5.2 KiB
1. JetStream JSON API Design
Date: 2020-04-30 Author: @ripienaar
Status
Partially Implemented
The JSON API has been updated, however the urlencoded replies have not been adopted.
Context
At present, the API encoding consists of mixed text and JSON, we should improve consistency and error handling.
Decision
We anticipate many types of consumer for JetStream including very small devices via MQTT and those written in languages without easy to use JSON support like C.
We will make it easy for light-weight clients by using +OK and -ERR style prefixes for client interactions,
but the server management API will be JSON all the time. The rationale being that such small devices will most
likely have their Consumers and Streams provisioned as part of the platform and often might not care for the
additional context.
In client responses we will use URL Encoding to provide additional context so multiple fields worth of context can be provided. URL Encoding has broad language support.
Admin APIs
Requests
All Admin APIs that today accept nil body should also accept an empty JSON document as request body.
Any API that responds with JSON should also accept JSON, for example to delete a message by sequence we accept
10 as body today, this would need to become {"seq": 10} or similar.
Responses
All responses will be JSON objects, a few examples will describe it best. Any error that happens has to be
communicated within the originally expected message type. Even the case where JetStream is not enabled for
an account, the response has to be a valid data type with the addition of error. When error is present
empty fields may be omitted as long as the response still adheres to the schema.
Successful Stream Info:
{
"type": "io.nats.jetstream.api.v1.stream_info",
"time": "2020-04-23T16:51:18.516363Z",
"config": {
"name": "STREAM",
"subjects": [
"js.in"
],
"retention": "limits",
"max_consumers": -1,
"max_msgs": -1,
"max_bytes": -1,
"max_age": 31536000,
"max_msg_size": -1,
"storage": "file",
"num_replicas": 1
},
"state": {
"messages": 95563,
"bytes": 40104315,
"first_seq": 34,
"last_seq": 95596,
"consumer_count": 1
}
}
Stream Info Error:
{
"type": "io.nats.jetstream.api.v1.stream_info",
"error": {
"description": "unknown stream bob",
"code": 400
}
}
Here we have a minimally correct response with the additional error object.
In the error struct we have description as a short human friendly explanation that should include enough context to
identify what Stream or Consumer acted on and whatever else we feel will help the user while not sharing privileged account
information. code will be HTTP like, 400 human error, 500 server error etc.
Ideally the error response includes a minimally valid body of what was requested but this can be very hard to implement correctly.
Today the list API's just return ["ORDERS"], these will become:
{
"type": "io.nats.jetstream.api.v1.stream_list",
"time": "2020-04-23T16:51:18.516363Z",
"streams": [
"ORDERS"
]
}
With the same error treatment when some error happens.
Client Interactions
To keep clients on small devices viable we will standardise our responses as in the examples below.
+OKeverything is fine no additional context+OK "stream=ORDERS&sequence=10"everything is fine, we have additional data to pass using standard URL encoding-ERRsomething failed, we have no reason-ERR "reason=reason%20for%20failure&stream=STREAM"something failed, we have additional context to provide using standard URL encoding
Additional information will be needed for example when there are overlapping Streams and one of them fail.
Implementation
While implementing this in JetStream the following pattern emerged:
type JSApiResponse struct {
Type string `json:"type"`
Error *ApiError `json:"error,omitempty"`
}
type ApiError struct {
Code int `json:"code"`
Description string `json:"description,omitempty"`
}
type JSApiConsumerCreateResponse struct {
JSApiResponse
*ConsumerInfo
}
This creates error responses without the valid ConsumerInfo fields but this is by far the most workable solution.
Validating this in JSON Schema draft 7 is a bit awkward, not impossible and specifically leads to some hard to parse validation errors, but it works.:
{
"$schema": "http://json-schema.org/draft-07/schema#",
"$id": "https://nats.io/schemas/jetstream/api/v1/consumer_create_response.json",
"description": "A response from the JetStream $JS.API.CONSUMER.CREATE API",
"title": "io.nats.jetstream.api.v1.consumer_create_response",
"type": "object",
"required": ["type"],
"oneOf": [
{
"$ref": "definitions.json#/definitions/consumer_info"
},
{
"$ref": "definitions.json#/definitions/error_response"
}
],
"properties": {
"type": {
"type": "string",
"const": "io.nats.jetstream.api.v1.consumer_create_response"
}
}
}
Consequences
URL Encoding does not carry data types and the response fields will need documenting.