Node.js is a good technology option on which to develop the interoperability layer core component for the following reasons:
- It is very lightweight
- It has a robust HTTP library
- It has robust support from 3rd party libraries for reading and modifying HTTP requests
- It is highly performant
Libraries to use
- Express - Express is a web application framework for node.js. It provides easy mechanisms to expose web endpoints.
- Connect - Connect provides middleware for node.js that allow HTTP requests and responses can be altered while in 'flight'. The express framework also uses connect middleware and middleware are exposed from the express framework as well.
General Overview
The Express library which includes the Connect middleware provides an easy way to modify HTTP request and responses as they are being processed. Each step that the OpenHIM needs to perform can be written as Connect middleware. Each middleware can handle a different aspect of processing that the OpenHIM need to perform such as authentication, authorization, message persistence and message routing. Developing each of these steps as connect middleware allows them to be easily reused and allows us to add new step for future versions.
More information on Connect middleware can be found here: http://www.senchalabs.org/connect/ and http://stephensugden.com/middleware_guide/
The express library gives us some convenience req and res objects that are designed to be used for web applications but they are equally useful for handling web services.
Logical Architecture
Authentication and Authorization
The are two possible combinations of authentication that the interoperability layer should provide to determine a applications identity:
- HTTP basic authentication
- ATNAs Node Authentication (PKI)
Once identify has been established the IL core component should check if that application has the authority to access the requested service.
The HIM should also provide a single-sign-on (SSO) facility to allow users of the HIE management application to have a single identity that they may used to access these applications. To do this the HIM should also act as an openid provider and provide functions to manage SSO users.
The two main workflows that we wish to enable for authentication and authorization are described in the following workflows:
Authentication
User details for authentication are stored in the MongoDB database is the following format. Either a password or a certificate (in binary form) is stored in this structure depending on whether the user chooses to use PKI or HTTP basic auth to authenticate applications.
{
"applicationID": "Musha_OpenMRS",
"name": "OpenMRS Musha instance",
"roles": [ "OpenMRS_PoC", "PoC" ],
"password-hash": "",
"cert": ""
}
When authentication is set to HTTP basic auth then connect middleware is setup to intercept the request as soon as it enters the HIM as shown above. This middleware will read application details out of the MongoDB store to determine if the application can be authenticated. If the application is rejected an error is returned else the request is considered authenticated and is then passed onto the authorization step.
To help perform the authentication the passport.js module will be use. This provides us with middleware for a number of different authentication schemes. Passport.js
Authorization
The OpenHIM only performs simple authorization based on the path that is being requested. It should be able to restrict access to certain paths to applications with particular roles. The channel description shown in the router section below shows that each path has one or more allowed roles or applications associated with it. The authorization component will check if the authenticated application has the authority to access the current path. It authorized the request will be passed on, else, the request will be denied and a HTTP 401 message will be returned.
Persistence
Each request and response will be persisted so that it can be logged and so that error'd transaction may be re-run. This persistence occurs at two stages. Firstly, once a request is authenticated and authorised and secondly once a response has been received from the external service. All the metadata about a transaction is stored in a single document in MongoDB. The relevant sections are just updated as new information is received. The structure of this information is shown below.
In addition the ability to store orchestration steps exists in the structure. We anticipate exposing a web service to enable mediators to report requests and responses that they make to/receive from external services and have these stored alongside the actual transaction.
We will use the plain mogodb node library rather than using an ORM such as Mongoose. Mongoose helps with schema validation, casting and write object business logic. We don't require these features as the documents that we are storing are relatively simple. Also, Mongoose want you to define data types for everything in the schema. It does not easily allow you to add dynamic elements to your documents. Thus, following the KISS principle we will not make use of any ORM until the need and reason for its use is evident.
{
"_id": "123",
"status": "Processing|Failed|Completed",
"applicationId": "Musha_OpenMRS",
"request": {
"path": "/api/test",
"headers": [
{ "header1": "value1" },
{ "header2": "value2" }
],
"requestParams": [
{ "param1": "value1" },
{ "param2": "value2" }
],
"body": "<HTTP body>",
"method": "POST",
"timestamp": "<ISO 8601>"
},
"response": {
"status": 201,
"body": "<HTTP body>",
"headers": [
{ "header1": "value1" },
{ "header2": "value2" }
],
"timestamp": "<ISO 8601>"
},
"routes": [
{
"name": "<route name>"
// Same structure as above
"request": { ... },
"response": { ... }
}
]
"orchestrations": [
{
"name": "<orchestration name>"
// Same structure as above
"request": { ... },
"response": { ... }
}
]
"properties": [ // optional meta data about a transaction
{ "prop1": "value1" },
{ "prop2": "value2" }
]
}
Router
The router allows request to be forwarded to one or more external services (these could be mediators or an actual HIE component). It does this by allowing the user to configure a number of channels. Each channel matches a certain path and contains a number of routes on which to forward requests. Request may be forwarded to multiple routes however there can only be one primary route. The primary route is a the route whose response is returned back to the service requester making use of the OpenHIM.
Channel will be able to be added, removed and updated dynamically as the application is running.
Channel may be access controlled via the 'allow' field. This field will specify a list of users or groups that are allowed to send requests to that channel. If a channel receives a request from an un-authorised source it will return an error.
A custom router will have to be developed that can route according to these rules. The router can be build using the node.js functions provides to make HTTP request and responses can be relayed using the .pipe() function.
[
{
"name": "Some Registry Channel",
"urlPattern": "test/sample/.+",
"allow": "*",
"routes": [
{
"name": "Some Registry",
"path": "some/other/path" // this is optional if left out original path is used
"host": "localhost",
"port": 8080
}
]
"properties": [ // optional meta data about a channel
{ "prop1": "value1" },
{ "prop2": "value2" }
]
},
{
"name": "Some Registry Channel",
"urlPattern": "test/sample2/.+/test2",
"allow": [ "Alice","Bob", "PoC" ],
"routes": [
{
"name": "Some Registry",
"host": "localhost",
"port": 8080,
"primary": true
},
{
"name": "Logger",
"host": "log-host",
"port": 4789
}
]
"properties": [ // optional meta data about a channel
{ "prop1": "value1" },
{ "prop2": "value2" }
]
}
]
Restful API
The OpenHIM must also expose a restful API that enables it to be configured and to allow access to the transaction that it has logged. This restful API will drive a web application that can allow the OpenHIM to be configured and will allow allow transaction to be viewed and monitored.
The API must supply CRUD access to the following constructs:
- transaction logs
- transaction channels
- users
It should also allow for the following actions:
- single and batch re-processing of transactions
- querying for monitoring statistics
The API is define in using RAML. You can view the details of the API here or you can view the raw RAML code here.