Provisioning GKE cluster for integration tests

[aszecowka]

Description
Implement installation and test of kyma project with GKE cluster using Prow.

AC:
Deliver artifacts (sources in github) necessary to create/re-create the following pipeline:

• activation of the process on pull request comment (1) -> @kfurgol Placeholder job for provisioning GKE cluster for integration tests #71
• checkout sources (2a) -> @strekm
• add "pipeline" script that integrates all steps. Hardcode values that are not available yet -> @Tomasz-Smelcerz-SAP -> PR Create job for integration tests #100
• build Kyma-Installer image (charts included Fix GKE upgrade plans: get helm certs before creating test resources #866 ) (2b, docker-in-docker?) -> @Tomasz-Smelcerz-SAP -> Build Kyma-Installer Docker image #122
• push Kyma-Installer image to gcr (pulls folder?) (2c) -> @Tomasz-Smelcerz-SAP -> Build Kyma-Installer Docker image #122
• reserve IP address (4a) -> @kfurgol Reserve static IP address on GCP #94
• generate domain name like pullxx-random-name.kyma.pro (4b) -> @kfurgol Generate type A DNS record #92
• create DNS A record for IP and domain (4c) -> @kfurgol Generate type A DNS record #92
• generate self signed certificate for the domain (5) -> @kfurgol Generate self signed certificate #97
• provision GKE cluster (6) -> @strekm, PR-1452
• provision and deprovision all external resources (IP, DNS, Cluster) -> @Tomasz-Smelcerz-SAP (Provision and clean-up all external resources #109)
• install tiller (7a) -> @jakkab -> PR Add Kyma installation to GKE integration script #121
• apply kyma-installer deploymenmt (with proper IP and domain) (7b) -> @jakkab -> PR Add Kyma installation to GKE integration script #121
• trigger installation (action=installl) (7c) -> @jakkab -> PR Add Kyma installation to GKE integration script #121
• release IP address and delete DNS record -> @kfurgol Release IP address from GCP #106 Remove DNS record of type A #107
  Stretch: Pipelines steps can be triggered independently (if easy enough to implement)

After provisioning trigger those actions (outside of scope of this task):

• run tests
• deprovision cluster (and IP address) -> @strekm working on deprovision cluster, PR-1452
• set github check state for pull request

Consider implementation as orchestration of small jobs doing one thing at once (e.g. set up DNS, provision cluster, generate TLS) - avoid implementation with one big script.

Reasons
This task is a preparation to move CI from minikube in docker to real cluster on GCE. Prow is the kyma community choice for CI/CD.
Minikube in docker has serious limitations (kubeadm is not supported), therefore other way of testing kyma is required.

[pbochynski]

Set up cluster on GKE flow:

• reserve IP address
• generate domain name like pullxx-random-name.kyma.pro
• create DNS A record for IP and domain
• generate self signed certificate for the domain
• provision GKE cluster
• install tiller
• apply kyma-installer deploymenmt (with proper IP and domain)
• trigger installation (action=installl)

[Tomasz-Smelcerz-SAP]

script that creates a dns entry (record type "A") for given IP address:

AC:

• script accepts IP address as an input
• script generates domain name like: pullXYZ-random-name.kyma.pro (I'd say about 12 random chars should be enough to avoid collisions)
• script adds "A" record in kyma.pro zone
• script verifies the DNS entry works as expected (by resolving host name and verifying it matches IP given as input).
• script waits up to five minutes for correct DNS propagation, then gives up.
• See example: ts-remove-me.kyma.pro

[pbochynski]

Please use build.kyma-project.io domain.

[Tomasz-Smelcerz-SAP]

Integration script first iteration AC:

• provision GKE cluster with hardcoded values (IP address, domain, cert).
• Introduce variables for every configuration value (IP address, domain, cert etc)
• Add several TODOs in the script to replace hardcoded values with actual ones.

[jakkab]

Running testing.sh requires helm command:

/home/prow/go/src/github.com/kyma-project/kyma/installation/scripts/testing.sh: line 203: helm: command not found

Both the bootstrap dockerfile and the actual image need to be updated.

[PK85]

Please use kubernetes version same as minikube which is 1.10. Maybe we should even add that to the installation guide.

[jakkab]

Running tests implemented in #139