清理UI中的鏡像

清理ui中的鏡像，如果直接通過圖形界面一個個的點擊刪除的話，在有大規模鏡像需要清理的時候，簡直就是災難，而且這種方式，實在太low。

我這里簡單寫了個腳本，以實現如下功能：

遍歷所有project
獲取project中所有tag數超過30的repositories
獲取這些tag數超過30的repositories的所有tag
基於時間排序，保留最新的30個tag
刪除其他tag

腳本示例：

#! /usr/bin/env python
# -*- coding:utf-8 -*-


import requests
import json


class RequestClient(object):

    def __init__(self,login_url, username, password):
        self.username = username
        self.password =  password
        self.login_url = login_url
        self.session = requests.Session()
        self.login()

    def login(self):
        self.session.post(self.login_url, params={"principal": self.username, "password": self.password})

class ClearHarbor(object):
    
    def __init__(self, harbor_domain, password, schema="https",
                 username="admin"):
        self.schema = schema
        self.harbor_domain = harbor_domain
        self.harbor_url = self.schema + "://" + self.harbor_domain
        self.login_url = self.harbor_url + "/login"
        self.api_url = self.harbor_url + "/api"
        self.pro_url = self.api_url + "/projects"
        self.repos_url = self.api_url + "/repositories"
        self.username = username
        self.password = password
        self.client = RequestClient(self.login_url, self.username, self.password)

    def __fetch_pros_obj(self):
        # TODO
        self.pros_obj = self.client.session.get(self.pro_url).json()
        return self.pros_obj

    def fetch_pros_id(self):
        self.pros_id = []
        # TODO
        pro_res = self.__fetch_pros_obj()
        for i in pro_res:
            self.pros_id.append(i['project_id'])
        return self.pros_id

    def fetch_del_repos_name(self, pro_id):
        self.del_repos_name = []
        repos_res = self.client.session.get(self.repos_url, params={"project_id": pro_id})
        # TODO
        for repo in repos_res.json():
            if repo["tags_count"] > 30: 
                self.del_repos_name.append(repo['name'])
        return self.del_repos_name

    def fetch_del_repos(self, repo_name):
        self.del_res = []
        tag_url = self.repos_url + "/" + repo_name + "/tags"
        # TODO
        tags = self.client.session.get(tag_url).json()
        tags_sort = sorted(tags, key=lambda a: a["created"])
        #print(tags_sort) 
        del_tags = tags_sort[0:len(tags_sort) -30]
        #print(del_tags)
        for tag in del_tags:
            del_repo_tag_url = tag_url + "/" + tag['name']
            print(del_repo_tag_url)
            del_res = self.client.session.delete(del_repo_tag_url)
            self.del_res.append(del_res)

        return self.del_res


if __name__ == "__main__":
   
    harbor_domain = "hub.test.com" 
    password = "xxxxxxx"
    res = ClearHarbor(harbor_domain,password)
    # 循環所有的project id
    for i in res.fetch_pros_id():
        # 獲取所有tag超過30的repos
        repos = res.fetch_del_repos_name(i)
        if repos:
            print(repos)   
            for repo in repos:
                del_repos = res.fetch_del_repos(repo)
                print(del_repos)

清理鏡像釋放空間

依舊為286m，到是什么原因呢，通過官方github查看使用文檔，終於找到了解決辦法，那就是在webui刪除鏡像是軟刪除，並不會釋放空間，在

webui界面刪除后必須停止harbor后再進行硬刪除，也就是空間釋放，垃圾回收，下面是官方文檔原文

Deleting repositories

Repository deletion runs in two steps.

First, delete a repository in Harbor's UI. This is soft deletion. You can delete the entire repository or just a tag of it. After the soft deletion, the repository is no longer managed in Harbor, however, the files of the repository still remain in Harbor's storage.

CAUTION: If both tag A and tag B refer to the same image, after deleting tag A, B will also get deleted. if you enabled content trust, you need to use notary command line tool to delete the tag's signature before you delete an image.

Next, delete the actual files of the repository using the registry's garbage collection(GC). Make sure that no one is pushing images or Harbor is not running at all before you perform a GC. If someone were pushing an image while GC is running, there is a risk that the image's layers will be mistakenly deleted which results in a corrupted image. So before running GC, a preferred approach is to stop Harbor first.

Run the below commands on the host which Harbor is deployed on to preview what files/images will be affected:

$ docker-compose stop

$ docker run -it --name gc --rm --volumes-from registry vmware/registry:2.6.2-photon garbage-collect --dry-run /etc/registry/config.yml

NOTE: The above option "--dry-run" will print the progress without removing any data.

Verify the result of the above test, then use the below commands to perform garbage collection and restart Harbor.

$ docker run -it --name gc --rm --volumes-from registry vmware/registry:2.6.2-photon garbage-collect  /etc/registry/config.yml

$ docker-compose start

For more information about GC, please see GC.

官方已經說的很明白了，第一個run是只打印出來已刪除鏡像，但不進行空間釋放和垃圾回收，執行下面的一個run，成功釋放空間了。