鲲鹏服务器,300I-Duo(310P3)npu卡,部署Qwen3-32B,容器内启动 ./mindieservice_daemon报权限错误
收藏回复举报
鲲鹏服务器,300I-Duo(310P3)npu卡,部署Qwen3-32B,容器内启动 ./mindieservice_daemon报权限错误
t('forum.solved') 已解决
新人帖
发表于2025-08-26 14:48:26
0 查看

前期配置:
1、系统:Ubuntu 22.04.5 LTS,aarch64,300I-Duo(310P3)npu卡。

2、镜像下载:1.0.0-300I-Duo-py311-openeuler24.03-lts

cke_4614.png

3、模型下载:魔乐社区(通过modelscope下载)

cke_5129.png

4、启动镜像前文件权限确认

HwHiAiUser用户组ID:

(base) ubuntu@ubuntu:/data/workspace/HuaWei_Ascend_300I/models/Qwen3_32B$ id HwHiAiUser
uid=1001(HwHiAiUser) gid=1001(HwHiAiUser) groups=1001(HwHiAiUser)

驱动目录权限:

(base) ubuntu@ubuntu:/data/workspace/HuaWei_Ascend_300I/models/Qwen3_32B$ ls -ld /usr/local/Ascend/driver
drwxr-xr-x 9 root root 4096 Aug  4 03:38 /usr/local/Ascend/driver

NPU卡权限:

(base) ubuntu@ubuntu:/data/workspace/HuaWei_Ascend_300I/models/Qwen3_32B$ ls -l /dev/davinci*
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 0 Aug  9 02:27 /dev/davinci0
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 1 Aug  9 02:27 /dev/davinci1
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 2 Aug  9 02:27 /dev/davinci2
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 3 Aug  9 02:27 /dev/davinci3
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 4 Aug  9 02:27 /dev/davinci4
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 5 Aug  9 02:27 /dev/davinci5
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 6 Aug  9 02:27 /dev/davinci6
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 507, 7 Aug  9 02:27 /dev/davinci7
crw-rw-rw- 1 HwHiAiUser HwHiAiUser 508, 0 Aug  9 02:26 /dev/davinci_manager

模型权限:

(base) ubuntu@ubuntu:/data/workspace/HuaWei_Ascend_300I/models/Qwen3_32B$ ls -l /data/workspace/HuaWei_Ascend_300I
total 4
drwxr-xr-x 3 ubuntu ubuntu 4096 Aug 25 07:31 models

根据权限,有两种进入容器的参数配置。

1.以普通用户组的身份进入容器

docker run -it --net=host --ipc=host \
    --user 1001:1001 \
    --name Qwen3-32B \
    --device=/dev/davinci_manager \
    --device=/dev/hisi_hdc \
    --device=/dev/devmm_svm \
    --device=/dev/davinci6 \
    --device=/dev/davinci7 \
    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver:ro \
    -v /usr/local/sbin:/usr/local/sbin:ro \
    -v /data/workspace/HuaWei_Ascend_300I/models:/home/models:rw \
swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-300I-Duo-py311-openeuler24.03-lts /bin/bash

先将模型目录权限给了HwHiAiUser:

(base) ubuntu@ubuntu:/data/workspace/HuaWei_Ascend_300I/models/Qwen3_32B$ ls -l /data/workspace/HuaWei_Ascend_300I/models
total 4
drwxrwxr-x 3 HwHiAiUser HwHiAiUser 4096 Aug 25 09:15 Qwen3_32B

但进入容器后没有权限执行以下命令

vim /usr/local/Ascend/mindie/latest/mindie-service/conf/config.json

此时可以以root用户身份再进容器修改配置文件。

但没有权限进入服务化目录拉起服务化。所以放弃,

2.以root用户身份进去容器

docker run -it --net=host --ipc=host \
    --user 1001:1001 \
    --name Qwen3-32B \
    --device=/dev/davinci_manager \
    --device=/dev/hisi_hdc \
    --device=/dev/devmm_svm \
    --device=/dev/davinci6 \
    --device=/dev/davinci7 \
    -v /usr/local/Ascend/driver:/usr/local/Ascend/driver:ro \
    -v /usr/local/sbin:/usr/local/sbin:ro \
    -v /data/workspace/HuaWei_Ascend_300I/models:/home/models:rw \
swr.cn-south-1.myhuaweicloud.com/ascendhub/mindie:1.0.0-300I-Duo-py311-openeuler24.03-lts /bin/bash

此时已经把模型目录权限修改回root下

(base) ubuntu@ubuntu:/data/workspace/HuaWei_Ascend_300I$ ls -al
total 12
drwxr-xr-x  3 root root 4096 Aug 25 07:30 .
drwxr-xr-x 12 root root 4096 Aug 25 07:12 ..
drwxr-xr-x  3 root root 4096 Aug 25 07:31 models

可以正常修改配置文件,并进入服务化目录,但是无法拉起服务化。一直报权限问题的错误:

[root@ubuntu bin]# cd /usr/local/Ascend/mindie/latest
[root@ubuntu latest]# ls -al mindie-service/conf/config.json
-rw-r----- 1 root root 3888 Aug 26 13:33 mindie-service/conf/config.json
[root@ubuntu latest]# chmod 750 mindie-service/conf/config.json
[root@ubuntu latest]# ls -al mindie-service/conf/config.json
-rwxr-x--- 1 root root 3888 Aug 26 13:33 mindie-service/conf/config.json
[root@ubuntu latest]# cd /usr/local/Ascend/mindie/latest/mindie-service/bin
[root@ubuntu bin]# ./mindieservice_daemon
Check path: config.json failed, by: Check Owner permission failed: Current permission is 7, but required no greater than 6. Required permission: 640, but got 750
ConfigManager: Get config path failed.
[2025-08-26 13:56:02.179002][107][ubuntu][system][start][mindie server][fail]
[2025-08-26 13:56:02.182122][107][ubuntu][system][start][endpoint][fail]
[2025-08-26 13:56:02.182140][107][ubuntu][system][start][mindie server][fail]
ERR: Failed to init endpoint! Please check the service log or console output.
Killed
[root@ubuntu bin]# cd /usr/local/Ascend/mindie/latest/mindie-service/conf
[root@ubuntu conf]# chmod 640 config.json
[root@ubuntu conf]# ls -al config.json
-rw-r----- 1 root root 3888 Aug 26 13:33 config.json
[root@ubuntu Ascend]# chmod 750 /usr/local/Ascend/mindie/latest/mindie-service/conf
[root@ubuntu Ascend]# chmod 750 /usr/local/Ascend/mindie/latest/mindie-service
[root@ubuntu Ascend]# chmod 750 /usr/local/Ascend/mindie/latest
[root@ubuntu Ascend]# cd /usr/local/Ascend/mindie/latest/mindie-service/bin
[root@ubuntu bin]# ./mindieservice_daemon
Check path: config.json failed, by: Check Owner group permission failed: Current permission is 6, but required no greater than 5. Required permission: 750, but got 664
Failed to check config.json under model weight path.
[2025-08-26 14:10:51.157738][116][ubuntu][system][start][mindie server][fail]
[2025-08-26 14:10:51.160895][116][ubuntu][system][start][endpoint][fail]
[2025-08-26 14:10:51.160913][116][ubuntu][system][start][mindie server][fail]
ERR: Failed to init endpoint! Please check the service log or console output.
Killed

已经按社区中提到的修改权限问题,但还是不行,这是怎么回事?

注意到几次权限错误信息有变化,by: Check Owner permission failed和by: Check Owner group permission failed:。

这种错误该怎么解决?或者是因为镜像的问题?

我要发帖子