fix(docker release): avoid duplicate occupation in docker image

[yanjing19989]

Using COPY --chmod=755 instead of chmod +x prevents Docker from creating duplicate layers for the binary file, which will resolve the abnormal increase in image space usage since v3.42.0.

Although the issue only occurs in CI, I also update the local Dockerfile for consistency.

Close: #8388

[Mmx233]

Are there any images built by CI in this branch? That layer should be tiny, I'm currios about how much exactly we can save. The root cause of image size increase, I guess, is that the binary size of alist increased.

[Mmx233]

Your are right, the size almost doubled because of chmod, we can save about 28MB.

[yanjing19989]

This change will reduce the size of CI-built Docker images by the equivalent of the AList binary file's size. Particularly for Linux (musl) like Alpine, this will cut down image size by over 80MB.

Additionally, musl-based systems are often more space-sensitive. During testing, I noticed that the AList binary on Linux (musl) is significantly larger than on other platforms (80MB vs. 30MB). I'm unsure whether this is due to limitations in the platform's compilation toolchain.

[Copilot]

Copilot wasn't able to review any files in this pull request.

Files not reviewed (2)

• Dockerfile: Language not supported
• Dockerfile.ci: Language not supported

[Mmx233]

I haven't find out root cause yet. This should not related to platform, I have some other images use same extra layer with alpine but size didn't increase. Currently I guess it may related to whether there is x permission on binary before copy and whether docker buildkit is enabled while building.

Whatever, the fix in this PR is fantastic. I haven't try but it should able to fix the problem.

[yanjing19989]

Your guess is correct. The Alist copied during the CI build lacks execute permissions, so running chmod +x would cause duplicate occupation. However, in the local multi-stage build, there is no issue with missing execute permissions, and running chmod +x does not produce any changes.

The additional Linux (musl) size issue mentioned in the previous comment is related to the executable file itself. If this can be optimized, the Docker image size could be significantly reduced once again.

Taking linux_amd64 as an example:

• Current v3.44.0 Docker image size: 194MB
• Docker image size after adopting this commit / local build: 106MB
• If the Linux (musl) executable size were similar to other platforms, the Docker image size could be as low as: ~55MB

[Mmx233]

Currently, musl build in CI for docker is important and it's hard to migrate to docker QEMU cross build or go toolchain cross build.

Firstly alist needs sqlite, which must enable CGO during build. So we can't use go toolchain to build directly. Secondly, alist used to be using QEMU build with docker multistage build, result in each action will run for more than an hour because of insufficiant memory. Now, after migrate to musl build, we have added more architactures to docker release work flow, so if we use QEMU again, build time will be much more higher than an hour.

As an anther workaround, it's possible to set buildkit parallel build limit to 1 to avoid using too much memory. However, using QEMU makes it almost impossible to reuse go build caches across the same workflow or multiple actions, and its emulation efficiency is significantly lower than that of the toolchain with musl, which also results in a substantial increase in CI time. So currently it's hard to migrate to other build method.

[Mmx233]

Hi @xhofe, this PR LGTM and is ready to be merged. We're currently discussing further optimizations based on this PR.

Also, there is another pending PR (#7812) with an little code logic imporve — we would appreciate it if you could take a look and consider merging it as well.

[xhofe]

Nice job 👍👍👍