Useful links

https://github.com/Azure/azure-resource-manager-rpc

https://github.com/microsoft/api-guidelines

https://docs.microsoft.com/en-gb/rest/api/cosmos-db

https://docs.microsoft.com/en-gb/rest/api/storageservices/queue-service-rest-api

https://github.com/jim-minter/go-cosmosdb

Getting started

• Copy env.example to env, edit the values and source it (. ./env)

• Deploy a CosmosDB SQL database and a v2 storage account to a resource group: az group create -g "$RP_RESOURCEGROUP" -l "$LOCATION" az group deployment create -g "$RP_RESOURCEGROUP" --mode complete --template-file deploy/rp.json --parameters "location=$LOCATION" "storageAccountName=$STORAGE_ACCOUNT" "databaseAccountName=$COSMOSDB_ACCOUNT"

• go run ./cmd/rp

Useful commands

export CLUSTER=cluster

curl -X PUT "localhost:8080/subscriptions/$AZURE_SUBSCRIPTION_ID/resourceGroups/$CLUSTER/providers/RedHat.OpenShift/OpenShiftClusters/$CLUSTER?api-version=2019-12-31-preview" -H 'Content-Type: application/json' -d '{"location":"'"$LOCATION"'", "properties": {"pullSecret": "'"$(base64 -w0 <<<$PULL_SECRET)"'"}}'

curl "localhost:8080/subscriptions/$AZURE_SUBSCRIPTION_ID/resourceGroups/$CLUSTER/providers/RedHat.OpenShift/OpenShiftClusters/$CLUSTER?api-version=2019-12-31-preview"

curl "localhost:8080/subscriptions/$AZURE_SUBSCRIPTION_ID/resourceGroups/$CLUSTER/providers/RedHat.OpenShift/OpenShiftClusters/$CLUSTER/credentials?api-version=2019-12-31-preview"

curl -X DELETE "localhost:8080/subscriptions/$AZURE_SUBSCRIPTION_ID/resourceGroups/$CLUSTER/providers/RedHat.OpenShift/OpenShiftClusters/$CLUSTER?api-version=2019-12-31-preview"

Basic architecture

• pkg/frontend is intended to become a spec-compliant RP web server. It is backed by CosmosDB. Incoming PUT/DELETE requests are written to the database with an Updating/Deleting provisioningState and unqueued flag set.

• pkg/queue/forwarder is a worker which spots unqueued documents, writes their IDs into a queue and unsets the flag. It is intended that a document ID may be accidently written to the queue more than once without ill effect.

• pkg/backend reads IDs off the queue and asynchronously updates desired state, finally updating the database document with a terminal provisioningState (Succeeded/Failed). The backend sends a heartbeat to the queue - if this fails, the ID will be picked up by a different worker.

• As CosmosDB does not support document patch, care is taken to correctly pass through any fields in the internal model which the reader is unaware of (see github.com/ugorji/go/codec.MissingFielder). This is intended to help in upgrade cases and (in the future) with multiple microservices reading from the database in parallel.

• Care is taken to correctly use optimistic concurrency to avoid document corruption through concurrent writes (see RetryOnPreconditionFailed).

• The pkg/api architecture differs somewhat from github.com/openshift/openshift-azure: the intention is to fix the broken merge semantics and try pushing validation into the versioned APIs to improve error reporting.

• Everything is intended to be crash/restart/upgrade-safe...