Optimize dockerfile

[devholic]

설명 (Description)

• Optimize dockerfile
  • Reduced image size by 94% (4.12GB -> 234MB)
  • Utilize cache when rebuilding resources
  • Change maintainer to Kakao Corp. <oss@kakaocorp.com>
  • Specify version to base image
  • Large model can be used by passing --rsc-dir=/khaiii/share/large

Also, I think it'd be better if pre-built image is provided through public registry (e.g. Dockerhub/kakaocorp) for faster use, for example:

I requsted CLA copy, but didn't received yet

체크 리스트 (Checklist)

• [x] master 브랜치가 아니라 develop 브랜치에 머지하도록 pull request를 작성 중이신가요? (Did you merge into develop branch not master?)
• [x] build/test/khaiii 프로그램을 실행하여 테스트가 성공했나요? (Did all tests are passed when you ran as build/test/khaiii)
• [ ] PyLint 툴을 실행하여 발생한 에러를 모두 수정하셨나요? (Did you fix all errors after running PyLint?)
• [ ] CppLint 툴을 실행하여 발생한 에러를 모두 수정하셨나요? (Did you fix all errors after running CppLint?)

[devholic]

@krikit

Because I cannot determine which Dockerfile is good for khaiii at this point. Sorry I have not enough knowledge for docker. ;)

I think this can be resolved with providing --target=builder option. If we build image with this option, there will be two differences with original image:

• Base image difference
  • Since original dockerfile's base tag is latest, we can't guarantee users about exact version of base image. (Because latest image can be updated anytime)
• It builds not only resource, but also large_resource 🙂

So I can say new dockerfile is identical with original dockerfile.

How about moving your Dockerfile into sub directory such as pytorch-1.1-optimized, instead of replacing the original Dockerfile?

I think providing multiple pytorch version would be better since there are performance issues in latest pytorch version (https://github.com/kakao/khaiii/issues/72), but I think optimized suffix can be removed since images are almost identical.

[violetblue]

@cla-bot check

[cla-bot[bot]]

The cla-bot has been summoned, and re-checked this pull request!