when a directory with symlinks is copied, the symlinks are badly created in the target image

[mtsyganov]

Actual behavior
In the multistage build process when a directory with symlinks is copied, the symlinks are badly created in the target image. Too many levels of symbolic links are generated.

Expected behavior
Symlinks are copied from source image into the target image as they are created in the source image

To Reproduce
Steps to reproduce the behavior:

1. create an image with symbolic link. For example /usr/local/psql/bin/postmaster -> postgres (you can check registry.gitlab.com/gitlab-org/build/cng/postgresql:9.6.8)
2. build a multistage image where directory with symlinks is copied to target image COPY --from=pg_image /usr/local/psql/bin/ /usr/bin/
3. run created image and check the copied symlinks. In the source image they are defined as /usr/local/psql/bin/postmaster -> postgres, but in the target image as /usr/bin/postmaster -> /usr/bin/postmaster/postgres. Correct symlink should be in the target image /usr/bin/postmaster -> postgres

Additional Information

• Dockerfile

FROM registry.gitlab.com/gitlab-org/build/cng/postgresql:9.6.8 as pg_image
FROM ubuntu
COPY --from=pg_image /usr/local/psql/bin/ /usr/bin/

[mtsyganov]

@priyawadhwa retested with latest debug image. problem is still reproducable

[priyawadhwa]

@mtsyganov sorry about that, I reopened the issue

[andrewrynhard]

@priyawadhwa Where is this on the list of priorities? This is a major blocker for my particular use case. If low on your list I'd consider picking this up, but would be really happy if a fix is coming soon.

EDIT:

Looks like #459 is in the works! Any chance someone take a look at this?

[priyawadhwa]

@mtsyganov and @andrewrynhard, I just merged #459, does that fix your respective bugs?

[miguelitoq76]

Dear @priyawadhwa we had the same error during the use of kaniko in a custom image,
this was a for us a problem, until we started to create a Dummy build via executor --no_push , to force a filesystem cleanup. See my post in #434

I assume, it would be a good idea, to start the build with a "delete FS", what do you think ?

[priyawadhwa]

@miguelitoq76 sorry, I'm a bit confused -- do you want kaniko to delete the filesystem before starting the image build? Wouldn't that essentially be the same as directly using the kaniko image (which is built from scratch)?

[tequilarista]

Triage - collapsing 'waiting for response' category into 'blocked'

[cvgw]

I am unable to repro this issue; leading me to think that it has since been fixed. If this issue can still be repro'd please share the appropriate steps to do so.

[Kami-no]

Still got this with fresh gcr.io/kaniko-project/executor:debug

GitLab CI + Kaniko

[cvgw]

@Kami-no would you mind sharing an example Dockerfile and the options you are passing to kaniko?

[raijinsetsu]

Is this related to #1036 ?