-type f | grep -Ei "log|err"
ss -lntp | grep cube
cat /proc/1949/cmdline | tr '\0' ' '
cubelet-start.sh: line 57: 6023 Killed
cubelet-start.sh: line 57: 1668 Killed
cubelet-start.sh: line 57: 1710 Killed
root 1949 0.5 0.1 4164304 221340 ? Sl 16:24 0:31
/usr/local/services/cubetoolbox/Cubelet/bin/cubelet
0.0.0.0:9999 cubelet
0.0.0.0:9998 cubelet
0.0.0.0:9966 cubelet
if [[ -n "${CUBE_SANDBOX_NODE_IP:-}" ]]; then
(
 export CUBE_SANDBOX_NODE_IP="${CUBE_SANDBOX_NODE_IP}"
 exec "${CUBELET_BIN}" ...
) &
fi

initial_pid=$!
6023
1668
1710
exec cubelet
refresh_pidfile_from_pattern
PATTERN
Killed 1710
stable pid=1949
cubelet-start.sh
        |
        +-- bash subshell PID 1710
                |
                exec cubelet
                |
                PID变成cubelet
PID 1949 cubelet
cubelet initial pid=6023

Killed 6023

cubelet stable pid=7079
cubelet initial pid=1668

Killed 1668

stable pid=2125
initial pid=1710

Killed 1710

stable pid=1949
Killed
cat /usr/local/services/cubetoolbox/scripts/systemd/cubelet-start.sh
PATTERN=
refresh_pidfile_from_pattern
grep PATTERN -n \
/usr/local/services/cubetoolbox/scripts/systemd/cubelet-start.sh
systemctl status cube-sandbox-cubelet
Active: active (running)
Main PID: 1949
cat /run/cube-sandbox-systemd/cubelet.pid
1949
Mem:
125Gi total
9.7Gi used
115Gi available

Swap:
0
dmesg -T | grep -Ei "oom|killed"
find /usr/local/services/cubetoolbox/Cubelet \
-type f | grep -Ei "log|err"
journalctl \
-u cube-sandbox-cubelet \
-f
curl http://127.0.0.1:9999
curl http://127.0.0.1:9966
curl -v http://127.0.0.1:9999/health
curl -v http://127.0.0.1:9999/healthz
curl -s http://127.0.0.1:9966/metrics | head
cube-lifecycle-manager
bootstrap complete
entries:0

proxy joined

replay done pushed:0 failed:0
cube-proxy
cube-webui
cube-egress
redis
mysql
minio
PID 1949
ports opened
cubelet-start.sh wrapper pid killed
cat /run/cube-sandbox-systemd/cubelet.pid
1949
Main PID: 1949 (cubelet)
cubelet-start.sh
        |
        +-- 临时启动PID 1710
        |
        +-- exec cubelet
        |
        +-- 最终PID 1949
        |
        +-- 写入pidfile
cubelet-start.sh: line 57: 1710 Killed
cubelet stable pid=1949
(
 export CUBE_SANDBOX_NODE_IP
 exec cubelet
) &
refresh_pidfile_from_pattern
/usr/local/services/cubetoolbox/Cubelet/bin/cubelet --config
1949
19:PATTERN="^${CUBELET_BIN} --config"
cat /proc/1949/cmdline | tr '\0' ' '
/usr/local/services/cubetoolbox/Cubelet/bin/cubelet --config ...
^cubelet --config
curl -s http://127.0.0.1:9966/metrics
404 page not found
/metrics
Received HTTP/0.9 when not allowed
curl http://127.0.0.1:9999/health
/data/cubelet
cubelet.sock
cubelet.sock.ttrpc
cubetap.sock
root/
state/
storage/
cubelet.sock.ttrpc
cubeops      3010
cube-api     3000
cubemaster   8089
lifecycle    8083
proxy        8082
/data/snapshot_pack/disks
drwxr-x--- 2 root root 6 disks
curl -s http://127.0.0.1:3010/api/v1/nodes | jq
ss -lntp | grep 3010
ss -lntp | grep cubelet
9966
9998
9999
curl -v http://127.0.0.1:9966/
curl -v http://127.0.0.1:9998/
/data/cube-shim/disks
ls -lah /data/cube-shim
find /usr/local/services/cubetoolbox \
-type f \
-name "*qemu*"
docker logs --tail=200 cube-proxy
error
failed
template
sandbox
create
node
cubelet
journalctl -u cube-sandbox-cubemaster -n 200 --no-pager
ps aux | grep cubemaster
find /data -maxdepth 3 \
-type d \
| grep -Ei "template|image|snapshot|box"
template/
snapshot/
image/
rootfs/
cubelet-start.sh: line 57:
1710 Killed
initial pid=1710
...
stable pid=1949
( 
 export CUBE_SANDBOX_NODE_IP
 exec cubelet ...
) &
1710
refresh_pidfile_from_pattern
1949
systemctl status
Main PID: 1949
curl http://127.0.0.1:9999/health
HTTP/0.9
9966
9998
9999
curl 9999
curl http://127.0.0.1:9966/metrics | head
404 page not found
cat /proc/1949/net/tcp
strings /usr/local/services/cubetoolbox/Cubelet/bin/cubelet | grep -Ei "metric|health|debug"
SELECT * FROM t_cube_template_definition
WHERE template_id='tpl-464f1b5538764bd7afa240ca'
rows:0
SELECT * FROM t_cube_snapshot
WHERE snapshot_id='tpl-464f1b5538764bd7afa240ca'
rows:0
template_id
tpl-464f1b5538764bd7afa240ca
t_cube_template_definition
/usr/local/services/cubetoolbox/CubeTemplateCenter
systemctl status cube-sandbox-cube-templatecenter
journalctl \
-u cube-sandbox-cube-templatecenter \
-n 200 \
--no-pager
mysql connect failed
redis failed
register template failed
grep -R "mysql\|database\|dsn" \
/usr/local/services/cubetoolbox \
-n
mysql://user:pass@tcp(...)
mysql -h xxx -u xxx -p
show tables;
select count(*) 
from t_cube_template_definition;
select 
 template_id,
 name,
 status
from t_cube_template_definition;
/usr/local/services/cubetoolbox/cube-image/
cube-guest-image-cpu.img
version
/data/snapshot_pack
drwxr-xr-x snapshot_pack
└── disks
Cubelet
CubeMaster
TemplateCenter
find /usr/local/services/cubetoolbox \
-type f \
| grep -Ei "init|template|image|import|snapshot"
cubetemplatecenter-postcheck.sh
cat \
/usr/local/services/cubetoolbox/scripts/systemd/cubetemplatecenter-postcheck.sh
ss -lntp | grep template
docker ps
CubeTemplateCenter/bin/templatecenter
systemctl status cube-sandbox-cube-templatecenter
cube-api
 |
 cube-master :8089
 |
 TemplateCenter
 |
 DB
 |
 image storage
cube-api        OK
cubemaster      OK
cubelet         OK
TemplateCenter  未确认
DB              有访问
Template数据    缺失
image           缺失
systemctl status cube-sandbox-cube-templatecenter

journalctl \
-u cube-sandbox-cube-templatecenter \
-n 100 \
--no-pager

grep -R "t_cube_template_definition" \
/usr/local/services/cubetoolbox \
-n
ls -lh \
/usr/local/services/cubetoolbox/cube-image/
Active: active (running)
Main PID: 5907
dao schema migration completed (driver= db=cube_mvp)
templatecenter successfully booted in 0.014476s
templatecenter listen address:
CUBE_TEMPLATE_CENTER_NODE_IP=192.168.0.6 ignored 
for the listen address:
http_bind="127.0.0.1" is set explicitly
HttpPort:8090
HttpBind:"127.0.0.1"
127.0.0.1:8090
/usr/local/services/cubetoolbox/CubeTemplateCenter/conf.yaml
http_bind: 127.0.0.1
http_bind: 192.168.0.6
http_bind: 0.0.0.0
systemctl restart cube-sandbox-cube-templatecenter
ss -lntp | grep 8090
LISTEN 0 4096 192.168.0.6:8090
LISTEN 0 4096 0.0.0.0:8090
grep -R "t_cube_template_definition" /usr/local/services/cubetoolbox
templatecenter binary 匹配
cubemaster binary 匹配
cubeops binary 匹配
DBName:"cube_mvp"
Addr:"127.0.0.1:3306"
User:"cube"
Pwd:"cube_pass"
mysql -h127.0.0.1 \
-u cube \
-pcube_pass \
cube_mvp
show tables;
t_cube_template_definition
show tables like '%template%';
select * from t_cube_template_definition;
id
name
version
image
status
created_at
cpu
gpu
ubuntu
centos
/usr/local/services/cubetoolbox/cube-image/

cube-guest-image-cpu.img 232M
version
CPU Guest Image
cube-guest-image-gpu.img
curl http://127.0.0.1:8090/
curl http://127.0.0.1:8090/health
curl -v http://127.0.0.1:8090/
MasterAddr:"http://127.0.0.1:8089"
CubeMaster
  |
  |
TemplateCenter
  |
  |
Cubelet
systemctl status cube-sandbox-cube-master
ss -lntp | grep 8089
127.0.0.1:8089
CubeOpsAddr:""
systemctl status cube-sandbox-cube-ops
mysql -h127.0.0.1 -ucube -pcube_pass cube_mvp
show tables like '%template%';

select count(*) from t_cube_template_definition;
curl -s http://127.0.0.1:8090/metrics | head
systemctl list-units \
| grep cube-sandbox
HttpPort:8090
HttpBind:"127.0.0.1"
templatecenter successfully booted in 0.014476s
templatecenter listen address:
CUBE_TEMPLATE_CENTER_NODE_IP=192.168.0.6 ignored for the listen address:
http_bind="127.0.0.1" is set explicitly
http_bind="192.168.0.6"
systemctl status cube-sandbox-cube-templatecenter -l
journalctl -xeu cube-sandbox-cube-templatecenter --no-pager
bind: cannot assign requested address
ip addr | grep 192.168.0.6
inet 192.168.0.6/xx
listen tcp 192.168.0.6:8090: bind: address already in use
ss -lntp | grep 8090
grep -n "http_bind" \
/usr/local/services/cubetoolbox/CubeTemplateCenter/conf.yaml
http_bind: "192.168.0.6"
http_bind=192.168.0.6
http_bind:192.168.0.6
http_bind: "0.0.0.0"
Common:
  HttpPort: 8090
  HttpBind: "0.0.0.0"
TemplateCenter
        |
        |
   192.168.0.6:8090
        |
其它Cube节点访问
0.0.0.0:8090
192.168.0.6:8090
vi /usr/local/services/cubetoolbox/CubeTemplateCenter/conf.yaml
http_bind: "0.0.0.0"
systemctl daemon-reload

systemctl restart cube-sandbox-cube-templatecenter
ss -lntp | grep 8090
LISTEN 0 4096 0.0.0.0:8090
MasterAddr:"http://127.0.0.1:8089"
grep -R "TemplateCenter" \
/usr/local/services/cubetoolbox/CubeMaster/conf.yaml
grep -R "CubeTemplateCenter" \
/usr/local/services/cubetoolbox
http://192.168.0.6:8090
LocalIp":"192.168.0.6"
CUBE_TEMPLATE_CENTER_NODE_IP=192.168.0.6
CUBE_TEMPLATE_CENTER_NODE_IP=192.168.0.6

http_bind=0.0.0.0

HttpPort=8090
http://192.168.0.6:8090
templatecenter successfully booted in 0.009064s
[one-click-systemd] ERROR wait_for_http timeout:
url=http://127.0.0.1:8090/health
curl: (7) Failed to connect to 127.0.0.1 port 8090
ExecStartPost=/usr/bin/bash ... cubetemplatecenter-postcheck.sh
status=1/FAILURE
templatecenter received signal terminated, shutting down
HttpBind: "127.0.0.1"
HttpBind: "192.168.0.6"
HttpBind: "0.0.0.0"
curl http://127.0.0.1:8090/health
192.168.0.6:8090
127.0.0.1:8090
ss -lntp | grep 8090
LISTEN 0 4096 192.168.0.6:8090
LISTEN 0 4096 0.0.0.0:8090
curl http://192.168.0.6:8090/health
curl http://127.0.0.1:8090/health
cat /usr/local/services/cubetoolbox/scripts/systemd/cubetemplatecenter-postcheck.sh
wait_for_http http://127.0.0.1:8090/health
wait_for_http http://192.168.0.6:8090/health
CUBE_TEMPLATE_CENTER_NODE_IP=192.168.0.6
wait_for_http http://${CUBE_TEMPLATE_CENTER_NODE_IP}:8090/health
systemctl daemon-reload

systemctl restart cube-sandbox-cube-templatecenter
Common:
  HttpBind: "0.0.0.0"
  HttpPort: 8090
curl http://127.0.0.1:8090/health
                    +-------------+
192.168.0.6 ------> |             |
                   | TemplateCtr |
127.0.0.1 --------> |             |
                    +-------------+
                         :8090
HttpBind: "0.0.0.0"
127.0.0.1,192.168.0.6
MasterAddr":"http://192.168.0.6:8089"
MasterAddr":"http://127.0.0.1:8089"
ss -lntp | grep 8089
192.168.0.6:8089
0.0.0.0:8089
Common:
  HttpBind: "0.0.0.0"
  HttpPort: 8090

  MasterAddr: "http://192.168.0.6:8089"
export CUBE_TEMPLATE_CENTER_NODE_IP=192.168.0.6
ExecStartPost=/usr/bin/bash /usr/local/services/cubetoolbox/scripts/systemd/cubetemplatecenter-postcheck.sh
curl -f http://127.0.0.1:8090/health
systemctl daemon-reload
systemctl restart cube-sandbox-cube-templatecenter

systemctl status cube-sandbox-cube-templatecenter -l

组件	建议
cubelet	>=2G
cube-proxy	512M
redis/mysql/minio	>=3G
模板构建	>=4G
运行 sandbox	8G+

项目	状态
宿主机内存	✅正常
磁盘	✅正常
Docker组件	✅正常
Lifecycle Manager	✅正常
Cubelet最终进程	✅正常
首次pid killed	⚠️启动脚本/自重启机制

组件	状态
cube-api	✅
cubeops	✅
cubemaster	✅
cube-proxy	✅
cube-lifecycle-manager	✅
redis	✅
mysql	✅
minio	✅
cubelet	✅
OOM	❌ 排除
磁盘压力	❌ 排除
cubelet启动失败	❌ 排除
template/image	⚠️ 未发现

组件	状态
CubeTemplateCenter	✅正常
MySQL	✅正常
Redis	未知
CubeMaster	需要确认
模板数据	未知
镜像	只有CPU

配置	效果
127.0.0.1	只能本机访问
192.168.0.6	只监听这个网卡
0.0.0.0	监听所有网卡

端口	可能用途
9966	metrics
9998	RPC
9999	HTTP API

端口	用途
9966	metrics
9998	ttrpc
9999	cubelet grpc/http2

基于TencentOS的PVMCubeSandBox模板构建失败

1. 根据 Job ID 查构建日志（第一步）

2. 查看 CubeSandbox 服务日志

3. 你的截图里有一个重要信息

4. 检查 Harbor 镜像访问

认证失败

TLS问题

5. 检查 MinIO/CFS

6. 检查 PVM/KVM

7. 删除失败模板

8. 建议你现在执行这几个命令

① 找 Job

② 看服务

③ 看最近错误

④ 看磁盘

当前重点：找真正的 builder 日志

1. 查看 cubelet 日志（优先）

2. 查看 lifecycle-manager

3. 查看 TemplateCenter

4. Docker 容器日志

5. 另外发现一个可能问题

情况A

情况B

6. 检查 MinIO 是否有模板残留

7. 建议下一步直接执行这几个

1. Cubelet（最重要）

2. Lifecycle Manager

3. TemplateCenter

4. 查看是否有临时构建容器

5. 检查 PVM 环境

6. 检查 Harbor 连通性

我目前最怀疑的两个点

可能1：Harbor 私有镜像无法拉取

可能2：PVM 模板转换失败

1. 最大可能原因：OOM Killer 杀掉 cubelet

2. 检查机器内存

3. 检查 systemd 是否限制内存

4. 检查 cgroup OOM

5. cubelet binary 单独启动测试

6. 检查 kernel 是否因为权限杀

7. lifecycle-manager 正常

8. 当前架构状态

9. 建议下一步执行

A. 看 OOM

B. 看机器资源

C. 看 cubelet service

D. 看进程

E. 看 kernel

关键证据

1. 宿主机资源完全正常

2. 当前 cubelet 实际运行正常

3. 真正异常是这个：

下一步重点查 cubelet-start.sh

第二个重点：为什么第一次启动被杀？

可能1：cubelet 自带 watchdog/reexec

可能2：动态配置 reload

3. 检查 cubelet 健康接口

4. 查看 cubelet 日志目录

5. 检查是否有 sandbox runtime

6. lifecycle-manager 状态正常

我建议现在执行这几个命令

① 看启动脚本第57行

② 看 cubelet 日志

③ 看端口

④ 看 cubelet完整启动参数

1. 先解释为什么看到 Killed

2. 但是有一个值得检查的问题

3. 当前 cubelet 状态确认

4. 当前资源没有问题

5. cubelet 自己日志在哪里？

6. 建议检查 cubelet health

7. 我认为当前真正问题点

1. cubelet PID 管理正常

2. Killed 不是异常

3. PATTERN 匹配也正确

4. cubelet API 判断

5. 当前目录状态正常

现在真正需要排查的是：为什么没有 Sandbox / Template

下一步重点检查

1. 看 cubelet 注册状态

2. `Killed` 不是异常