feat(di): allow controllers to be scoped to environment

[abenerd]

Information

Type	Breaking change
Feature	No

---

Allow controllers to be scoped to an environment.

import {Get} from "@tsed/schema";
import {Controller} from "@tsed/di";

@Controller({
  path: "local",
  environments: ["development"]
})
export class LocalCtrl {
  @Get()
  index(): string {
    return [];
  }
}

Todos

• Tests
• Coverage
• Example
• Documentation

[github-actions]

Benchmarks

• Machine: linux x64 | 4 vCPUs | 15.6GB Mem
• Node: v16.20.2
• Run: Tue Dec 12 2023 09:24:19 GMT+0000 (Coordinated Universal Time)
• Method: autocannon -c 100 -d 10 -p 10 localhost:3000 (two rounds; one to warm-up, one to measure)

	Version	Router	Requests/s	Latency	Throughput/Mb
fastify	3.29.4	✓	54844.8	17.74	9.78
nest-fastify	8.4.3	✓	47352.7	20.62	8.44
koa	2.13.4	✗	38380.8	25.55	6.84
fastify-injector	3.29.4	✓	27820.4	35.38	6.34
express	4.18.1	✓	12025.8	82.45	2.14
tsed-koa	7.52.0	✓	11960.4	82.89	10.45
fastify-big-json	3.29.4	✓	11090.5	89.41	127.61
nest	8.4.3	✓	10449.8	94.88	2.51
express-injector	4.18.1	✗	10175.2	97.38	2.32
tsed-express	7.52.0	✓	9460.2	104.91	1.73
express-morgan	4.18.1	✗	6802.4	145.70	1.21

Explanation

The benchmark shows a performance difference between the frameworks. We note that Ts.ED is often last. In fact, Ts.ED uses features useful to a production application which reduce its performance.

For example, Ts.ED initializes a sandbox (async_hook) for each request in order to work in an isolated context if necessary.
It also initializes the elements necessary for monitoring requests in a log manager.

All this at a necessary cost that reflects the reality of a production application ;)

[github-actions]

Benchmarks

• Machine: linux x64 | 4 vCPUs | 15.6GB Mem
• Node: v16.20.2
• Run: Tue Dec 12 2023 09:35:29 GMT+0000 (Coordinated Universal Time)
• Method: autocannon -c 100 -d 10 -p 10 localhost:3000 (two rounds; one to warm-up, one to measure)

	Version	Router	Requests/s	Latency	Throughput/Mb
fastify	3.29.4	✓	53996.8	18.03	9.63
nest-fastify	8.4.3	✓	48112.0	20.29	8.58
koa	2.13.4	✗	38430.6	25.50	6.85
fastify-injector	3.29.4	✓	28270.4	34.84	6.44
express	4.18.1	✓	12053.6	82.25	2.15
fastify-big-json	3.29.4	✓	11616.4	85.28	133.66
tsed-koa	7.52.0	✓	11601.6	85.32	10.14
nest	8.4.3	✓	10865.1	91.18	2.61
express-injector	4.18.1	✗	10164.7	97.58	2.32
tsed-express	7.52.0	✓	9338.7	106.03	1.71
express-morgan	4.18.1	✗	7005.6	141.44	1.25

Explanation

The benchmark shows a performance difference between the frameworks. We note that Ts.ED is often last. In fact, Ts.ED uses features useful to a production application which reduce its performance.

For example, Ts.ED initializes a sandbox (async_hook) for each request in order to work in an isolated context if necessary.
It also initializes the elements necessary for monitoring requests in a log manager.

All this at a necessary cost that reflects the reality of a production application ;)

[github-actions]

Benchmarks

• Machine: linux x64 | 4 vCPUs | 15.6GB Mem
• Node: v16.20.2
• Run: Tue Dec 12 2023 13:47:26 GMT+0000 (Coordinated Universal Time)
• Method: autocannon -c 100 -d 10 -p 10 localhost:3000 (two rounds; one to warm-up, one to measure)

	Version	Router	Requests/s	Latency	Throughput/Mb
fastify	3.29.4	✓	57040.0	17.03	10.17
nest-fastify	8.4.3	✓	49241.6	19.85	8.78
koa	2.13.4	✗	40327.3	24.29	7.19
fastify-injector	3.29.4	✓	28645.1	34.37	6.53
express	4.18.1	✓	12398.9	79.97	2.21
tsed-koa	7.52.0	✓	11813.5	83.87	10.32
fastify-big-json	3.29.4	✓	11630.2	85.20	133.80
nest	8.4.3	✓	10965.5	90.45	2.64
express-injector	4.18.1	✗	10341.6	95.90	2.36
tsed-express	7.52.0	✓	9544.5	103.86	1.75
express-morgan	4.18.1	✗	7485.5	132.33	1.33

Explanation

The benchmark shows a performance difference between the frameworks. We note that Ts.ED is often last. In fact, Ts.ED uses features useful to a production application which reduce its performance.

For example, Ts.ED initializes a sandbox (async_hook) for each request in order to work in an isolated context if necessary.
It also initializes the elements necessary for monitoring requests in a log manager.

All this at a necessary cost that reflects the reality of a production application ;)

[Romakita]

Hello @abenerd
it's a really good idea ! But your option is limited to the controller decorator and it would be great to have it for any kind of provider :). I suggest you to perform the filter implementation on DI level directly.

So to do that:

• Edit the Provider model and add the environments property: https://github.com/tsedio/tsed/blob/production/packages/di/src/common/domain/Provider.ts
• Edit ProviderOpts and add the environments property: https://github.com/tsedio/tsed/blob/production/packages/di/src/common/interfaces/ProviderOpts.ts
• Filter mounted controller: https://github.com/tsedio/tsed/blob/production/packages/di/src/common/utils/resolveControllers.ts#L50

export function resolveControllers(settings: Partial<TsED.Configuration>): TokenRoute[] {
  const providers = lookupProperties.flatMap((property) => getTokens(settings[property]));
  
   /// get env from settings
  const env = settings.env

  /// filter this list
  return [...resolveRecursively(providers), ...providers].filter((provider) => !!provider.route) as any[];
}

• Update createContainer implementation to take a env parameter to filter providers:
  • https://github.com/tsedio/tsed/blob/79b855d1374c18a86b39653381fe162d7413f3c8/packages/di/src/common/services/InjectorService.ts#L314C1-L314C41

  • tsed/packages/di/src/node/services/DITest.ts

    Line 54 in 79b855d

    await DITest.injector.load(createContainer());

Implementation: https://github.com/tsedio/tsed/blob/production/packages/di/src/common/utils/createContainer.ts#L5

export function createContainer(rootModule?: Type<any>) {
 // GlobalProviders.entries() => returns Provider
  const container = new Container(GlobalProviders.entries());

  if (rootModule) { // can be removed because there is no usage of rootModule signature in the code base (I check on tsed/cli also
    container.delete(rootModule);
  }

  return container;
}

Suggested:

import {Env} from "@tsed/core"
export function createContainer(env?: Env) { //or settings: DIConfiguration ?
  /// change the code implementation to 
}

To cover all this, we can do a test as follows:

@Controller({path: "/", environments: [Env.DEV]})
class MyController {
}

describe('Test env', () => {
   describe('Env = DEV', ()=> {
     beforeEach(() => PlatformTest.create({env: Env.DEV})
     afterEach(() => Platform.reset())
     
     it('should return the MyController provider', () => {
         expect(PlatformTest.injector.get(MyController)).toBeInstanceof(MyController)
     })
   })

    describe('Env = PROD', ()=> {
       beforeEach(() => PlatformTest.create({env: Env.PROD})
       afterEach(() => Platform.reset())
     
       it('should return the MyController provider', () => {
          expect(PlatformTest.injector.get(MyController)).not.toBeDefined()
      })
    })
})

See you :)

[abenerd]

You're right that would be even better!
In your snippets you're using the Env Enum, I would actually like to avoid that and use a simple string to give people more control over what their environments are called, names like staging, and super-cute-cats would not be possible with that. @Romakita

[Romakita]

Yes you can use a simple string type instead of enum ;)