Vapor 3

Transcript

1. Vapor 3 By Joannis Orlandos
2. — Insanely fast & efficient — Clean, simple and maintainable — Reliable at small and large scale — Flexibility — A big and friendly community
3. Examples
4. Example: File uploads router.post("files", String.parameter) { request -> Future<HTTPResponse> in // Unsafe! Don't let users decide file paths let filename = try request.parameter(String.self) let file = try File(atPath: "/path/to/" + filename, flags: [.mustCreate]) let written = try request.http.body.output(to: file) return written.transform(to: HTTPResponse(status: .ok)) } — Would normally require fileSize of RAM — Now requires as little as 2KB
5. Example: File uploading with Base64 router.post("files", String.parameter) { request -> Future<HTTPResponse> in // Unsafe! Don't let users decide file paths let filename = try request.parameter(String.self) let file = try File(atPath: "/path/to/" + filename, flags: [.mustCreate]) let base64encoder = Base64Encoder().stream() let written = try request.http.body.stream(to: base64encoder).output(to: file) return written.transform(to: HTTPResponse(status: .ok)) } — Would normally need fileSize * 2.25 RAM — Now only needs as little as 3KB of RAM
6. Example: File downloads router.get("somepicture") { request -> Response in return try request.streamFile(at: "myfile.png") }
7. Example: HTTP Proxy behaviour router.get("json") { req -> Future<Response> in let client = try req.make(Client.self) // Return the response from this URL return client.get("http://example.vapor.codes/json") }
8. Reactive — Event driven programming — Not blocked by external sources — Low memory usage — Consistent performance
9. Why reactive programming?
10. Comparison to humans 1. Frontend and backend dev are working 2. Frontend dev needs an API and asks the backend dev 3. Backend dev blocks frontend dev 4. Frontend dev is now in "idle" state 5. Backend dev finished API 6. Frontend dev can now continue work
11. Ideal scenario 1. Frontend and backend dev are working 2. Frontend dev needs an API and asks the backend dev 3. Backend dev starts development 4. Frontend dev hops to the next task 5. Backend dev finished API 6. Frontend dev can continue the previous task
12. Vapor 3 "un-blocks" the application and developers working with it
13. "Multi-reactor" pa!ern — Reactors are Event Loops — One "Reactor" per virtual core — Independent Extra benefits — No locks, no deadlocks — No Data races — ✅ Stability
14. Thread safety 1. Frontend and backend dev are working 2. Frontend dev needs an API and asks the manager 3. Manager blocks both frontend and backend devs 4. Backend dev gets the API request from the manager 5. Backend dev starts development. Frontend dev is idle 6. Backend dev finishes his task. Hands it to manager 7. Manager hands the task to the frontend dev 8. Both continue working
15. Conclusion about locks — More overhead (idle time) — Less productivity — More complexity (bugs) — Not common between humans either
16. The Swi! Language
17. Primary language types — Compiled — JIT Compiled — Interpretedt
18. Type Languages Compiled C, C++, Swift, Go JIT Compiled Java, NodeJS, PHP 7 Interpreted Traditional JS, PHP 5.6
19. Statistical results
20. Name Req/sec Avg. Responsetime Base usage Top usage Vapor Engine 103_000 1.03ms 1.6MB 3.2MB Vapor 3 80_000 1.82ms 1.7MB 3.4MB Gin 77_000 1.26ms 6.5MB 14.0MB ExpressJS 46_000 2.40ms 16MB / Thread 64MB / Thread Laravel 1_600 14.20ms 30MB / Request 60MB / Request — High DDoS resistance — Low memory, High throughput — Configurable memory versus CPU tradeoff
21. Planned optimizations — Custom HTTP scanner — Better TCP buffer reuse — Protocol boundary scanning — Lazy protocol parsing