Fix frequently failing test suite for `unit tests` job

The unit tests job generates frequently fails without relation to changes in code. This is disturbing and makes the whole suite false-positive. We should fix this.

/cc @ayufan @nick.thomas

There are two major sources of failures:

• executors/docker/machine race condition
• executors/shell failures (timeout-related?)

Sometimes we have connectivity failures which break the tests in odd ways - inability to clone the gitlab-test repo is one such example. This is made worse by GitLab.com infrastructure problems.

Mentioned in merge request !370 (merged)

I've created !370 (merged) to fix this problem.

• executors/docker/machine race condition

A proof of concept of a fix for this is in https://gitlab.com/gitlab-org/gitlab-ci-multi-runner/merge_requests/370/diffs#diff-3

• executors/shell failures (timeout-related?)

Sometimes we have connectivity failures which break the tests in odd ways - inability to clone the gitlab-test repo is one such example. This is made worse by GitLab.com infrastructure problems.

Those two are basically the same. Tests in executors/shell are using external gitlab-test repository and timeouts are generated (I'm sure in 99%) by networking problems and/or GitLab.com performance.

I've changed common/support.go to add an option to use locally generated git repositories instead of external one on GitLab.com. Currently it works only for tests in shell executor and for common/build_test.go. I'm still thinking on how to expose such repository for tests of docker/kubernetes/virtualbox/parallels executors.

Added Doing label

Status changed to closed by commit 056473be6d9f11c7a5acc5bd9e42cb7319003726