[v1.7.x] Fix CI error

[ciyongch]

Description

Fix the following CI error:

[2020-05-28T17:49:56.861Z] + ln -s /usr/bin/ninja-build /usr/bin/ninja

[2020-05-28T17:49:56.861Z] ln: failed to create symbolic link '/usr/bin/ninja': File exists

[2020-05-28T17:49:57.421Z] The command '/bin/sh -c /work/centos7_core.sh' returned a non-zero code: 1

@TaoLv @leezu @pengzhao-intel @ChaiBapchya

[mxnet-bot]

Hey @ciyongch , Thanks for submitting the PR
All tests are already queued to run once. If tests fail, you can trigger one or more tests again with the following commands:

• To trigger all jobs: @mxnet-bot run ci [all]
• To trigger specific jobs: @mxnet-bot run ci [job1, job2]

---

CI supported jobs: [windows-cpu, unix-cpu, sanity, centos-gpu, website, clang, unix-gpu, centos-cpu, edge, miscellaneous, windows-gpu]

---

Note:
Only following 3 categories can trigger CI :PR Author, MXNet Committer, Jenkins Admin.
All CI tests must pass before the PR can be merged.

[nickguletskii]

Hi @ciyongch, it seems that the reason for the CI failures in the v1.7.x branch are the same as in v1.x: the ninja package in CentOS now ships with the required symlink, which makes it unnecessary to create it manually. I fixed this issue in v1.x by removing the build step that creates the symlink: #18437

[ciyongch]

@nickguletskii Thanks for point it out for the fix:), would you mind to backport that PR into v1.7.x branch as well?
Regarding the rest of symlink change in this PR, I suggest to use ln -sf instead of ln -s, what do you think?

[nickguletskii]

@nickguletskii Thanks for point it out for the fix:), would you mind to backport that PR into v1.7.x branch as well?
Regarding the rest of symlink change in this PR, I suggest to use ln -sf instead of ln -s, what do you think?

I've created a backport here: #18456 .
Concerning the other symlink changes: let's see if the CI passes without the ninja link first. Personally, I don't think overwriting symlinks managed by the system's package manager is a very good idea, even in containers, so unless the CI jobs fail without ln -sf, I don't think these changes are necessary.

[ciyongch]

Thanks for your help and comments @nickguletskii , let's just keep the rest as is if the CI passed. My initial thoughts was to handle all the potential of this kind of issue, as ln -sf is used widely in the current build script.

[leezu]

#18456