julz / cube Goto Github PK

Acceptance:

1. Deploy eirini following instructions in repo
2. cf push, cf scale -i 3
3. cf apps, see that it correctly reports number of reported and requested instances (compare with kubectl get deployments)
4. Manually kill some pods using kubectl, see that this is reflected in cf apps

The cloud controller sends stop staging request to cube in case staging needs to be stopped. Therefor, the stopStaging handler needs to be implemented.

Here is some information about Kube logging: Kube Logging

We need this to be able to re-use a single k8s cluster for two different pipelines.

Pass the name as a parameter to sync process when it is started.

This is a catch-all story to cover the first MVP goal of CF Smoke Tests 'apps' test suite passing

Prerequisites:

• We will need to figure out how to deal with the lack of loggregator APIs at this point, possibly by forking the CATs to allow the logging to be done directly via kubectl get logs or similar.

Acceptance:

1. Follow instructions on smoke tests repo pointed at a CF deployed with eirini (following instructions on eirini repo for this), they pass, weeee!

Check with Adam (Loggregator) if they have an early piece we could use.

Other options? CF plugin.

What does CATs need to pass WRT logs. (Run them all and report on what fails why)

e.g. https://github.com/alecthomas/gometalinter

See previous spike and try out option 3 from story #30

bulk loop works for a demo, but we need to implement immediate endpoints so we don't have to wait for the loop to run to kick off a deployment and so cf apps comes back with the right things

should be deployable to CF, this should also allow wiring sync and registry together since we can actually have the networking work between the registry and the syncer (so we can remove the hardcoded busybox).

Acceptance:

1. Deploy eirini following instructions on repo
2. cf push a hello world app, see that I can curl it on the route the cli tells me about
3. cf push a hello world app and map-route a different route, see that I can curl that, too

The download of app-bits requires auth. Therefor, the admin user and password are provided as environment variables to the staging jobs on kubernetes. You can lookup the env vars in kubectl describe job, which is very insecure.

The recipe should retrieve credentials from a vault or the like.

Test and investigate the best way for CAPI to inteface with Eirini and to Diego in the long run.

Output from the story is a design on what would needed to be done in Capi (PR) and what needs to be done Eirini to work

Also add config to provide the namespace to deploy to.

The recipe executes the binary of packs/build. A nicer solution would be to import packs and use it as a library.

@julz @andrew-edgar This is probably not a complete story yet, but we wanted to capture the requirement that came up as part of the discussion in this PR.

TaskDesirer needs to able to cancel (and potentially delete) a task.

In Diego these are documented as DeleteTask and CancelTask.

AC:

• Deploy CF with Eirini
• cf push an app that has at least one task (see docs)
• Cancel the running task
• Check that the task is no longer running usingcf tasks my-app

/CC @JulzDiverse

Acceptance could be:
Create the app with Diego.
Restart with Eirini mode
cc will see app is not running in K8S
cc will start the LRP in Eirini
Expect the app to appear in K8S

Right now when a staging succeeded successfully, we leave the job description behind in k8s (because in a generic k8s scenario you might want to start that job again). But in Eirini we never want to run the same staging job again. Instead of leaving cruft behind, we should clean up after ourselves.

Implementation notes:

• The staging completion callback needs to be enhanced to perform the delete.
• Use the DeleteTask function to perform this action

AC:

• Stage an app
• Check k8s that the staging job description is gone

Create a new 1-click bosh-lite pipeline

The recipe uses "unzip" with exec.Cmd to unzip app-bits/app-packages. This should be replaced with a go implementation of a zipper.

Extracted from #17

Acceptance:

1. Deploy and enable eirini following instructions on repo
2. cf push --no-start
3. cf start
4. See that the app is immediately started (i.e. deployments created and set to non-zero instance count), where immediately means doesn't wait for convergence loop (this may require a bit of white-box accepting, depending on how quick convergence is)