我回答的前提是:

容器使用标签运行。 相同的标签可以指向不同的图像UUID，因为我们喜欢/觉得合适。 对图像的更新可以提交到新的图像层

方法

首先使用安全补丁更新脚本构建所有容器 为以下内容构建一个自动化流程 使用安全补丁脚本将现有映像运行到新容器 提交对映像的更改为 现有标记->，然后逐个重新启动容器 新版本标签->用新标签替换少数容器->验证->将所有容器移动到新标签

此外，可以升级基本映像/可以定期构建具有完整的新基本映像的容器，这是维护者认为必要的

优势

我们在创建新的安全补丁映像时保留映像的旧版本，因此如果需要，我们可以回滚到以前正在运行的映像 我们保留了docker缓存，从而减少了网络传输(只有被更改的层连接到网络上) 升级过程可以在过渡到产品化之前在阶段中进行验证 这可以是一个受控的过程，因此只有在必要/被认为重要时才可以推送安全补丁。

我回答的前提是:

容器使用标签运行。 相同的标签可以指向不同的图像UUID，因为我们喜欢/觉得合适。 对图像的更新可以提交到新的图像层

方法

首先使用安全补丁更新脚本构建所有容器 为以下内容构建一个自动化流程 使用安全补丁脚本将现有映像运行到新容器 提交对映像的更改为 现有标记->，然后逐个重新启动容器 新版本标签->用新标签替换少数容器->验证->将所有容器移动到新标签

此外，可以升级基本映像/可以定期构建具有完整的新基本映像的容器，这是维护者认为必要的

优势

我们在创建新的安全补丁映像时保留映像的旧版本，因此如果需要，我们可以回滚到以前正在运行的映像 我们保留了docker缓存，从而减少了网络传输(只有被更改的层连接到网络上) 升级过程可以在过渡到产品化之前在阶段中进行验证 这可以是一个受控的过程，因此只有在必要/被认为重要时才可以推送安全补丁。

我不打算讨论您是否希望在生产中进行无人值守的更新(我认为不需要)。我只是把这个留在这里作为参考，以防有人觉得有用。在终端中使用以下命令将所有docker映像更新到最新版本:

# docker images | awk '(NR>1) && ($2!~/none/) {print $1":"$2}' | xargs -L1 docker pull

You can use Watchtower to watch for updates to the image a container is instantiated from and automatically pull the update and restart the container using the updated image. However, that doesn't solve the problem of rebuilding your own custom images when there's a change to the upstream image it's based on. You could view this as a two-part problem: (1) knowing when an upstream image has been updated, and (2) doing the actual image rebuild. (1) can be solved fairly easily, but (2) depends a lot on your local build environment/practices, so it's probably much harder to create a generalized solution for that.

If you're able to use Docker Hub's automated builds, the whole problem can be solved relatively cleanly using the repository links feature, which lets you trigger a rebuild automatically when a linked repository (probably an upstream one) is updated. You can also configure a webhook to notify you when an automated build occurs. If you want an email or SMS notification, you could connect the webhook to IFTTT Maker. I found the IFTTT user interface to be kind of confusing, but you would configure the Docker webhook to post to https://maker.ifttt.com/trigger/`docker_xyz_image_built`/with/key/`your_key`.

如果你需要在本地构建，你至少可以通过在Docker Hub中创建一个连接到你感兴趣的repo的虚拟repo来解决上游映像更新时收到通知的问题。虚拟回购的唯一目的是在重新构建时触发一个webhook(这意味着它的一个链接回购被更新)。如果你能收到这个webhook，你甚至可以用它来触发你这边的重建。

更新:使用Dependabot - https://dependabot.com/docker/

BLUF:找到正确的插入点来监视容器的更改是一个挑战。如果DockerHub能解决这个问题，那就太好了。(存储库链接已经提到过，但注意在DockerHub上设置它们时-“每当在DockerHub上更新基本映像时，都会在此存储库中触发一个构建。只适用于非官方图片。”)

当我自己尝试解决这个问题时，我看到了几个webhook的建议，所以我想详细说明我使用过的几个解决方案。

Use microbadger.com to track changes in a container and use it's notification webhook feature to trigger an action. I set this up with zapier.com (but you can use any customizable webhook service) to create a new issue in my github repository that uses Alpine as a base image. Pros: You can review the changes reported by microbadger in github before taking action. Cons: Microbadger doesn't let you track a specific tag. Looks like it only tracks 'latest'. Track the RSS feed for git commits to an upstream container. ex. https://github.com/gliderlabs/docker-alpine/commits/rootfs/library-3.8/x86_64. I used zapier.com to monitor this feed and to trigger an automatic build of my container in Travis-CI anytime something is committed. This is a little extreme but you can change the trigger to do other things like open an issue in your git repository for manual intervention. Pros: Closer to an automated pipline. The Travis-CI build just checks to see if your container has issues with whatever was committed to the base image repository. It's up to you if your CI service takes any further action. Cons: Tracking the commit feed isn't perfect. Lots of things get committed to the repository that don't affect the build of the base image. Doesn't take in to account any issues with frequency/number of commits and any API throttling.

One of the ways to do it is to drive this through your CI/CD systems. Once your parent image is built, have something that scans your git repos for images using that parent. If found, you'd then send a pull request to bump to new versions of the image. The pull request, if all tests pass, would be merged and you'd have a new child image based on updated parent. An example of a tool that takes this approach can be found here: https://engineering.salesforce.com/open-sourcing-dockerfile-image-update-6400121c1a75 .

如果你不控制你的父映像，就像你依赖于官方ubuntu映像的情况一样，你可以写一些工具来检测父映像标记或校验和(不是一回事，标记是可变的)的变化，并相应地调用子映像构建。