Go Clouddriver

Go Clouddriver is a rewrite of Spinnaker's Clouddriver microservice. It has an observed 95%+ decrease in CPU and memory load for Kubernetes operations.

Go Clouddriver brings many features to the table which allow it to perform better than Clouddriver OSS for production loads:

• it does not rely on kubectl and instead interfaces directly with the Kubernetes API for all operations
• it utilizes an in-memory cache store for Kubernetes API discovery
• it stores Kubernetes providers in a database, fronted by a simple CRUD API
• it removes over-complicated strategies such as Cache All The Stuff, instead making live calls for all operations

Go Clouddriver is not an API complete implementation of Clouddriver OSS and only handles Kubernetes providers and operations. It is meant to be run in tandem with Clouddriver OSS. Visit the wiki for feature support and intallation instructions.

Getting Started

Testing

Run from the root directory

make tools test

Building

Run from the root directory

make build

Running Locally

1. Go Clouddriver generates its access tokens using Arcade as a sidecar, so a working instance of Arcade will need to be running locally in order for Go Clouddriver to talk to Kubernetes clusters.

2. Export the Arcade API key (the same one you set up in step 1).

export ARCADE_API_KEY=test

3. Run Go Clouddriver.

make run

4. Create your first Kubernetes provider! Go Clouddriver runs on port 7002, so you'll make a POST to localhost:7002/v1/kubernetes/providers.

curl -XPOST localhost:7002/v1/kubernetes/providers -d '{
  "name": "test-provider",
  "host": "https://test-host",
  "caData": "test",
  "permissions": {
    "read": [
      "test-group"
    ],
    "write": [
      "test-group"
    ]
  }
}' | jq

And you should see the response

{
  "name": "test-provider",
  "host": "https://test-host",
  "caData": "test",
  "permissions": {
    "read": [
      "test-group"
    ],
    "write": [
      "test-group"
    ]
  }
}

Running the command again will return a 409 Conflict unless you change the name of the provider.

5. List your providers by calling the /credentials endpoint.

curl localhost:7002/credentials | jq

Configuration

Environment Variable	Description	Notes	Default Value
ARCADE_API_KEY	Needed to talk to Arcade.	Required for most operations.
ARTIFACTS_CREDENTIALS_CONFIG_DIR	Sets the directory for artifacts configuration.	Optional. Leave unset to use OSS Clouddriver's Artifacts API.
KUBERNETES_USE_DISK_CACHE	Stores Kubernetes API discovery on disk instead of in-memory.		false
DB_HOST	Used to connect to MySQL database.	If not set will default to local SQLite database.
DB_NAME	Used to connect to MySQL database.	If not set will default to local SQLite database.
DB_PASS	Used to connect to MySQL database.	If not set will default to local SQLite database.
DB_USER	Used to connect to MySQL database.	If not set will default to local SQLite database.
VERBOSE_REQUEST_LOGGING	Logs all incoming request information.	Should only be used in non-production for testing.	false

MySQL Indexes and Cleanup

Go Clouddriver stores all deployed resource requests in its kubernetes_resources table, which needs to be cleaned up periodically. It also requires a few indexes to work efficiently and properly for continued deployments over long periods of time.

Indexes

First, an index to help the Applications API remain efficient.

CREATE INDEX kind_account_name_kind_name_spinnaker_app_idx ON kubernetes_resources(account_name, kind, name, spinnaker_app);

Next, an index to assist in pulling "cluster" kinds from the kubernetes_resources table.

CREATE INDEX kind_idx ON kubernetes_resources(kind);

Next, an index to assist the Task API.

CREATE INDEX task_id_idx ON kubernetes_resources(task_id);

Finally, a couple of indexes on the provider read/write permissions tables to help the Credentials API and any queries to select providers from the database.

CREATE INDEX account_name_idx ON provider_read_permissions(account_name);
CREATE INDEX account_name_idx ON provider_write_permissions(account_name);

SQL Cleanup

The kubernetes_resources tables is incredibly important for storing the most-recent state of a cluster for the Kubernetes kinds we care about.

It is recommended to run the following cleanup statements daily to maintain a healthy state of onboarded Kubernetes clusters.

DELETE FROM kubernetes_resources where cluster = '' and timestamp < (NOW() - INTERVAL 6 HOUR);

DELETE t1 FROM kubernetes_resources t1
INNER JOIN kubernetes_resources t2
WHERE
    t1.id < t2.id AND
    t1.account_name = t2.account_name AND
    t1.api_group = t2.api_group AND
    t1.name = t2.name AND
    t1.namespace = t2.namespace AND
    t1.resource = t2.resource AND
    t1.version = t2.version AND
    t1.kind = t2.kind AND
    t1.cluster = t2.cluster AND
    t1.timestamp < (NOW() - INTERVAL 6 HOUR);