星辰狂澜's Blog

发表于 分类于
admin

docker pull 通过 docker-group 既可以拉取 docker-hosted 镜像又可以拉取 docker-proxy 的镜像。

由于 nexus3 的限制，如果只代理至 docker-group 的话，无法推送镜像，所以需要将推送代理至 docker-hosted。

默认的 GET 与 HEAD 请求代理至 docker-group，其余的 PUT POST 请求代理至 docker-group。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
http {

    map $request_method $repo_name {
        default hosted;
        GET group;
        HEAD group;
    }

    server {

        location /v2/ {
            client_max_body_size 8000M;
            rewrite /v2/(.*) /repository/docker-$repo_name/v2/$1 break;
            proxy_pass http://${IP}:8081;
            proxy_set_header Host $host;
            proxy_set_header X-Real-IP $remote_addr;
            proxy_set_header X-Real-PORT $remote_port;
            proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
            proxy_set_header X-Forwarded-Proto $scheme;
        }
    }

}

发表于 分类于

Proto & Buf

  • buf.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
# yaml-language-server: $schema=https://json.schemastore.org/buf.json

version: v2
deps:
  - buf.build/googleapis/googleapis
  - buf.build/envoyproxy/protoc-gen-validate
  - buf.build/grpc-ecosystem/grpc-gateway
lint:
  use:
    - BASIC
  ignore_only:
    IMPORT_NO_PUBLIC:
      - common/common.proto
breaking:
  use:
    - FILE
  • buf.gen.yaml
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
# yaml-language-server: $schema=https://json.schemastore.org/buf.gen.json

version: v2
plugins:
  # https://github.com/protocolbuffers/protobuf-go
  - local: protoc-gen-go
    out: gen_go
    opt:
      - paths=source_relative

  # https://github.com/grpc/grpc-go
  - local: protoc-gen-go-grpc
    out: gen_go
    opt:
      - paths=source_relative

  # github.com/envoyproxy/protoc-gen-validate
  - local: protoc-gen-validate
    out: gen_go
    opt:
      - paths=source_relative
      - lang=go

  # github.com/grpc-ecosystem/grpc-gateway/v2/protoc-gen-grpc-gateway
  # - local: protoc-gen-grpc-gateway
  #   out: gen_go
  #   opt:
  #     - paths=source_relative

  # github.com/go-kratos/kratos/cmd/protoc-gen-go-http/v2
  - local: protoc-gen-go-http
    out: gen_go
    opt:
      - paths=source_relative

  # github.com/grpc-ecosystem/grpc-gateway/v2/protoc-gen-openapiv2
  - local: protoc-gen-openapiv2
    strategy: all
    out: gen_doc
    opt:
      - allow_merge=true
      - include_package_in_tags=true
      - fqn_for_openapi_name=true
      - use_go_templates=true
      - disable_default_errors=true
      - enums_as_ints=true
      - output_format=yaml
      - preserve_rpc_order=true

流水线

服务连接管理

webui

构建集群

webui

生成并发布 Proto

将 api-pb 所生成的 go 代码推送至 api-go 仓库，以便 goprivate 使用。

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
defaultWorkspace: api_pb

sources:
  api_pb:
    type: codeup
    endpoint: git@codeup.aliyun.com:starudream/os/api-pb.git
    certificate:
      type: serviceConnection
      serviceConnection: ${云效Codeup服务连接ID}
    triggerEvents:
      - push
  api_go:
    type: codeup
    endpoint: git@codeup.aliyun.com:starudream/os/api-go.git
    certificate:
      type: serviceConnection
      serviceConnection: ${云效Codeup服务连接ID}
    triggerEvents:
      - tagPush # 防止 push 时触发该工作流

stages:
  proto:
    jobs:
      gen:
        timeoutMinutes: 30
        runsOn:
          group: private/${私有构建集群ID}
        steps:
          gen:
            step: CustomEnvironmentBuild
            with:
              image: registry.cn-shanghai.aliyuncs.com/starudream/go-dev
              privateRegistry: true
              certificate:
                type: serviceConnection
                serviceConnection: ${ACR服务连接ID}
              run: |
                go version
                buf --version
                make clean gen
                tar -czf pb_gen_${PIPELINE_ID}_${BUILD_NUMBER}.tar.gz gen_go gen_doc
                mkdir -p /cache/proto/ && mv -f pb_gen_${PIPELINE_ID}_${BUILD_NUMBER}.tar.gz /cache/proto/

      pub:
        timeoutMinutes: 30
        runsOn:
          group: private/${私有构建集群ID}
          vm: true
        needs: gen
        steps:
          go:
            step: Command
            with:
              run: |
                pushd /data/yunxiao/cache/proto/ >/dev/null
                rm -rf ${PIPELINE_ID}_${BUILD_NUMBER} && mkdir -p ${PIPELINE_ID}_${BUILD_NUMBER}
                tar -xzf pb_gen_${PIPELINE_ID}_${BUILD_NUMBER}.tar.gz -C ${PIPELINE_ID}_${BUILD_NUMBER}
                mv -f ${WORK_SPACE}/api_go/.git ${PIPELINE_ID}_${BUILD_NUMBER}/gen_go/.git

                pushd ${PIPELINE_ID}_${BUILD_NUMBER}/gen_go >/dev/null
                git config --global user.name "ci-bot"
                git config --global user.email "ci-bot@mail.starudream.cn"
                git add . && git commit -m "${CI_COMMIT_ID_1} -> ${CI_COMMIT_TITLE_1}" && git push
                popd >/dev/null

                rm -rf ${PIPELINE_ID}_${BUILD_NUMBER}
webui

GoPrivate

Configuration

1
2
3
4
go env -w CGO_ENABLED=0
go env -w GO111MODULE=on
go env -w GOPRIVATE=codeup.aliyun.com
go env -w GOPROXY=https://goproxy.cn,direct

ssh key

1
replace codeup.aliyun.com/starudream/os/api-go => codeup.aliyun.com/starudream/os/api-go.git

需要手动指定版本，不能更新，比较麻烦。

netrc

  • windows: $HOME/_netrc
  • linux: $HOME/.netrc

可以使用 go get -u 自动更新。

Backend

1
2
3
4
5
6
7
8
module codeup.aliyun.com/starudream/os/backend

go 1.23.0

require (
	codeup.aliyun.com/starudream/os/api-go v0.0.0-20241111023212-d66b05b24117
	codeup.aliyun.com/starudream/os/lib-go v0.0.0-20241111082247-43cf457f0683
)

发表于 分类于

docker-compose.yaml

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
networks:
  ollama:
    external: true

services:
  ollama:
    image: ollama/ollama:0.4.1
    container_name: ollama
    restart: always
    networks:
      - ollama
    deploy:
      resources:
        reservations:
          # noinspection ComposeUnknownValues
          devices:
            - driver: nvidia
              capabilities: [ gpu ]
    environment:
      - OLLAMA_KEEP_ALIVE=24h
    volumes:
      - "${COMPOSE_DATA_DIR:-/data}/ollama/ollama:/root/.ollama"
    ports:
      - "11434:11434"

  ollama-webui:
    image: ghcr.io/open-webui/open-webui:0.3.35
    container_name: ollama-webui
    restart: always
    networks:
      - ollama
    environment:
      - OLLAMA_BASE_URL=http://ollama:11434
      - ENV=dev
      - WEBUI_AUTH=False
      - WHISPER_MODEL_AUTO_UPDATE=False
      - RAG_EMBEDDING_MODEL_AUTO_UPDATE=False
      - RAG_RERANKING_MODEL_AUTO_UPDATE=False
    volumes:
      - "${COMPOSE_DATA_DIR:-/data}/ollama/webui:/app/backend/data"
    ports:
      - "8080:8080"

拉取模型

  • https://ollama.com/library
  • https://ollama.com/library/qwen2.5
1
docker exec -it ollama bash
1
2
3
ollama list
ollama pull qwen2.5
ollama list

使用模型

打开 http://${IP}:8080 访问 webui，左上角选择模型 qwen2.5：

webui 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 565.57.01              Driver Version: 565.57.01      CUDA Version: 12.7     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  Tesla T4                       Off |   00000000:00:10.0 Off |                    0 |
| N/A   33C    P0             25W /   70W |    5255MiB /  15360MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  Tesla T4                       Off |   00000000:00:11.0 Off |                    0 |
| N/A   24C    P8              9W /   70W |       3MiB /  15360MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|    0   N/A  N/A    326540      C   ...unners/cuda_v12/ollama_llama_server       5252MiB |
+-----------------------------------------------------------------------------------------+

发表于 分类于

查看 GPU 信息

1
lspci | grep -i nvidia

驱动

1
yum install -y kernel-devel gcc

Test

1
nvidia-smi

容器工具包

  • Install Guide
  • USTC Mirrors

Install

1
2
curl -s -L https://mirrors.ustc.edu.cn/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo | \
  sed 's#nvidia.github.io#mirrors.ustc.edu.cn#g' | tee /etc/yum.repos.d/nvidia-container-toolkit.repo
1
yum install -y nvidia-container-toolkit

Configure

1
nvidia-ctk runtime configure --runtime=docker

equivalent to edit /etc/docker/daemon.json add following content:

1
2
3
4
5
6
7
8
{
  "runtimes": {
    "nvidia": {
      "args": [],
      "path": "nvidia-container-runtime"
    }
  }
}

then restart docker daemon

1
systemctl restart docker

Test

1
docker run --rm --runtime=nvidia --gpus all ubuntu:24.04 nvidia-smi
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
+-----------------------------------------------------------------------------------------+
| NVIDIA-SMI 565.57.01              Driver Version: 565.57.01      CUDA Version: 12.7     |
|-----------------------------------------+------------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id          Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |           Memory-Usage | GPU-Util  Compute M. |
|                                         |                        |               MIG M. |
|=========================================+========================+======================|
|   0  Tesla T4                       Off |   00000000:00:10.0 Off |                    0 |
| N/A   24C    P8              8W /   70W |       3MiB /  15360MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+
|   1  Tesla T4                       Off |   00000000:00:11.0 Off |                    0 |
| N/A   25C    P8              9W /   70W |       3MiB /  15360MiB |      0%      Default |
|                                         |                        |                  N/A |
+-----------------------------------------+------------------------+----------------------+

+-----------------------------------------------------------------------------------------+
| Processes:                                                                              |
|  GPU   GI   CI        PID   Type   Process name                              GPU Memory |
|        ID   ID                                                               Usage      |
|=========================================================================================|
|  No running processes found                                                             |
+-----------------------------------------------------------------------------------------+

发表于 分类于

Import

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
#!/usr/bin/env bash

set -e

REGISTRY="hub.starudream.local"

function load() {
  if [ ! -f "$1" ]; then
    echo "file not found: $1"
    exit 1
  fi
  res=$(docker image load -i "$1" 2>/dev/null | grep "Loaded image:" || true)
  images=${res//Loaded image: /}
  for image in $images; do
    echo "-> loaded $image"
    if [ -z $REGISTRY ]; then
      continue
    fi
    cnt=$(echo "$image" | awk -F/ "{print NF-1}")
    name=$image
    case $cnt in
    0)
      name="library/$image"
      ;;
    1) ;;
    *)
      part=${image%/*}
      left=${part%/*}
      name=${image/$left\//}
      ;;
    esac
    docker image tag "$image" "$REGISTRY/$name"
    docker image push "$REGISTRY/$name"
    echo "=> pushed $image to $REGISTRY/$name"
    docker image rm "$REGISTRY/$name"
    docker image rm "$image" || true
  done
}

cnt=$(grep -c $REGISTRY /etc/hosts || true)
if [ "$cnt" -eq 0 ]; then
  REGISTRY=""
elif [ "$PUSH_REGISTRY" = 0 ]; then
  REGISTRY=""
fi

if [ -z "$1" ]; then
  files=$(find . -type f -name "*.tar.gz" | sort -u)
  for file in $files; do
    echo ">> found $file"
    load "$file"
  done
else
  load "$1"
fi

Export

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
#!/usr/bin/env bash

set -e

CUR_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
ROOT_DIR=$(dirname "$CUR_DIR")

mkdir -p "$ROOT_DIR"/images

function save() {
  if [ ! -f "$1" ]; then
    echo "file not found: $1"
    exit 1
  fi
  images=$(docker-compose -f "$1" config --images | sort -u)
  for image in $images; do
    name=${image////-}
    name=${name//:/_}
    if [ -f "$ROOT_DIR/images/$name.tar.gz" ]; then
      echo "-> exist $name.tar.gz, skip"
      continue
    fi
    docker image pull --platform linux/amd64 "$image"
    echo "=> save $image to $name.tar.gz"
    docker image save "$image" | gzip > "$ROOT_DIR/images/$name.tar.gz"
  done
}

if [ -z "$1" ]; then
  files=$(find . -type f -name "docker-compose*.yaml" | sort -u)
  for file in $files; do
    echo ">> found $file"
    save "$file"
  done
else
  save "$1"
fi