syncsets

Sync Sets

Prototype for an operator like application not depending on Kubernetes CRDs or API, breaking out the SyncSets functionality from OpenShift Hive.

Planned Architecture

• Simulate Kubernetes operator style applications, but without actually being depending on the Kubernetes apiserver, CRDs, or etcd.
• Generate restapi and Go client with go-swagger from Go types. ("code first")
• Rest API backed by Postgres JSONB document storage.
  • Types will support some common metadata such as labels, only now they will actually be indexed.
• Watch will be provided by RabbitMQ or some AMQP bus for pub/sub.
  • Hoping for very minimal bus usage with no logic.
  • Rest API publishes events to queues for each API type.
  • Establishing a watch means subscribing to a queue for an API type, and if you're listening you will receive messages on API events.
• Controllers to be horizontally scalable.
  • Leverage a second type of queue where only one listener can pickup an event.
  • Can run multiple pods and workloads will be distributed to only one as they pull work from the queue.

Development

Install go-swagger

Used to generate API server, go client code, and documentation.

go get -u github.com/go-swagger/go-swagger/cmd/swagger

Presently using "code first" with go-swagger where types/handlers have appropriate godoc annotations to generate swagger.yaml, from which virtually everything under client/ and restapi/ is generated.

Install HTTPPie

Just an easier option than curl. Used in some commands in this README.

Launch RabbitMQ

This project presently aims to use RabbitMQ for pub/sub consumers who wish to watch API events.

Using the RabbitMQ Operator, this process involves some resources I had to patch to work on OpenShift.

kubectl apply -f manifests/rabbitmq-operator/
kubectl apply -f manifests/namespace.yaml
kubectl apply -f manifests/rabbitmq-cluster.yaml
oc adm policy add-scc-to-user rabbitmq-cluster -z rabbitmq-server

Once running you can check in with:

oc rsh rabbitmq-server-0 rabbitmqctl cluster_status

Create a PostgreSQL Database

Several options here:

1. On OpenShift: Use the OpenShift Template, create a new project in the console, select to add a database and choose postgresql. * Crunchy PostgreSQL operator appears much too complicated and possibly broken. * TODO: Try EnterpriseDB PostgreSQL operator.
2. On plain Kubernetes/Kind: Run a postgresql pod: kubectl create -f manifests/postgresql/postgresql.yaml * This won't work on OpenShift as the official Docker images assume root. * TODO: Update the manifest to use the OpenShift image and have one manifest that works on both.
3. Amazon RDS: Choose free tier and public access.

Note your password and ensure you've created a database called syncsets:

Establish a local port forward if running on OpenShift or Kube.

The POSTGRES_PARAMS env var will be used both for goose schema migrations and the api server itself.

You should be able to connect to your local database with:

kubectl port-forward svc/postgresql 5432:5432
export POSTGRES_PARAMS="user=postgres password=helloworld dbname=syncsets sslmode=disable host=localhost"
psql $POSTGRES_PARAMS

Database Schema

Install goose for managing database schema migrations, and create (or update) the schema:

go get -u github.com/pressly/goose/cmd/goose
goose postgres $POSTGRES_PARAMS up

Testing Locally

Ensure you have postgresql properly configured and reachable from localhost per above.

Compile your current code, generate server/client, and run the API locally:

make run

Push some data with httpie:

echo '{"name": "cluster1", "namespace": "foo", "kubeconfig": "foobar"}' | http POST localhost:7070/v1/clusters

Your database should now have an entry in the clusters table.

psql $POSTGRES_PARAMS -c "select * from clusters"

Testing In-Cluster

IMG="quay.io/dgoodwin/syncsets:latest" make docker-push deploy