Status

Current state: Under Discussion, VotingAccepted

Discussion threads: here (the original), here (a new one as PonyMail didn't manage to associate the new messages with the old one)

...

As JSON null values aren't allowed in the connector configs, null will serve as a tombstore value used for deleting existing fields from configs.

Responses follow the model of the configuration PUT endpoint. If the patch was successfully applied, the response code is 200 and the body is a JSON object with the updated connector information (name, type, config, and tasks), for example:

{
    "name": "my-connector",
    "config":
    {
        "name": "my-connector",
        "sample_config_2": "test_config_2",
        "sample_config": "test_config_new"
    },
    "tasks":
    [
        {
            "connector": "my-connector",
            "task": 0
        },
        {
            "connector": "my-connector",
            "task": 1
        }
    ],
    "type": "sink"
}

In case of errors, the response code matches the error type (e.g. 400 in case of a config validation error; 404 if the connector is not found; 500 in case of other server-side errors) and the body is a JSON object with the error details:

{
    "error_code": 400,
    "message": "Connector configuration is invalid and contains the following 1 error(s):\n...\nYou can also find the above list of errors at the endpoint `/{connectorType}/config/validate`"
}

Proposed Changes

We will add PATCH method definition into ConnectorsResource, which will call the actual logic in Herder and its child classes StandaloneHerder and DistributedHerder. Due to the way how the herders are implemented now, it's difficult to guarantee 100% race-free partial updates. Besides, other similar race scenarios exist, so we keep out of scope of this KIP. The implementation will attempt to keep the race window as narrow as possible.

...