davidrs 0.1.1

A small, feature-gated framework for AWS Lambda functions: one ordered pipeline per trigger, absolute invocation deadlines, bounded work and safe built-in error responses.
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
# HTTP

[`Api`](crate::http::Api) serves one route from API Gateway (an HTTP API, a REST API with the `apigw-rest` feature, or an Application Load Balancer with `alb`) or a Lambda Function URL. Enable `http`. Add `validate` for Garde checks on the body, and `problem` when the error body should be RFC 9457 problem details. Streamed responses are [Streaming](crate::guide::streaming). Who may call, and how many times, is [Access control](crate::guide::access). [Getting started](crate::guide::getting_started) builds the same shape with a path parameter and runs it locally.

The pipeline is configured once, with an operation name, a [`Policy`](crate::http::Policy) and an [`ErrorRenderer`](crate::http::ErrorRenderer). Every invocation then runs the same steps, under one deadline, and every failure leaves through the same renderer. The handler receives the shared state, a decoded input and a [`Context`](crate::Context), and returns a value or a [`Failure`](crate::http::Failure). One `Api` does not route. Several operations are several functions, unless they truly share one decoder. The program under [The pipeline](#the-pipeline) is a complete function. The tables after it are what you need when a status, a header or a decoder is wrong.

## Why it exists

A Lambda behind API Gateway is easy to write and easy to get subtly wrong:

- **A 500 leaks internals.** `format!("{error}")` in an error body sends a table name, an address or an upstream service's message to the client. Here every 5xx renders a fixed message, however the failure was built.
- **A success answers 200 with a broken body.** Serialization that fails after the status is chosen sends an empty or truncated `200`. Here serialization is a step of the pipeline, and its failure is a rendered `500`.
- **The rate limit reads the body first.** A limiter that runs after decoding pays for parsing every refused request. Here admission runs before anything is parsed.
- **The body is unbounded.** A decoder that reads whatever arrives can be handed megabytes. Here every decoder, including one that reads raw bytes, sits behind a body limit.
- **Lambda's timeout answers for you.** A handler still awaiting when the function is stopped leaves the client a gateway error with no body. Here the pipeline stops 100 ms early and renders a `504`.
- **Headers go missing on the error path.** Cache headers set in the success path, or rate-limit headers set before a refusal, are easy to lose. Here a finalizer sees every response, and admission headers reach successes and failures alike.

## The pipeline

Each invocation runs these steps in order:

| Step | Runs | Produces |
| --- | --- | --- |
| 1. admission | async, before any parsing | headers for the response, or a failure |
| 2. decode | sync, after the body limit | the handler's input, or a `400`, `413` or `415` |
| 3. policy | async | the scope the handler runs with, or a `401` or `403` |
| 4. handler | async | a value that implements [`IntoResponse`](crate::http::IntoResponse), or a failure |
| 5. serialization | sync | the response, or a `500` |

Steps 1 to 5 share the invocation deadline minus a 100 ms reserve; when it runs out the pending step is dropped and a `504` (`ERROR_TIMEOUT`) is rendered. A failure from any step is logged with its code and kind only, never its message, then rendered. The finalizer set with [`Api::finalize`](crate::http::Api::finalize) runs last, on every response.

A complete function is one expression in `main`:

```rust,no_run
use std::sync::Arc;

use davidrs::http::{Api, Failure, HeaderValue, Json, PlainErrors, Public, Request};
use davidrs::{Context, RuntimeError};
use serde::{Deserialize, Serialize};

/// The shared state, built once per process.
struct App {
    prefix: String,
}

#[derive(Deserialize)]
struct NewItem {
    name: String,
}

#[derive(Serialize)]
struct Item {
    id: String,
}

async fn create(app: Arc<App>, input: NewItem, _context: Context<()>) -> Result<Json<Item>, Failure> {
    Ok(Json(Item { id: format!("{}{}", app.prefix, input.name) }))
}

#[tokio::main]
async fn main() -> Result<(), RuntimeError> {
    let app = Arc::new(App { prefix: "item-".to_owned() });
    Api::new("create-item", Public, PlainErrors)
        .body_limit(64 * 1024)
        .finalize(|_invocation, response| {
            response.headers_mut().insert("cache-control", HeaderValue::from_static("no-store"));
        })
        .run(app, |request: &Request<'_>| request.json::<NewItem>(), create)
        .await
}
```

[`Api::run`](crate::http::Api::run) starts the Lambda loop. [`Api::handle`](crate::http::Api::handle) runs one request and returns the response, which is how a test drives a route without the runtime. A request built by hand has no Lambda context; it gets an empty request id and Lambda's maximum budget.

```rust
use std::sync::Arc;

use davidrs::http::{Api, Body, Failure, Json, PlainErrors, Public, Request, StatusCode};
use davidrs::Context;

async fn double(_app: Arc<()>, n: u32, _context: Context<()>) -> Result<Json<u32>, Failure> {
    Ok(Json(n * 2))
}

# #[tokio::main(flavor = "current_thread")]
# async fn main() {
let api = Api::new("double", Public, PlainErrors);
let decode = |request: &Request<'_>| request.json::<u32>();

let request = lambda_http::http::Request::builder()
    .method("POST")
    .uri("https://example.com/double")
    .body(Body::Text("21".to_owned()))
    .unwrap();
let response = api.handle(Arc::new(()), request, &decode, &double).await;
assert_eq!(response.status(), StatusCode::OK);
assert_eq!(response.body(), &Body::Text("42".to_owned()));

let garbage = lambda_http::http::Request::builder()
    .method("POST")
    .uri("https://example.com/double")
    .body(Body::Text("twenty-one".to_owned()))
    .unwrap();
let response = api.handle(Arc::new(()), garbage, &decode, &double).await;
assert_eq!(response.status(), StatusCode::BAD_REQUEST);
# }
```

Use it for any request–response endpoint behind API Gateway or a Function URL: a create or update that takes a JSON body, a lookup by path parameter, a listing with query filters, a webhook whose signature a policy checks over the raw bytes.

It does not route. One `Api` is one operation; several operations are several functions, or a decoder that dispatches on [`Request::method`](crate::http::Request::method) when they truly belong together. There is no middleware stack: the extension points are admission, policy, renderer and finalizer, each with one job. Serialization cannot be interrupted, so keep response values bounded. For streamed responses, server-sent events and CORS, use the `http-stream` feature's `StreamApi`.

## Reading the request

The decoder and the policy receive a [`Request`](crate::http::Request), a borrowed view of the native request that carries the body limit. Every typed read returns a [`Failure`](crate::http::Failure) of kind `Decode` instead of panicking:

| Read | Returns | Refuses with |
| --- | --- | --- |
| [`path`](crate::http::Request::path) | path parameters as `T` | `400 ERROR_INVALID_PATH` |
| [`query`](crate::http::Request::query) | the query string as `T` | `400 ERROR_INVALID_QUERY` |
| [`query_pairs`](crate::http::Request::query_pairs) | every `(name, value)`, repetitions kept | — |
| [`json`](crate::http::Request::json) | the body as `T` | `413`, `415`, `400 ERROR_MALFORMED_BODY` |
| [`json_bounded`](crate::http::Request::json_bounded) | the same, under a tighter limit | the same |
| [`json_text`](crate::http::Request::json_text) | the body as UTF-8, for your own decoder | `413`, `415`, `400` |
| [`form`](crate::http::Request::form) | an `application/x-www-form-urlencoded` body as `T` | `413`, `415`, `400 ERROR_MALFORMED_BODY` |
| [`raw_body`](crate::http::Request::raw_body) | the exact bytes, for a signature | — |
| [`media_type`](crate::http::Request::media_type) | `Content-Type`, lowercased, without parameters | — |
| [`source_ip`](crate::http::Request::source_ip) | the gateway's `sourceIp`, or `"unknown"` | — |

A JSON body with no `Content-Type` is accepted, because many clients send JSON without declaring it; a body that declares another media type is a `415` and is never parsed. An empty body is a `400`. The body limit is 1 MiB ([`DEFAULT_BODY_LIMIT`](crate::http::DEFAULT_BODY_LIMIT)) unless [`Api::body_limit`](crate::http::Api::body_limit) sets another, and the pipeline checks it before the decoder runs.

```rust
use davidrs::http::{Body, Request};
use serde::Deserialize;

#[derive(Deserialize)]
struct Filter {
    status: String,
    page: Option<u32>,
}

#[derive(Debug, Deserialize)]
struct Note {
    text: String,
}

let native = lambda_http::http::Request::builder()
    .method("POST")
    .uri("https://example.com/orders/notes?status=open&page=2")
    .header("content-type", "application/json; charset=utf-8")
    .body(Body::Text(r#"{"text":"call back"}"#.to_owned()))
    .unwrap();
let request = Request::new(&native);

let filter: Filter = request.query().unwrap();
assert_eq!((filter.status.as_str(), filter.page), ("open", Some(2)));
assert_eq!(request.json::<Note>().unwrap().text, "call back");
assert_eq!(request.media_type().as_deref(), Some("application/json"));

let tight = request.json_bounded::<Note>(8).unwrap_err();
assert_eq!(tight.status().as_u16(), 413);
```

Reach for [`form`](crate::http::Request::form) when an HTML form, a webhook or an OAuth client posts `application/x-www-form-urlencoded`, for [`json_text`](crate::http::Request::json_text) when an application decoder needs the text, for [`raw_body`](crate::http::Request::raw_body) when a policy verifies an HMAC over the exact bytes, for [`query_pairs`](crate::http::Request::query_pairs) when a list arrives as `?tag=a&tag=b`, and for [`source_ip`](crate::http::Request::source_ip) as a rate-limit key. Behind an Application Load Balancer (feature `alb`), `source_ip` is the last address of `X-Forwarded-For`, the one the load balancer appends in its default `append` mode.

For gateway events, `source_ip` ignores `X-Forwarded-For`: trusting a forwarded header needs a proxy policy of the application's own, and without one any caller could choose its own rate-limit key. The view never copies the body and never awaits; for anything it does not expose, [`Request::native`](crate::http::Request::native) returns the underlying request.

## Answering

A handler returns any [`IntoResponse`](crate::http::IntoResponse):

| Value | Response |
| --- | --- |
| [`Json(value)`](crate::http::Json) | `200`, `application/json`, the serialized value |
| [`NoContent`](crate::http::NoContent) | `204`, no body |
| `(StatusCode, T)` | `T`'s response with that status |
| `(StatusCode, HeaderMap, T)` | the same, with those headers set over `T`'s |
| `Option<T>` | `T`'s response, or `404 ERROR_NOT_FOUND` for `None` |
| [`HttpResponse`](crate::http::HttpResponse) | itself, unchanged |

Conversion returns a `Result`: when a value cannot be serialized, the result is a `500 FAULT_SERIALIZATION` of kind `Serialization`, and a status chosen in a tuple cannot hide it.

```rust
use davidrs::http::{HeaderMap, HeaderValue, IntoResponse, Json, NoContent, StatusCode};

let mut headers = HeaderMap::new();
headers.insert("location", HeaderValue::from_static("/orders/7"));
let created = (StatusCode::CREATED, headers, Json(serde_json::json!({"id": 7})))
    .into_response()
    .unwrap();
assert_eq!(created.status(), StatusCode::CREATED);
assert_eq!(created.headers()["location"], "/orders/7");

let missing = Option::<NoContent>::None.into_response().unwrap_err();
assert_eq!(missing.status(), StatusCode::NOT_FOUND);
```

Use `Option` for a lookup by id, a tuple for `201 Created` with a `Location`, `NoContent` for a delete, and a built [`HttpResponse`](crate::http::HttpResponse) for a CSV or any body that is not JSON. There is no content negotiation and no compression: a handler that needs them builds the response itself.

## Failures

A [`Failure`](crate::http::Failure) is what a handler returns when it cannot succeed. It carries:

| Part | Rendered | Purpose |
| --- | --- | --- |
| status | yes | the HTTP status |
| code | yes | a stable identifier clients can match on |
| message | 4xx only | what the client may read |
| detail | never | text for a deliberate diagnostic |
| kind | never | the step that failed: admission, decode, policy, handler, serialization, deadline |
| headers | yes | `Retry-After`, `Allow`, rate-limit fields |

A renderer must read the client-visible message through [`public_message`](crate::http::Failure::public_message), which returns [`INTERNAL_MESSAGE`](crate::http::INTERNAL_MESSAGE) for every 5xx. `Display` shows the same thing, so an accidental `{failure}` in a body cannot leak either. [`internal_detail`](crate::http::Failure::internal_detail) returns the unredacted message and the detail for a log line you write on purpose; the pipeline's own log records only the operation, request id, code and kind.

```rust
use davidrs::http::{Failure, FailureKind, StatusCode, INTERNAL_MESSAGE};

let closed = Failure::new(StatusCode::CONFLICT, "ERROR_ORDER_CLOSED", "The order is closed")
    .with_detail("order 7 closed at 12:00");
assert_eq!(closed.public_message(), "The order is closed");
assert_eq!(closed.kind(), FailureKind::Handler);

let upstream = Failure::new(StatusCode::BAD_GATEWAY, "FAULT_UPSTREAM", "connection reset by 10.0.0.4");
assert_eq!(upstream.public_message(), INTERNAL_MESSAGE);
assert_eq!(upstream.to_string(), "502 InternalServerError");
assert!(upstream.internal_detail().contains("10.0.0.4"));

let io = std::io::Error::other("disk full");
let stored = Failure::from_error("FAULT_STORE", &io);
assert_eq!(stored.status(), StatusCode::INTERNAL_SERVER_ERROR);
assert_eq!(stored.internal_detail(), "disk full");

let busy = Failure::new(StatusCode::SERVICE_UNAVAILABLE, "FAULT_BUSY", "Try later")
    .with_header("retry-after", "5");
assert_eq!(busy.headers()[0].1, "5");
```

Build a 4xx with [`Failure::new`](crate::http::Failure::new) and a message the client can act on. Build a 5xx with [`Failure::internal`](crate::http::Failure::internal) or [`Failure::from_error`](crate::http::Failure::from_error), which keeps an error's whole source chain as the detail, or convert a [`RuntimeError`](crate::RuntimeError) with `?`. A header whose name or value is invalid is dropped instead of panicking.

A `Failure` does not log itself and does not decide whether a request may be retried: that is the status and the operation's own contract.

### Error catalogs

An [`ErrorCatalog`](crate::http::ErrorCatalog) declares a service's failures in one place, each an [`ErrorDefinition`](crate::http::ErrorDefinition) with a code, a status and a message. A `const` slice of definitions is a catalog; a code it does not know becomes a `500 FAULT_UNHANDLED` rather than a panic.

```rust
use davidrs::http::{ErrorCatalog, ErrorDefinition, StatusCode};

const ERRORS: &[ErrorDefinition] = &[
    ErrorDefinition::new("ERROR_ORDER_NOT_FOUND", StatusCode::NOT_FOUND, "The order does not exist"),
    ErrorDefinition::new("ERROR_ORDER_CLOSED", StatusCode::CONFLICT, "The order is closed"),
];

let failure = ERRORS.failure("ERROR_ORDER_CLOSED");
assert_eq!(failure.status(), StatusCode::CONFLICT);
assert_eq!(ERRORS.failure("ERROR_TYPO").status(), StatusCode::INTERNAL_SERVER_ERROR);
```

Use a catalog when clients document your error codes, or when several functions share one vocabulary. The pipeline never requires one, and implementing the trait over a map or a generated table works the same way.

## Renderers

An [`ErrorRenderer`](crate::http::ErrorRenderer) turns a failure into the response the client sees. It cannot fail: whatever it builds is what goes on the wire. Two are included:

- [`PlainErrors`](crate::http::PlainErrors) writes `{"errorCode": "...", "errorMessage": "..."}` as `application/json`.
- [`ProblemErrors`](crate::http::ProblemErrors), with the `problem` feature, writes RFC 9457 `application/problem+json`: `status`, a `title` from the status, and the public message as `detail`. [`with_type_base`](crate::http::ProblemErrors::with_type_base) turns the code into the `type` URI; without it `type` is omitted, which means `about:blank`.

Both copy the failure's headers onto the response.

```rust
use davidrs::http::{Body, ErrorRenderer, Failure, PlainErrors, ProblemErrors, StatusCode};

let failure = Failure::new(StatusCode::NOT_FOUND, "ERROR_NOT_FOUND", "No such order");

let plain = PlainErrors.render(&failure);
assert_eq!(
    plain.body(),
    &Body::Text(r#"{"errorCode":"ERROR_NOT_FOUND","errorMessage":"No such order"}"#.to_owned())
);

let problem = ProblemErrors::default()
    .with_type_base("https://example.com/errors/")
    .render(&failure);
assert_eq!(problem.headers()["content-type"], "application/problem+json");
let Body::Text(text) = problem.body() else { unreachable!() };
let json: serde_json::Value = serde_json::from_str(text).unwrap();
assert_eq!(json["type"], "https://example.com/errors/ERROR_NOT_FOUND");
assert_eq!(json["detail"], "No such order");
```

When clients expect another shape, write your own. [`literal`](crate::http::literal) builds the response from a status, a media type and a body you already serialized; if the media type is not a valid header value it returns an empty `500` rather than a half-built response. Read the message through `public_message`, and copy [`headers`](crate::http::Failure::headers) so a `429` keeps its `Retry-After`:

```rust
use davidrs::http::{literal, ErrorRenderer, Failure, HttpResponse, StatusCode};

struct XmlErrors;

impl ErrorRenderer for XmlErrors {
    fn render(&self, failure: &Failure) -> HttpResponse {
        let body = format!("<error code=\"{}\">{}</error>", failure.code(), failure.public_message());
        let mut response = literal(failure.status(), "application/xml", body);
        for (name, value) in failure.headers() {
            response.headers_mut().insert(name.clone(), value.clone());
        }
        response
    }
}

let response = XmlErrors.render(&Failure::internal("FAULT_STORE", "table unreachable"));
assert_eq!(response.status(), StatusCode::INTERNAL_SERVER_ERROR);
```

The renderer owns the envelope and nothing else: it cannot change which step failed, and it never sees the success path. Headers every response needs belong in the finalizer.

## Policies and admission

A [`Policy`](crate::http::Policy) decides who is calling and returns a `Scope`, the authority the handler runs with. The scope is evidence: it exists only because the policy succeeded, so keep its fields private and construct it nowhere else. The handler reads it from [`Context::scope`](crate::Context::scope) and never re-reads a header. [`Public`](crate::http::Public) admits everyone with the scope `()`.

```rust
use std::sync::Arc;

use davidrs::http::{Api, Body, Failure, Json, PlainErrors, Policy, Request, StatusCode};
use davidrs::{Context, Invocation};

/// The caller a request was authorized for.
pub struct Caller {
    user: String,
}

/// Admits the users on an allowlist, named by a header a gateway authorizer sets.
struct Allowlist(Vec<String>);

impl Policy for Allowlist {
    type Scope = Caller;

    async fn authorize(&self, request: &Request<'_>, _: &Invocation) -> Result<Caller, Failure> {
        let user = request.header("x-user").unwrap_or_default();
        if !self.0.iter().any(|allowed| allowed == user) {
            return Err(Failure::new(StatusCode::FORBIDDEN, "ERROR_FORBIDDEN", "Not allowed"));
        }
        Ok(Caller { user: user.to_owned() })
    }
}

async fn whoami(_app: Arc<()>, _input: (), context: Context<Caller>) -> Result<Json<String>, Failure> {
    Ok(Json(context.scope().user.clone()))
}

# #[tokio::main(flavor = "current_thread")]
# async fn main() {
let api = Api::new("whoami", Allowlist(vec!["ada".to_owned()]), PlainErrors);
let request = lambda_http::http::Request::builder()
    .uri("https://example.com/whoami")
    .header("x-user", "grace")
    .body(Body::Empty)
    .unwrap();
let response = api.handle(Arc::new(()), request, &|_: &Request<'_>| Ok(()), &whoami).await;
assert_eq!(response.status(), StatusCode::FORBIDDEN);
# }
```

The policy runs after decoding, so a policy that verifies a signature can read [`raw_body`](crate::http::Request::raw_body), and a malformed request is refused before any key is fetched. Token verification itself is the `auth` feature's job; the policy decides what a verified token may do.

An [`Admission`](crate::http::Admission) runs first, before the body is read, and either admits the request with headers for the response or refuses it. [`AdmitAll`](crate::http::AdmitAll) is the default. Use admission for rate limits, quotas and maintenance switches: anything that must refuse cheaply. It cannot see the decoded input or the caller's scope; a limit that depends on who the caller is belongs in the policy or the handler.

### Rate-limit headers

A rate limiter reports its budget with the IETF `RateLimit-Policy` and `RateLimit` header fields. [`RateLimit`](crate::http::RateLimit) formats them from a [`RateLimitConfig`](crate::http::RateLimitConfig), the requests left and when the window resets, whatever store keeps the counter; it stores nothing itself. An admission returns [`RateLimit::headers`](crate::http::RateLimit::headers), and the pipeline attaches them to the response whether the request then succeeds or fails; on a refusal, attach them to the `429` itself.

```rust
use std::time::{Duration, SystemTime};

use davidrs::http::{Admission, Failure, RateLimit, RateLimitConfig, Request, StatusCode};
use davidrs::Invocation;

const SEARCH: RateLimitConfig = RateLimitConfig::new("search", 60, Duration::from_secs(300));

/// A limiter whose store is left out: every caller has 7 requests left.
struct PerAddress;

impl PerAddress {
    /// Spends one request for `key`; returns what is left and when the window
    /// resets.
    async fn spend(&self, _key: &str) -> (i64, SystemTime) {
        (7, SystemTime::now() + Duration::from_secs(90))
    }
}

impl Admission for PerAddress {
    async fn check(&self, request: &Request<'_>, _: &Invocation) -> Result<Vec<(String, String)>, Failure> {
        let (remaining, resets_at) = self.spend(&request.source_ip()).await;
        let limit = RateLimit::new(&SEARCH, remaining, resets_at);
        if limit.is_exceeded() {
            return Err(Failure::new(StatusCode::TOO_MANY_REQUESTS, "ERROR_RATE_LIMITED", "Too many requests")
                .with_headers(limit.headers()));
        }
        Ok(limit.headers())
    }
}

# #[tokio::main(flavor = "current_thread")]
# async fn main() {
let native = lambda_http::http::Request::builder().body(davidrs::http::Body::Empty).unwrap();
let invocation = davidrs::Invocation::new("r", davidrs::Deadline::after(Duration::from_secs(1)));
let headers = PerAddress.check(&Request::new(&native), &invocation).await.unwrap();
assert_eq!(headers[0], ("RateLimit-Policy".to_owned(), "\"default\";q=60;w=300".to_owned()));
assert!(headers[1].1.starts_with("\"default\";r=7;t="));
# }
```

## The pipeline's own codes

Most codes come from your service. The pipeline refuses some requests before a handler runs, and those failures use the constants in [`codes`](crate::http::codes):

| Code | Status | Raised when |
| --- | --- | --- |
| `ERROR_INVALID_PATH` | 400 | path parameters do not deserialize |
| `ERROR_INVALID_QUERY` | 400 | the query string does not deserialize |
| `ERROR_MALFORMED_BODY` | 400 | the body is empty, not UTF-8 or not valid JSON |
| `ERROR_INVALID_BODY` | 400 | the body fails validation |
| `ERROR_BODY_TOO_LARGE` | 413 | the body is over the limit |
| `ERROR_UNSUPPORTED_MEDIA_TYPE` | 415 | the body declares a media type other than JSON |
| `ERROR_NOT_FOUND` | 404 | a handler returned `None` |
| `ERROR_TIMEOUT` | 504 | the pipeline's deadline ran out (a converted [`RuntimeError`](crate::RuntimeError) deadline is a `500`) |
| `FAULT_SERIALIZATION` | 500 | the success value did not serialize |
| `ERROR_LIMIT_EXCEEDED`, `FAULT_UNHANDLED` | 500 | a [`RuntimeError`](crate::RuntimeError) was converted, or a catalog did not know a code |

The streamed pipeline adds its own for CORS, methods and `Accept`; [`codes`](crate::http::codes) lists them all. A code's prefix is not a retry instruction: `ERROR_TIMEOUT` is a server-side failure.

These spellings are defaults, not a contract you are stuck with. When your clients expect another vocabulary, map the codes in your renderer, which sees the whole failure:

```rust
use davidrs::http::{codes, literal, ErrorRenderer, Failure, HttpResponse, StatusCode};

struct SnakeCase;

impl ErrorRenderer for SnakeCase {
    fn render(&self, failure: &Failure) -> HttpResponse {
        let code = match failure.code() {
            codes::MALFORMED_BODY | codes::INVALID_BODY => "invalid_request",
            codes::TIMEOUT => "timeout",
            other => other,
        };
        let body = serde_json::json!({"code": code, "message": failure.public_message()}).to_string();
        let mut response = literal(failure.status(), "application/json", body);
        for (name, value) in failure.headers() {
            response.headers_mut().insert(name.clone(), value.clone());
        }
        response
    }
}

let response = SnakeCase.render(&Failure::new(StatusCode::BAD_REQUEST, codes::MALFORMED_BODY, "Invalid request body"));
assert_eq!(response.status(), StatusCode::BAD_REQUEST);
```

## Validation with Garde

With the `validate` feature, [`Request::validated_json`](crate::http::Request::validated_json) decodes a JSON body and checks it with a [Garde](https://docs.rs/garde) `Validate` derive. A body that decodes but breaks a rule is a `400 ERROR_INVALID_BODY`, and the handler never runs, so it only ever sees values that hold.

```rust
use davidrs::http::{codes, Body, Request};

#[derive(Debug, serde::Deserialize, garde::Validate)]
#[serde(deny_unknown_fields)]
struct NewOrder {
    #[garde(length(min = 1, max = 64))]
    reference: String,
    #[garde(range(min = 1, max = 100))]
    quantity: u32,
}

let native = lambda_http::http::Request::builder()
    .body(Body::Text(r#"{"reference":"A-1","quantity":0}"#.to_owned()))
    .unwrap();
let failure = Request::new(&native).validated_json::<NewOrder>().unwrap_err();
assert_eq!(failure.code(), codes::INVALID_BODY);
```

Use it for field constraints a type cannot express: lengths, ranges, nested items with `#[garde(dive)]`. The failure says that validation failed, not which field: a client that needs every invalid field listed gets that from the [`schema`](crate::http::schema) checks, and a rule that needs context (the caller, a stored value) runs in the handler with `validate_with`. Garde's own default features (regex, email, URL, phone numbers) are not enabled.

## Problem details

With the `problem` feature, [`ProblemErrors`](crate::http::ProblemErrors) renders every failure as RFC 9457 `application/problem+json`, the format standard HTTP clients and API gateways recognise. It is a renderer like any other: pass it to [`Api::new`](crate::http::Api::new) and every failure, from admission to serialization, uses it. The public-message rule holds, so a 5xx `detail` is always `InternalServerError`. It writes no extension members: a client that needs more than `type`, `title`, `status` and `detail` needs a renderer of its own.

## If the response is wrong

The status tells you which step ran. The handler is not always the place to look.

| What the client receives | The step that produced it | What to change |
| --- | --- | --- |
| `400` `ERROR_INVALID_PATH` or `ERROR_INVALID_QUERY` | the decoder, before the handler | The path or query does not match the struct. Check the API Gateway route and the field names. |
| `400` `ERROR_MALFORMED_BODY` | the decoder | The body is empty, not UTF-8, or not JSON. An empty body is a `400`, not a missing optional. |
| `413` `ERROR_BODY_TOO_LARGE` | the body limit, before the decoder | Raise [`Api::body_limit`](crate::http::Api::body_limit) for this route, or send less. The default is 1 MiB. |
| `415` `ERROR_UNSUPPORTED_MEDIA_TYPE` | the decoder | The body declared a media type other than JSON. A missing `Content-Type` is accepted. Another type is not, and the body is not parsed. |
| `401` or `403` | the policy | The handler did not run. Fix the policy, or the authorizer in front of it. Do not re-read the header in the handler. |
| `404` `ERROR_NOT_FOUND` | `Option::None` from the handler | The lookup found nothing. That is a response, not a missing route. This pipeline does not route. |
| `500` `FAULT_SERIALIZATION` | serializing a success value | The handler returned `Ok` with a value that cannot be serialized. The client does not receive a `200` with a broken body. |
| `500` with `errorMessage` `InternalServerError` | the handler, or a converted [`RuntimeError`](crate::RuntimeError) | The public message is fixed. Read the failure's code, and log [`internal_detail`](crate::http::Failure::internal_detail) yourself if you need the cause. |
| `504` `ERROR_TIMEOUT` | the deadline, 100 ms before Lambda's timeout | The pending step was dropped so the pipeline could still answer. Shorten the work, or raise the function timeout and keep it under the gateway's integration timeout. |
| API Gateway `502` `{"message":"Internal server error"}` | not this pipeline | The invocation failed before a response was posted, or the function was killed. Return [`Failure`](crate::http::Failure) from the handler. Do not let `main` see a request error. |