Data-driven API design

We live in a world of APIs. LeaseWeb is an infrastructure provider that has taken on an API-first approach. This means that we first build an API for our products. Only after the API is finished we build the UI for the product and we do this on top of our own API. This “eat your own dog food” principle ensures a high quality API in which bugs are found before the customer encounters them.

Our “bare metal” (formerly: dedicated server) team has an API for switch port management. Opening a switch port may (dependent on the switch brand) take up several seconds. This is because (for instance) a SSH connection has to be made to the switch, then an XML command and answer has to be sent back and forth. Also, some switches may wait a few seconds to respond with a status, as they are trying to identify the link status, after the port is opened.

But what if a customer requests a “faster” API? We can’t make opening or closing a switch port go faster. What to do? If any API call would only modify a single record in the database, then the execution time would be predictable (and always low). To achieve this there are two API designs possible: “asynchronous” and “data-driven”. In this post we will explore both and show you why a data-driven API is a better choice for consistent APIs.

Synchronous API design

This is the naive implementation. It allows you to Open/Enable the switch port attached to the server with a synchronous call:

POST /bareMetals/{bareMetalId}/switchPort/open

Parameters:

bareMetalId (integer) - The id of the bare metal server

Example request:

POST /v1/bareMetals/1234/switchPort/open HTTP/1.1
Host: api.leaseweb.com
Accept: application/json
X-Lsw-Auth: 4e7d4f2d-e683-4192-a113-61dad6ac9a15

Example response:

HTTP/1.1 200 OK
Content-Type: text/html

{
	"switchPort": {
		"status": "opened",
		"switchNode": 1,
		"serverId": 1234,
		"serverName": "My Server"
	}
}

Note that this call may take several seconds to complete and does not follow the RESTful resource naming scheme.

Asynchronous API design

If the call is asynchronous then the request is the same:

POST /bareMetals/{bareMetalId}/switchPort/open

Parameters:

bareMetalId (integer) - The id of the bare metal server

Example request:

POST /v1/bareMetals/1234/switchPort/open HTTP/1.1
Host: api.leaseweb.com
Accept: application/json
X-Lsw-Auth: 4e7d4f2d-e683-4192-a113-61dad6ac9a15

Example response:

HTTP/1.1 200 OK
Content-Type: text/html

{
	"jobId": "f5825d99-62ed-433b-930b-1b9561fe4a6e",
	"createdAt": "2015-01-14T15:09:10+01:00"
}

The response is a job identifier that then has to be polled for status, like this:

GET /bareMetals/{bareMetalId}/jobs/{jobId}

Parameters:

bareMetalId (integer) - The id of the bare metal server
jobId (string) - The id of a job

Example request:

GET /v1/bareMetals/1234/jobs/f5825d99-62ed-433b-930b-1b9561fe4a6e HTTP/1.1
Host: api.leaseweb.com
Accept: application/json
X-Lsw-Auth: 4e7d4f2d-e683-4192-a113-61dad6ac9a15

Example response:

HTTP/1.1 200 OK
Content-Type: text/html

{
	"id": "f5825d99-62ed-433b-930b-1b9561fe4a6e",
	"status": 3,
	"statusDescription": "Failed",
	"createdAt": "2015-01-14T15:09:10+01:00",
	"updatedAt": "2015-01-14T15:09:12+01:00"
}

All reponses can be sent really fast as they are simple database lookups. It is important to realize that you need some sort of job manager in the back that does actually execute the jobs and updates the job status accordingly. Note that the commands do not follow the RESTful resource naming scheme.

Data-driven API design

When following data-driven API design you are simply modeling the database tables to facilitate the functionality. It allows you to have fast replies and also follows the RESTful resource naming scheme. You may get the status of the switch port attached to the server with a call like this:

GET /bareMetals/{bareMetalId}/switchPorts

Parameters:

bareMetalId (integer) - The id of the bare metal server

Example request:

GET /v1/bareMetals/1234/switchPorts HTTP/1.1
Host: api.leaseweb.com
Accept: application/json
X-Lsw-Auth: 4e7d4f2d-e683-4192-a113-61dad6ac9a15

Example response:

HTTP/1.1 200 OK
Content-Type: text/html

{
	"switchPorts": [
		{
			"status": "opened",
			"switchNode": 1,
			"serverId": 1234,
			"serverName": "My Server"
		}
	]
}

When you are opening the switch port you are actually inserting a record in the switchPortOpenRequests table with the standard POST as prescribed by the RESTful resource naming scheme, like this:

POST /bareMetals/{bareMetalId}/switchPortOpenRequests

Parameters:

bareMetalId (integer) - The id of the bare metal server

Example request:

POST /v1/bareMetals/1234/switchPortOpenRequests HTTP/1.1
Host: api.leaseweb.com
Accept: application/json
X-Lsw-Auth: 4e7d4f2d-e683-4192-a113-61dad6ac9a15

Example response:

HTTP/1.1 200 OK
Content-Type: text/html

{
	"id": "f5825d99-62ed-433b-930b-1b9561fe4a6e",
	"status": 1,
	"statusDescription": "Pending",
	"createdAt": "2015-01-14T15:09:10+01:00",
	"updatedAt": "2015-01-14T15:09:10+01:00"
}

Just like with the asynchronous model you can poll for status. Note that there is no job identifier, just a normal identifier, as the request is a normal resource that follows the RESTful resource naming scheme. You do still need to have a background process that watches the switchPortOpenRequests table to execute these requests when they are in “pending” state. This background process is responsible for updating both the¬†switchPortOpenRequests and the switchPorts table with the correct status information.

GET /bareMetals/{bareMetalId}/switchPortOpenRequests/{id}

Parameters:

bareMetalId (integer) - The id of the bare metal server
id (string) - The id of the request

Example request:

GET /v1/bareMetals/1234/switchPortOpenRequests/f5825d99-62ed-433b-930b-1b9561fe4a6e HTTP/1.1
Host: api.leaseweb.com
Accept: application/json
X-Lsw-Auth: 4e7d4f2d-e683-4192-a113-61dad6ac9a15

Example response:

HTTP/1.1 200 OK
Content-Type: text/html

{
	"id": "f5825d99-62ed-433b-930b-1b9561fe4a6e",
	"status": 3,
	"statusDescription": "Failed",
	"createdAt": "2015-01-14T15:09:10+01:00",
	"updatedAt": "2015-01-14T15:09:12+01:00"
}

As you can see this last way of modeling does not only reduce every API call to be a database lookup, it also allows you to follow the (predictable) RESTful resource naming scheme.

Generating a data-driven API

The advantage of following an API naming scheme is that the API becomes predictable and can be generated. The PHP-CRUD-API project facilitates creating a full-featured API by uploading a single PHP file. This API will follow the resource naming scheme for your MySQL, PostgreSQL or MS SQL Server tables. With a data-driven API design this should not limit you to create any API functionality. Try it!

Leave a Reply

Your email address will not be published. Required fields are marked *