Cleanup and bump glide.yaml and fix unit tests

[zhouzhuojie]

This pr is not urgent at all. Just want to send it anyway incase there's a need to bump glide in the future.

Spent some time clean up glide.yaml and finally got a successful upgrade. Especially the logrus package upgrade, I need to do something like, because glide is case sensitive, and downstream packages require both.

- package: github.com/sirupsen/logrus
- package: github.com/Sirupsen/logrus
  repo: git@github.com:/sirupsen/logrus
  vcs: git

Here's how to run a full clean glide upgrade

glide cc; rm -rf ~/.glide; rm -rf vendor; rm glide.lock; glide --debug up

Also, since some unit tests are broken, it's better to be another pr, but it relies on the glide update, so I've put them here. Really hope that Github can support stack code review.

server/plugins/codeflow/api_auth_test.go
server/plugins/git_sync/git_sync_test.go
server/plugins/kubedeploy/deployments_test.go
server/plugins/kubedeploy/loadbalancers_test.go
server/plugins/route53/route53_test.go
server/plugins/webhooks/github/github_webhooks_test.go

[zhouzhuojie]

this one is needed to get the most recent pflag update, otherwise it breaks cobra.

[zhouzhuojie]

glide is case sensitive

[zhouzhuojie]

Also fixed circleci by reusing our docker image to test golang. Now, everything good to go, including tests in all the _test.go, with complete CI support

version: 2
jobs:
  build:
    working_directory: ~/codeflow
    machine:
    ¦ services:
    ¦   - docker
    steps:
    ¦ - checkout
    ¦ - run: cp ./dashboard/.env ./dashboard/.env.development
    ¦ - run: docker build . -t codeflow-test
    ¦ - run: docker run --rm codeflow-test make test

[zhouzhuojie]

I removed it because the package name should be package kubedeploy, which means it doesn't test the desired package before. Also, the agent is REALLY hard to setup the mock to pass the tests listed here. My 2 cents is that unit tests should be as unit as possible, for example, lots of the logic can be simply tested against the functions defined in kubedeploy/deployment, instead of calling it from agent -- kind of more like an integration test to me.

[jeremyd]

This is an integration test, while we may not run it out of CI it's required when making changes to kubedeploy. All it needs is a test kube cluster to point at.. Usually we point at dev. I'd like to keep this test, is it getting in the way of the unit tests?

[zhouzhuojie]

Oh, maybe I need to setup a test cluster to get it working, do we have some notes/docs about how to set it up?

If it's not CI testable, for example, hard to setup kubedeploy in CI, maybe we can put it as code and run it manually as an integration test.

[kruppel]

@zhouzhuojie Why's this change required?

[zhouzhuojie]

It's not required, I can change it back - probably doesn't matter, before I was just testing and got glide upgrade errors

[jeremyd]

agreed, lets keep the orig repository here.

[kruppel]

This could definitely be sped up considerably.

[zhouzhuojie]

Yup, about 8min :(. Most of the time in the docker build is spent in npm install. I experimented with a lot of options in circleci, (https://circleci.com/gh/checkr/codeflow) really hard to get the libgit2 dependency working correctly, and then the docker build is kind of the last thing I can think to make it fully work. We can have a separate base image (without the npm update/install, pacman install etc.) to speed it up a little bit.

[jeremyd]

I think it's ok to build the docker here for now, the make test also does an npm install but I'm guessing it caches better? The other option might be to somehow wait for codeflow to build the docker image and then simply pull it.

[stale-bot]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.