coze离线部署问题解决

coze部署参考链接：coze部署
找到coze的git仓库，直接下载coze-studio-main源码即可：官方coze仓库

1. coze离线部署运行docker-compose --profile ‘*’ up -d时报错

Error response form daemon:Get "https://registry-1.docker.io/v2/ ": context deadline exceeded(Client.Timeout exceeded while awaiting headers

原因：未联网，需要在外网下载镜像后导入docker中，因为外网Windows机没有装docker，使用python代码的方式进行下载(需要科学上网)：

import os
import io
import json
import tarfile
import requests
from time import sleep

REGISTRY = "registry-1.docker.io"

def get_token(repo, scope="pull"):
    url = f"https://auth.docker.io/token?service=registry.docker.io&scope=repository:{repo}:{scope}"
    r = requests.get(url, timeout=30)
    r.raise_for_status()
    return r.json()["token"]

def get_manifest(repo, reference, token):
    headers = {
        "Authorization": f"Bearer {token}",
        "Accept": (
            "application/vnd.docker.distribution.manifest.v2+json,"
            "application/vnd.docker.distribution.manifest.list.v2+json,"
            "application/vnd.oci.image.index.v1+json,"
            "application/vnd.oci.image.manifest.v1+json"
        )
    }
    url = f"https://{REGISTRY}/v2/{repo}/manifests/{reference}"
    r = requests.get(url, headers=headers, timeout=60)
    r.raise_for_status()
    return r.json(), r.headers.get("Content-Type")

def download_blob(repo, digest, token, outdir, max_retries=5):
    """下载镜像层，支持断点续传 + 自动刷新 token"""
    path = os.path.join(outdir, digest.replace(":", "_"))
    url = f"https://{REGISTRY}/v2/{repo}/blobs/{digest}"

    for attempt in range(max_retries):
        try:
            mode = "ab" if os.path.exists(path) else "wb"
            existing_size = os.path.getsize(path) if os.path.exists(path) else 0

            headers = {"Authorization": f"Bearer {token}"}
            if existing_size > 0:
                headers["Range"] = f"bytes={existing_size}-"

            with requests.get(url, headers=headers, stream=True, timeout=120) as r:
                if r.status_code == 401:  # token 失效，重新获取
                    print(f"🔑 Token expired, refreshing for {digest}")
                    token = get_token(repo, "pull")
                    continue
                r.raise_for_status()

                with open(path, mode) as f:
                    for chunk in r.iter_content(1024 * 1024):  # 1MB buffer
                        if chunk:
                            f.write(chunk)

            return path  # 成功就返回

        except Exception as e:
            print(f"⚠️ Error downloading {digest} (attempt {attempt+1}/{max_retries}): {e}")
            sleep(3)

    raise RuntimeError(f"❌ Failed to download {digest} after {max_retries} attempts")

def save_tar(repo, tag, config_file, layer_files, outname):
    with tarfile.open(outname, "w") as tar:
        # manifest.json
        manifest_json = [{
            "Config": os.path.basename(config_file),
            "RepoTags": [f"{repo}:{tag}"],
            "Layers": [os.path.basename(x) for x in layer_files]
        }]
        data = json.dumps(manifest_json, indent=2).encode("utf-8")
        info = tarfile.TarInfo("manifest.json")
        info.size = len(data)
        tar.addfile(info, io.BytesIO(data))

        # repositories
        repositories = {repo: {tag: manifest_json[0]["Layers"][0]}}
        data = json.dumps(repositories, indent=2).encode("utf-8")
        info = tarfile.TarInfo("repositories")
        info.size = len(data)
        tar.addfile(info, io.BytesIO(data))

        # config.json (真实的 config)
        tar.add(config_file, arcname=os.path.basename(config_file))

        # layers
        for lf in layer_files:
            tar.add(lf, arcname=os.path.basename(lf))

def pull_image(image, outdir="images"):
    if ":" in image:
        repo, tag = image.split(":", 1)
    else:
        repo, tag = image, "latest"

    if "/" not in repo:
        repo = f"library/{repo}"  # 补全前缀

    token = get_token(repo)
    manifest, ctype = get_manifest(repo, tag, token)

    # 选择 amd64/linux 架构
    if "list.v2" in ctype or "image.index" in ctype:
        amd64 = next(
            m for m in manifest["manifests"]
            if m["platform"]["architecture"] == "amd64" and m["platform"]["os"] == "linux"
        )
        manifest, ctype = get_manifest(repo, amd64["digest"], token)

    config_digest = manifest["config"]["digest"]
    layers = [l["digest"] for l in manifest["layers"]]

    os.makedirs(outdir, exist_ok=True)

    # 下载 config.json （非常关键，确保 entrypoint/cmd 等信息保留）
    config_file = download_blob(repo, config_digest, token, outdir)
    layer_files = [download_blob(repo, d, token, outdir) for d in layers]

    outname = os.path.join(outdir, f"{repo.replace('/', '_')}_{tag}.tar")
    save_tar(repo, tag, config_file, layer_files, outname)
    print(f"✅ Saved {image} -> {outname}")

if __name__ == "__main__":
    images = [
    	"golang:1.24-alpine",
    	"alpine:3.20",
    	"alpine:3.22.0",
        "mysql:8.4.5",
        "bitnami/redis:8.0",
        "bitnami/elasticsearch:8.18.0",
        "minio/minio:RELEASE.2025-06-13T11-33-47Z-cpuv1",
        "bitnami/etcd:3.5",
        "milvusdb/milvus:v2.5.10",
        "nsqio/nsq:v1.2.1",
        "opencoze/opencoze:latest",
        "cozedev/coze-studio-server:latest",
        "cozedev/coze-studio-web:latest"
    ]
    for img in images:
        pull_image(img)

下载完成后如下所示：

将其导入到内网机中，然后docker进行load安装：

docker load -i /path/XXX.tar

docker load命令参考

完成之后，执行以下命令进行coze配置：

此时遇到问题：

输入命令:

docker-compose run coze-server /bin/sh
ls -l /app

可以看到没有bootstrap.sh文件：

因此需要自己将其导入，在文件路径coze-studio/backend/script路径下能够找到bootstrap.sh

所以修改docker目录下的docker-compose.yml文件，在第342行添加以下内容：

• …/backend/script/bootstrap.sh:app/bootstrap.sh
  

保存之后再次运行：

docker-compose run coze-server /bin/sh
ls -l /app

注意文件权限，如果权限不全，则运行:

chmod 777 bootstrap.sh

然后再次运行

docker–compose --profile ‘*’ up -d

结果如下：

在浏览器输入:localhost:8888,注册使用即可

2. Coze发布的Agent在其他服务器上远程访问

在docker文件夹下输入:

vim .env

找到WEB_LISTEN_ADDR,将127.0.0.1：8888改为0.0.0.0:8888即可；