Skip to content

Simulated API Gateway HTTP APIs

Yulin includes a simulated API Gateway v2 service, reachable as simAws.apiGatewayV2(). It covers HTTP APIs with a Lambda proxy integration, so a handler that runs behind an HTTP API can be tested against a real HTTP request rather than against a hand-built event.

WebSocket APIs are not simulated, and neither are REST APIs, which are the older API Gateway v1 service and a separate SDK client.

CreateApiCommand creates an HTTP API and returns the endpoint API Gateway generates for it.

/**
* Creating a simulated API Gateway HTTP API.
*/
import { CreateApiCommand, GetApiCommand } from "@aws-sdk/client-apigatewayv2";
import { SimAws } from "@kensio/yulin";
const simAws = new SimAws();
const apiGateway = simAws
.account("555555555555")
.region("eu-west-1")
.apiGatewayV2();
const created = await apiGateway.createApi(
new CreateApiCommand({ Name: "orders", ProtocolType: "HTTP" }),
);
console.log(created.ApiId);
console.log(created.ApiEndpoint);
const fetched = await apiGateway.getApi(
new GetApiCommand({ ApiId: created.ApiId }),
);
console.log(fetched.Name);

The endpoint names the API id and the region, as a real one does:

https://a1b2c3d4e5.execute-api.eu-west-1.amazonaws.com

A name is not an identity here, as it is not on real AWS. Two APIs in one Account and Region may share a name, and only the id tells them apart.

An API needs three more resources before it serves anything: an integration naming the function, a route pointing at that integration, and a stage to serve it from. The function also has to allow API Gateway to invoke it, which is a permission on the function rather than anything on the API. See Granting the API permission to invoke the function. Once all of that exists, serveSimAws answers requests to the generated endpoint by invoking the function.

Pass the API endpoint through srv.localUrl(...), which keeps the endpoint’s hostname but sends the request to the local server, in the same way it adapts simulated S3 website and Lambda Function URL endpoints.

/**
* Serving a simulated HTTP API that proxies to a simulated Lambda function.
*/
import {
CreateApiCommand,
CreateIntegrationCommand,
CreateRouteCommand,
CreateStageCommand,
} from "@aws-sdk/client-apigatewayv2";
import {
AddPermissionCommand,
CreateFunctionCommand,
} from "@aws-sdk/client-lambda";
import { SimAws } from "@kensio/yulin";
import type { SimPayload2Event } from "@kensio/yulin/apigatewayv2";
import { makeLambdaZipFileInput } from "@kensio/yulin/lambda";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const { FunctionArn } = await simAws.lambda().createFunction(
new CreateFunctionCommand({
FunctionName: "orders",
Role: "arn:aws:iam::111111111111:role/OrdersRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimPayload2Event) => ({
statusCode: 200,
headers: { "content-type": "text/plain" },
body: `orders limit ${event.queryStringParameters?.["limit"] ?? "none"}`,
})),
},
}),
);
const apiGateway = simAws.apiGatewayV2();
const { ApiId, ApiEndpoint } = await apiGateway.createApi(
new CreateApiCommand({ Name: "orders", ProtocolType: "HTTP" }),
);
const { IntegrationId } = await apiGateway.createIntegration(
new CreateIntegrationCommand({
ApiId,
IntegrationType: "AWS_PROXY",
IntegrationUri: FunctionArn,
PayloadFormatVersion: "2.0",
}),
);
await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey: "$default",
Target: `integrations/${IntegrationId}`,
}),
);
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "$default", AutoDeploy: true }),
);
await simAws.lambda().addPermission(
new AddPermissionCommand({
FunctionName: "orders",
StatementId: "api-gateway-invoke",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/*/*`,
}),
);
const srv = await serveSimAws({ simAws });
const response = await fetch(srv.localUrl(`${ApiEndpoint}/orders?limit=10`));
console.log(response.status);
console.log(await response.text());
await srv.close();

The $default route matches any method and path, so every request to the endpoint reaches the function. The integration URI is the function’s ARN, and the function may be in another Account or Region: it is looked up where its ARN says it is.

Granting the API permission to invoke the function

Section titled “Granting the API permission to invoke the function”

A Lambda proxy integration does not work until the function’s resource policy allows apigateway.amazonaws.com to invoke it. The console adds that permission for you; an integration created through CloudFormation, the CLI or an SDK does not, which is what AddPermissionCommand in the example above is for. Without it the request is answered with a 500 and {"message":"Internal Server Error"}, and the handler does not run. CDK’s HttpLambdaIntegration emits the same grant as an AWS::Lambda::Permission.

Each request is authorized as lambda:InvokeFunction on the function ARN, with the caller being the service principal apigateway.amazonaws.com. The function’s own resource policy is what decides: a service principal has no identity policies of its own. The route that matched is supplied as AWS:SourceArn, so a permission may be granted for one route and not another:

arn:aws:execute-api:<region>:<account>:<apiId>/<stage>/<METHOD>/<route path>
  • The Account and Region are the API’s, not the function’s.
  • The stage is the one that served the request, so $default for the default stage.
  • The method is the request’s own, so a GET reaching a route keyed ANY /orders gives GET.
  • The path is the matched route key’s template with its parameter braces intact, so a request to /orders/42 on the route GET /orders/{orderId} gives orders/{orderId}. A SourceArn is written against route keys rather than against paths, and IAM treats a brace as an ordinary character.
  • The $default route has no method and no path of its own, so both collapse into one $default segment: <apiId>/<stage>/$default.

A SourceArn may wildcard any part of that, which is what the usual grant does: <apiId>/*/* allows every route of the API on every stage.

The API’s own Account is supplied as AWS:SourceAccount, so a permission carrying a SourceAccount matches when it names the Account the API belongs to. That is the Account the source ARN names, not the one owning the function, which matters for an integration reaching across Accounts. CDK writes both keys for some grants, and a permission carrying either or both is evaluated on what it says.

AWS:SourceArn and AWS:SourceAccount are the only condition keys supplied here. A permission that also carries PrincipalOrgID or InvokedViaFunctionUrl never matches, since nothing gives those keys a value at request time, so the request is refused with the same 500.

Neither the method nor the path is documented by AWS as the value API Gateway supplies. Both are inferred from the permission patterns AWS and CDK write, which is recorded next to the code that builds the ARN.

A route key is either the literal $default or an upper-case HTTP method and a path separated by one space:

GET /pets
GET /pets/{petId}
ANY /admin/{proxy+}
$default

The method is one of GET, POST, PUT, PATCH, DELETE, HEAD, OPTIONS or ANY, where ANY matches whatever method the request used.

A path is made of three kinds of segment:

  • A literal, such as pets, matching that one segment.
  • A parameter, such as {petId}, matching exactly one segment, whatever is in it.
  • A greedy parameter, such as {proxy+}, matching everything left of the path. It is only ever the last segment, and it needs at least one segment to match, so GET /pets/{proxy+} matches /pets/cat/1 but not /pets.

A route key that cannot be read is refused by CreateRoute with a BadRequestException, which is where real API Gateway refuses it too. That covers a lower-case method, an unbalanced brace, and a greedy parameter anywhere but the end.

A parameter name is not part of a route’s identity, so GET /pets/{id} and GET /pets/{petId} are the same route key and creating the second gives a ConflictException.

One path may not name the same parameter twice, so GET /pets/{id}/toys/{id} is refused. A handler reads path parameters off one object, so only one of the two captures could arrive. Whether real API Gateway refuses it is not established, so this is stricter than AWS rather than known to match it.

More than one route may match a request, and the most specific one takes it. In order:

  1. A route matching the whole path beats a route ending in a greedy parameter, which beats $default.
  2. An exact method beats ANY.
  3. The path decides, segment by segment from the left, with a literal beating a {name} parameter and a {name} beating a {name+}. Comparing left to right is also what makes the longest literal prefix win between two greedy routes, so GET /pets/dog/{proxy+} takes /pets/dog/collars/1 ahead of GET /pets/{proxy+}.

AWS’s worked example, which is encoded as a test here:

Request Route selected
GET /pets/dog/1 GET /pets/dog/1
GET /pets/dog/2 GET /pets/dog/{id}
GET /pets/cat/1 GET /pets/{proxy+}
POST /test/5 ANY /{proxy+}

from the routes GET /pets/dog/1, GET /pets/dog/{id}, GET /pets/{proxy+}, ANY /{proxy+} and $default.

Rule 1 is documented by AWS, as is the literal-beating-parameter part of rule 3. Rule 2, the longest-literal-prefix part of rule 3, and the place of the method comparison above the path comparison rather than below it, are observed rather than documented. Each is marked in the code next to the rule it governs.

A request whose path matches a route but whose method does not simply matches nothing. An API with no ANY route, no greedy route and no $default route to catch it answers 404, not 405.

What a route captured reaches the handler as event.pathParameters. A named stage is served under its own path segment, and stage selection runs before route selection, so the routes never see the stage name.

/**
* Matching a simulated HTTP API request by route key, path parameter and stage.
*/
import {
CreateApiCommand,
CreateIntegrationCommand,
CreateRouteCommand,
CreateStageCommand,
} from "@aws-sdk/client-apigatewayv2";
import {
AddPermissionCommand,
CreateFunctionCommand,
} from "@aws-sdk/client-lambda";
import { SimAws } from "@kensio/yulin";
import type { SimPayload2Event } from "@kensio/yulin/apigatewayv2";
import { makeLambdaZipFileInput } from "@kensio/yulin/lambda";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const { FunctionArn } = await simAws.lambda().createFunction(
new CreateFunctionCommand({
FunctionName: "pets",
Role: "arn:aws:iam::111111111111:role/PetsRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimPayload2Event) => ({
statusCode: 200,
headers: { "content-type": "application/json" },
body: JSON.stringify({
routeKey: event.routeKey,
rawPath: event.rawPath,
stage: event.requestContext.stage,
petId: event.pathParameters?.["petId"],
}),
})),
},
}),
);
const apiGateway = simAws.apiGatewayV2();
const { ApiId, ApiEndpoint } = await apiGateway.createApi(
new CreateApiCommand({ Name: "pets", ProtocolType: "HTTP" }),
);
const { IntegrationId } = await apiGateway.createIntegration(
new CreateIntegrationCommand({
ApiId,
IntegrationType: "AWS_PROXY",
IntegrationUri: FunctionArn,
PayloadFormatVersion: "2.0",
}),
);
for (const RouteKey of [
"GET /pets",
"GET /pets/{petId}",
"ANY /admin/{proxy+}",
"$default",
]) {
await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey,
Target: `integrations/${IntegrationId}`,
}),
);
}
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "dev", AutoDeploy: true }),
);
await simAws.lambda().addPermission(
new AddPermissionCommand({
FunctionName: "pets",
StatementId: "api-gateway-invoke",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/*/*`,
}),
);
const srv = await serveSimAws({ simAws });
const response = await fetch(srv.localUrl(`${ApiEndpoint}/dev/pets/6`));
console.log(await response.json());
await srv.close();

That handler reports four fields of its event, so the response body it produces is:

{
"routeKey": "GET /pets/{petId}",
"rawPath": "/dev/pets/6",
"stage": "dev",
"petId": "6"
}

rawPath and requestContext.http.path keep the stage segment, while routeKey and pathParameters come from the path the routes matched, which is /pets/6. The stage prefix in the path is corroborated by the documented $context.path access log variable, “the request path, for example /{stage}/root/child”, rather than by the payload format page.

A stage name is $default, or up to 128 alphanumerics, hyphens and underscores. The $default stage is served at the root of the endpoint instead of under a name, and both kinds can exist on one API at once. An explicit stage match wins over any route match: with a $default stage and a stage named pets, a request for /pets/dog is served by the stage pets on the route path /dog. A request reaching an API with no stage for it, and no $default stage, is a 404.

pathParameters is left out of the event entirely when the matched route captured nothing, including on a $default match. StageVariables set on the stage arrive as event.stageVariables, and are left out the same way when the stage has none.

Protecting a route with a Cognito user pool

Section titled “Protecting a route with a Cognito user pool”

A JWT authorizer verifies a signed token before the integration is invoked. CreateAuthorizerCommand creates one, and a route asks for it with AuthorizationType: "JWT" and the authorizer’s id.

The issuer is a URL. Point it at a simulated Cognito user pool and the pool’s own signing key verifies the token, so a token from InitiateAuthCommand or AdminInitiateAuthCommand reaches the route and nothing else does. The audience is the app client ids the authorizer admits.

/**
* Protecting a simulated HTTP API route with a Cognito user pool.
*/
import {
CreateApiCommand,
CreateAuthorizerCommand,
CreateIntegrationCommand,
CreateRouteCommand,
CreateStageCommand,
} from "@aws-sdk/client-apigatewayv2";
import {
AdminCreateUserCommand,
AdminInitiateAuthCommand,
AdminSetUserPasswordCommand,
CreateUserPoolClientCommand,
CreateUserPoolCommand,
} from "@aws-sdk/client-cognito-identity-provider";
import {
AddPermissionCommand,
CreateFunctionCommand,
} from "@aws-sdk/client-lambda";
import { SimAws } from "@kensio/yulin";
import type { SimPayload2Event } from "@kensio/yulin/apigatewayv2";
import { makeLambdaZipFileInput } from "@kensio/yulin/lambda";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const cognito = simAws.cognitoIdentityProvider();
const pool = await cognito.createUserPool(
new CreateUserPoolCommand({ PoolName: "myapp-users" }),
);
const UserPoolId = pool.UserPool!.Id!;
const appClient = await cognito.createUserPoolClient(
new CreateUserPoolClientCommand({
UserPoolId,
ClientName: "web",
ExplicitAuthFlows: ["ALLOW_ADMIN_USER_PASSWORD_AUTH"],
}),
);
const ClientId = appClient.UserPoolClient!.ClientId!;
await cognito.adminCreateUser(
new AdminCreateUserCommand({ UserPoolId, Username: "ada" }),
);
await cognito.adminSetUserPassword(
new AdminSetUserPasswordCommand({
UserPoolId,
Username: "ada",
Password: "Correct-horse-1",
Permanent: true,
}),
);
const { FunctionArn } = await simAws.lambda().createFunction(
new CreateFunctionCommand({
FunctionName: "orders",
Role: "arn:aws:iam::111111111111:role/OrdersRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimPayload2Event) => ({
statusCode: 200,
headers: { "content-type": "text/plain" },
body: `orders for ${
event.requestContext.authorizer?.jwt?.claims["username"] ?? "nobody"
}`,
})),
},
}),
);
const apiGateway = simAws.apiGatewayV2();
const { ApiId, ApiEndpoint } = await apiGateway.createApi(
new CreateApiCommand({ Name: "orders", ProtocolType: "HTTP" }),
);
const { IntegrationId } = await apiGateway.createIntegration(
new CreateIntegrationCommand({
ApiId,
IntegrationType: "AWS_PROXY",
IntegrationUri: FunctionArn,
PayloadFormatVersion: "2.0",
}),
);
const { AuthorizerId } = await apiGateway.createAuthorizer(
new CreateAuthorizerCommand({
ApiId,
Name: "pool-authorizer",
AuthorizerType: "JWT",
IdentitySource: ["$request.header.Authorization"],
JwtConfiguration: {
Issuer: `https://cognito-idp.us-east-1.amazonaws.com/${UserPoolId}`,
Audience: [ClientId],
},
}),
);
await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey: "GET /orders",
Target: `integrations/${IntegrationId}`,
AuthorizationType: "JWT",
AuthorizerId,
}),
);
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "$default", AutoDeploy: true }),
);
await simAws.lambda().addPermission(
new AddPermissionCommand({
FunctionName: "orders",
StatementId: "api-gateway-invoke",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/*/*`,
}),
);
const signedIn = await cognito.adminInitiateAuth(
new AdminInitiateAuthCommand({
UserPoolId,
ClientId,
AuthFlow: "ADMIN_USER_PASSWORD_AUTH",
AuthParameters: { USERNAME: "ada", PASSWORD: "Correct-horse-1" },
}),
);
const accessToken = signedIn.AuthenticationResult!.AccessToken!;
const srv = await serveSimAws({ simAws });
const url = srv.localUrl(`${ApiEndpoint}/orders`);
const anonymous = await fetch(url);
console.log(anonymous.status); // 401
console.log(anonymous.headers.get("www-authenticate")); // "Bearer"
const authorized = await fetch(url, {
headers: { authorization: `Bearer ${accessToken}` },
});
console.log(await authorized.text()); // "orders for ada"
// Advancing the simulation's clock past the token's expiry closes the route
// to the same token, with nothing reissued.
await simAws.clock().advanceBy({ hours: 2 });
const expired = await fetch(url, {
headers: { authorization: `Bearer ${accessToken}` },
});
console.log(expired.status); // 401
await srv.close();

The verification is real. The token is parsed, its alg has to be RS256, its kid has to name a key the issuer publishes, and the signature is checked against that key with node:crypto. Nothing is fetched over the network, and no verification library is involved.

A token that is missing, unreadable, signed with an unsupported algorithm, signed by an unknown key, or carrying a claim that does not hold is answered with a 401, {"message":"Unauthorized"} and a www-authenticate: Bearer header. The integration is never invoked. The client is told nothing about which check failed, which is what real API Gateway does, except for an audience that does not match: that one carries error_description="the token does not have a valid audience", the one description AWS publishes.

The claims are checked in the order AWS documents: the issuer, then the audience, then exp, nbf and iat. There is no allowance for clock skew, and every timestamp comes from the simulation’s clock, so simAws.clock().advanceBy(...) expires a token that was accepted a moment before.

IdentitySource takes one entry, either $request.header.<name> or $request.querystring.<name>. A Bearer prefix on the value, followed by whitespace, is stripped case-insensitively and is not required. Anything else is refused by CreateAuthorizer, because an authorizer looking for the token in a place nothing puts it refuses every request for a reason that reads like a signing problem.

Route scopes, and access tokens versus ID tokens

Section titled “Route scopes, and access tokens versus ID tokens”

AuthorizationScopes on a route is checked against the token’s scope claim, split on whitespace. The check is any-of: one matching scope is enough. A verified token that matches none of them is answered with a 403 and {"message":"Forbidden"}.

await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey: "GET /orders",
Target: `integrations/${IntegrationId}`,
AuthorizationType: "JWT",
AuthorizerId,
AuthorizationScopes: ["aws.cognito.signin.user.admin"],
}),
);

An ID token passes an authorizer that configures only an audience. Nothing checks token_use, which is what real API Gateway does, and an ID token’s aud is the app client id, so it matches. AWS documents this and recommends route scopes as the way to tell the two apart. A Cognito ID token has no scope claim at all, so any route scope refuses it.

Sign-in through the user pool API issues one scope, aws.cognito.signin.user.admin, and that is the only scope a simulated flow can put in a token. Resource servers, custom scopes and the client credentials grant are not simulated, so no other route scope is satisfiable.

An accepted token arrives as event.requestContext.authorizer.jwt:

{
"claims": {
"sub": "0a1b2c3d-...",
"iss": "https://cognito-idp.us-east-1.amazonaws.com/us-east-1_abc123",
"client_id": "1h57kf5cpparf3m47el34md5m9",
"token_use": "access",
"scope": "aws.cognito.signin.user.admin",
"username": "ada",
"cognito:groups": "[Admins Readers]",
"exp": "1785675600"
},
"scopes": ["aws.cognito.signin.user.admin"]
}

Every claim value is a string, whatever type it was signed as. A list claim such as cognito:groups is rendered the way Go prints a slice, so two groups arrive as [Admins Readers] rather than as JSON or as a comma-separated list. scopes is null, not an empty list, when the token carries no scope claim. None of that is published by AWS; all of it is what the real endpoint was observed to send.

A route with AuthorizationType: "NONE" has no caller to describe, so requestContext.authorizer is left out of its events entirely.

A route declared AuthorizationType: "AWS_IAM" reaches its integration only when the caller is allowed execute-api:Invoke on the ARN of the route being called. The route takes no authorizer, and naming one is refused: IAM itself is what decides.

The caller comes from the request, through either a SigV4 signature or an x-sim-aws-caller header naming a principal directly. A request that offers neither is anonymous, owns no policies, and is refused. See callers of HTTP requests in the IAM docs for how that resolution works and how to sign a served request.

The ARN a request is authorized against is:

arn:aws:execute-api:<region>:<account>:<apiId>/<stage>/<METHOD>/<path>
  • The Account and Region are the API’s own, not the caller’s.
  • The stage is the one that served the request, so $default for the default stage.
  • The method is the one the client sent, upper case, so a GET reaching a route keyed ANY /orders gives GET.
  • The path is the request path with the stage segment and the leading slash taken off, so /dev/orders/42 served from stage dev gives orders/42. It is the path asked for rather than the route key, because this is the resource a policy is written against by hand. A request to the API root gives an ARN ending /GET/.

An identity policy may wildcard any part of that. <apiId>/*, <apiId>/$default/* and <apiId>/*/GET/orders/* all allow a GET of /orders/42 on the default stage.

/**
* Protecting a simulated HTTP API route with IAM.
*/
import {
CreateApiCommand,
CreateIntegrationCommand,
CreateRouteCommand,
CreateStageCommand,
} from "@aws-sdk/client-apigatewayv2";
import { CreateRoleCommand, PutRolePolicyCommand } from "@aws-sdk/client-iam";
import {
AddPermissionCommand,
CreateFunctionCommand,
} from "@aws-sdk/client-lambda";
import { SimAws } from "@kensio/yulin";
import type { SimPayload2Event } from "@kensio/yulin/apigatewayv2";
import { makeLambdaZipFileInput } from "@kensio/yulin/lambda";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const { FunctionArn } = await simAws.lambda().createFunction(
new CreateFunctionCommand({
FunctionName: "orders",
Role: "arn:aws:iam::888888888888:role/OrdersRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimPayload2Event) => ({
statusCode: 200,
headers: { "content-type": "text/plain" },
body: `orders for ${
event.requestContext.authorizer?.iam?.userArn ?? "nobody"
}`,
})),
},
}),
);
const apiGateway = simAws.apiGatewayV2();
const { ApiId, ApiEndpoint } = await apiGateway.createApi(
new CreateApiCommand({ Name: "orders", ProtocolType: "HTTP" }),
);
const { IntegrationId } = await apiGateway.createIntegration(
new CreateIntegrationCommand({
ApiId,
IntegrationType: "AWS_PROXY",
IntegrationUri: FunctionArn,
PayloadFormatVersion: "2.0",
}),
);
await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey: "GET /orders/{orderId}",
Target: `integrations/${IntegrationId}`,
AuthorizationType: "AWS_IAM",
}),
);
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "$default", AutoDeploy: true }),
);
await simAws.lambda().addPermission(
new AddPermissionCommand({
FunctionName: "orders",
StatementId: "api-gateway-invoke",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/*/*`,
}),
);
// A Role of the API's own Account, allowed to call the orders routes of this
// API on the default stage.
await simAws.iam().createRole(
new CreateRoleCommand({
RoleName: "Reporter",
AssumeRolePolicyDocument: JSON.stringify({
Version: "2012-10-17",
Statement: [
{
Effect: "Allow",
Principal: { AWS: "arn:aws:iam::888888888888:root" },
Action: "sts:AssumeRole",
},
],
}),
}),
);
await simAws.iam().putRolePolicy(
new PutRolePolicyCommand({
RoleName: "Reporter",
PolicyName: "InvokeOrders",
PolicyDocument: JSON.stringify({
Version: "2012-10-17",
Statement: [
{
Effect: "Allow",
Action: "execute-api:Invoke",
Resource: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/$default/GET/orders/*`,
},
],
}),
}),
);
const srv = await serveSimAws({ simAws });
const url = srv.localUrl(`${ApiEndpoint}/orders/42`);
const anonymous = await fetch(url);
console.log(anonymous.status); // 403
console.log(await anonymous.text()); // '{"message":"Forbidden"}'
const reporter = await fetch(url, {
headers: { "x-sim-aws-caller": "arn:aws:iam::888888888888:role/Reporter" },
});
console.log(await reporter.text()); // "orders for arn:aws:iam::888888888888:role/Reporter"
await srv.close();

A caller IAM does not allow is answered with a 403 and {"message":"Forbidden"}, and the integration is never invoked. That is the answer for an unsigned request too: the serving boundary resolves it to an anonymous caller, and nothing allows an anonymous caller anything. An explicit Deny beats an Allow, as it does in any IAM evaluation.

A request whose signature is malformed, or is scoped to another service, does not reach the route at all. The serving boundary refuses it first, with {"Message":"Forbidden"} and a capital M.

An admitted request carries its caller into the event as requestContext.authorizer.iam:

{
"accessKey": "",
"accountId": "888888888888",
"callerId": "arn:aws:iam::888888888888:role/Reporter",
"cognitoIdentity": null,
"principalOrgId": null,
"userArn": "arn:aws:iam::888888888888:role/Reporter",
"userId": "arn:aws:iam::888888888888:role/Reporter"
}

accountId and userArn come from the resolved principal’s ARN. requestContext.accountId is that Account too, rather than anonymous. The block is the same one a Lambda Function URL produces, so handler code reading it behaves the same behind either.

A principal of another Account is refused, whatever its own Account allows it. A cross-Account request needs an Allow from the resource side as well, and an HTTP API has nowhere to put one: unlike a REST API, it has no resource policy at all. Real AWS behaves the same way.

The way through, here and on AWS, is for that principal to assume a Role in the API’s Account through STS and sign with the session credentials. The request is then made by a principal of the API’s own Account.

Protecting a route with a Lambda authorizer

Section titled “Protecting a route with a Lambda authorizer”

A Lambda REQUEST authorizer runs a function of its own before the integration is invoked, and that function decides. CreateAuthorizerCommand with AuthorizerType: "REQUEST" creates one, and a route asks for it with AuthorizationType: "CUSTOM" and the authorizer’s id.

IdentitySource is what the request has to carry before the function is invoked at all. Each entry is $request.header.<name> or $request.querystring.<name>, and a request missing any one of them is refused without the function running.

EnableSimpleResponses: true asks the function for { isAuthorized, context }. With it off, the function answers a principalId and an IAM policyDocument instead. Either way, the context it returns reaches the integration handler as event.requestContext.authorizer.lambda.

/**
* Protecting a simulated HTTP API route with a Lambda REQUEST authorizer.
*/
import {
CreateApiCommand,
CreateAuthorizerCommand,
CreateIntegrationCommand,
CreateRouteCommand,
CreateStageCommand,
} from "@aws-sdk/client-apigatewayv2";
import {
AddPermissionCommand,
CreateFunctionCommand,
} from "@aws-sdk/client-lambda";
import { SimAws } from "@kensio/yulin";
import type {
SimHttpApiAuthorizerEvent,
SimPayload2Event,
} from "@kensio/yulin/apigatewayv2";
import { makeLambdaZipFileInput } from "@kensio/yulin/lambda";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const lambda = simAws.lambda();
const { FunctionArn: AuthorizerFunctionArn } = await lambda.createFunction(
new CreateFunctionCommand({
FunctionName: "session-authorizer",
Role: "arn:aws:iam::888888888888:role/AuthorizerRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimHttpApiAuthorizerEvent) => ({
isAuthorized: event.identitySource[0] === "session=valid",
context: { tenant: "acme" },
})),
},
}),
);
const { FunctionArn } = await lambda.createFunction(
new CreateFunctionCommand({
FunctionName: "account",
Role: "arn:aws:iam::888888888888:role/AccountRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimPayload2Event) => ({
statusCode: 200,
headers: { "content-type": "application/json" },
body: JSON.stringify(event.requestContext.authorizer?.lambda),
})),
},
}),
);
const apiGateway = simAws.apiGatewayV2();
const { ApiId, ApiEndpoint } = await apiGateway.createApi(
new CreateApiCommand({ Name: "account", ProtocolType: "HTTP" }),
);
const { IntegrationId } = await apiGateway.createIntegration(
new CreateIntegrationCommand({
ApiId,
IntegrationType: "AWS_PROXY",
IntegrationUri: FunctionArn,
PayloadFormatVersion: "2.0",
}),
);
const { AuthorizerId } = await apiGateway.createAuthorizer(
new CreateAuthorizerCommand({
ApiId,
Name: "session-cookie",
AuthorizerType: "REQUEST",
AuthorizerUri: AuthorizerFunctionArn,
AuthorizerPayloadFormatVersion: "2.0",
EnableSimpleResponses: true,
IdentitySource: ["$request.header.cookie"],
}),
);
await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey: "GET /account",
Target: `integrations/${IntegrationId}`,
AuthorizationType: "CUSTOM",
AuthorizerId,
}),
);
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "$default", AutoDeploy: true }),
);
// Each function needs its own grant: the integration is invoked under the ARN
// of the route, and the authorizer under an ARN naming the authorizer.
await lambda.addPermission(
new AddPermissionCommand({
FunctionName: "account",
StatementId: "api-gateway-invoke",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/*/*`,
}),
);
await lambda.addPermission(
new AddPermissionCommand({
FunctionName: "session-authorizer",
StatementId: "api-gateway-invoke-authorizer",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/authorizers/${AuthorizerId}`,
}),
);
const srv = await serveSimAws({ simAws });
const url = srv.localUrl(`${ApiEndpoint}/account`);
const refused = await fetch(url, { headers: { cookie: "session=expired" } });
console.log(refused.status); // 403
const admitted = await fetch(url, { headers: { cookie: "session=valid" } });
console.log(await admitted.text()); // '{"tenant":"acme"}'
await srv.close();

The authorizer function is invoked once per request reaching the route. Nothing is cached between requests until the authorizer is given an AuthorizerResultTtlInSeconds, which is what AWS defaults to as well. See Caching the authorizer’s decision.

The function is invoked with the payload format 2.0 request event, plus three members of its own, exported as SimHttpApiAuthorizerEvent:

{
"version": "2.0",
"type": "REQUEST",
"routeArn": "arn:aws:execute-api:us-east-1:888888888888:a1b2c3d4e5/$default/GET/account",
"identitySource": ["session=valid"],
"routeKey": "GET /account",
"rawPath": "/account",
"rawQueryString": "",
"headers": { "cookie": "session=valid" },
"requestContext": { "...": "as the integration receives it" }
}

identitySource carries the values found at the authorizer’s identity sources, in the order they were configured, rather than the expressions that found them. routeArn names the route the request matched, with the route key’s path template rather than the path asked for, so GET /orders/{orderId} is <apiId>/<stage>/GET/orders/{orderId} whichever order was asked for.

There is no body and no isBase64Encoded. AWS’s published example of this event carries neither, so an authorizer cannot read the request body here, as it cannot on AWS.

With EnableSimpleResponses left off, the function answers a principalId and an IAM policy document, and simulated IAM evaluates it for execute-api:Invoke against the route ARN. That is the same evaluation an AWS_IAM route goes through, with one difference: the returned document is the whole decision, since there is no IAM principal behind it. principalId is a name the function chose for the caller and grants nothing.

makeLambdaZipFileInput((event: SimHttpApiAuthorizerEvent) => ({
principalId: "user-1",
policyDocument: {
Version: "2012-10-17",
Statement: [
{
Effect: event.identitySource[0] === "session=valid" ? "Allow" : "Deny",
Action: "execute-api:Invoke",
Resource: event.routeArn,
},
],
},
context: { tenant: "acme" },
}));

An explicit Deny beats an Allow, and a document allowing nothing relevant refuses the request.

  • A request missing any configured identity source is a 401 and {"message":"Unauthorized"}, and the authorizer function is never invoked.
  • isAuthorized: false, or a policy that does not allow execute-api:Invoke on the route ARN, is a 403 and {"message":"Forbidden"}.
  • Returning { "errorMessage": "Unauthorized" } is a 401. That is the only way an authorizer produces one, and it is read whichever response format the authorizer is configured for.
  • A function that throws, returns a shape neither response format matches, returns a policy document IAM cannot read, or has no invoke permission, is a 500 and {"message":"Internal Server Error"}. The caller is told nothing further, as it is told nothing about a failed integration.

The integration is never invoked in any of these cases.

The authorizer’s function is invoked under arn:aws:execute-api:<region>:<account>:<apiId>/authorizers/<authorizerId>, which is the SourceArn AWS documents for granting API Gateway permission to invoke one. That ARN names no stage and no route, so it is a different grant from the integration’s, and a function used for both needs both.

AuthorizerResultTtlInSeconds holds a decision for that many seconds, and a request presenting the same identity source values within it is served from that decision without the function running again. AWS accepts up to 3600, and 0, which is the default, holds nothing.

The key is the identity source values and nothing else, so one decision covers every route of the API that uses the authorizer. Adding $context.routeKey as an identity source puts the route in the key, which is what AWS documents for caching per route.

/**
* Caching a simulated HTTP API Lambda authorizer's decision.
*/
import {
CreateApiCommand,
CreateAuthorizerCommand,
CreateIntegrationCommand,
CreateRouteCommand,
CreateStageCommand,
} from "@aws-sdk/client-apigatewayv2";
import {
AddPermissionCommand,
CreateFunctionCommand,
} from "@aws-sdk/client-lambda";
import { SimAws } from "@kensio/yulin";
import type { SimPayload2Event } from "@kensio/yulin/apigatewayv2";
import { makeLambdaZipFileInput } from "@kensio/yulin/lambda";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const lambda = simAws.lambda();
// An authorizer counting its own invocations, so the caller can see which
// decision served each request.
const counter = { invocations: 0 };
const { FunctionArn: AuthorizerFunctionArn } = await lambda.createFunction(
new CreateFunctionCommand({
FunctionName: "session-authorizer",
Role: "arn:aws:iam::888888888888:role/AuthorizerRole",
Code: {
ZipFile: makeLambdaZipFileInput(() => {
counter.invocations += 1;
return { isAuthorized: true, context: { ...counter } };
}),
},
}),
);
const { FunctionArn } = await lambda.createFunction(
new CreateFunctionCommand({
FunctionName: "account",
Role: "arn:aws:iam::888888888888:role/AccountRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimPayload2Event) => ({
statusCode: 200,
headers: { "content-type": "application/json" },
body: JSON.stringify(event.requestContext.authorizer?.lambda),
})),
},
}),
);
const apiGateway = simAws.apiGatewayV2();
const { ApiId, ApiEndpoint } = await apiGateway.createApi(
new CreateApiCommand({ Name: "account", ProtocolType: "HTTP" }),
);
const { IntegrationId } = await apiGateway.createIntegration(
new CreateIntegrationCommand({
ApiId,
IntegrationType: "AWS_PROXY",
IntegrationUri: FunctionArn,
PayloadFormatVersion: "2.0",
}),
);
const { AuthorizerId } = await apiGateway.createAuthorizer(
new CreateAuthorizerCommand({
ApiId,
Name: "session-cookie",
AuthorizerType: "REQUEST",
AuthorizerUri: AuthorizerFunctionArn,
AuthorizerPayloadFormatVersion: "2.0",
EnableSimpleResponses: true,
IdentitySource: ["$request.header.cookie"],
AuthorizerResultTtlInSeconds: 300,
}),
);
await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey: "GET /account",
Target: `integrations/${IntegrationId}`,
AuthorizationType: "CUSTOM",
AuthorizerId,
}),
);
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "$default", AutoDeploy: true }),
);
for (const [FunctionName, SourceArn] of [
["account", `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/*/*`],
[
"session-authorizer",
`arn:aws:execute-api:us-east-1:888888888888:${ApiId}/authorizers/${AuthorizerId}`,
],
]) {
await lambda.addPermission(
new AddPermissionCommand({
FunctionName,
StatementId: "api-gateway-invoke",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn,
}),
);
}
const srv = await serveSimAws({ simAws });
const url = srv.localUrl(`${ApiEndpoint}/account`);
const call = async (): Promise<unknown> => {
const response = await fetch(url, { headers: { cookie: "session=valid" } });
return await response.json();
};
console.log(await call()); // { invocations: 1 }
console.log(await call()); // { invocations: 1 }, held rather than asked again
// Simulated time passing the TTL drops the decision.
await simAws.clock().advanceBy({ minutes: 6 });
console.log(await call()); // { invocations: 2 }
await srv.close();

A refusal is held the same way an admission is, so a session the authorizer rejected stays rejected until the TTL expires. What is not held is an authorizer that could not answer at all: a function that threw, or replied in neither format, is asked again on the next request.

Expiry is checked against the simulation’s clock, so simAws.clock().advanceBy(...) expires a decision that was being reused a moment before, and no test has to wait.

The context the authorizer returned arrives as event.requestContext.authorizer.lambda, with its values as the function returned them rather than stringified. An authorizer that allowed the request and returned no context leaves null there, so the block still says which kind of authorizer ran.

The handler is invoked with the API Gateway HTTP API payload format 2.0 event, exported as SimPayload2Event:

{
"version": "2.0",
"routeKey": "$default",
"rawPath": "/orders",
"rawQueryString": "limit=10",
"headers": { "host": "...", "x-forwarded-proto": "https" },
"queryStringParameters": { "limit": "10" },
"cookies": ["session=abc"],
"requestContext": {
"accountId": "anonymous",
"apiId": "a1b2c3d4e5",
"domainName": "a1b2c3d4e5.execute-api.eu-west-1.amazonaws.com",
"domainPrefix": "a1b2c3d4e5",
"http": {
"method": "GET",
"path": "/orders",
"protocol": "HTTP/1.1",
"sourceIp": "127.0.0.1",
"userAgent": "..."
},
"requestId": "...",
"routeKey": "$default",
"stage": "$default",
"time": "02/Aug/2026:11:00:00 +0000",
"timeEpoch": 1785668400000
},
"isBase64Encoded": false
}

A field with nothing in it is left out rather than set to null, which is what real API Gateway does: cookies, queryStringParameters, body, pathParameters and stageVariables are absent when the request has nothing for them. rawQueryString is the exception and is always present, as an empty string when there was no query.

Repeated query parameters are joined with commas, cookies travel in cookies rather than in a cookie header, and a body is passed through as text for a text content type and base64-encoded otherwise, with isBase64Encoded saying which happened.

The headers API Gateway sets itself replace whatever the client sent under those names: host is the API’s own hostname rather than the localhost one the request arrived at, x-forwarded-proto is https, x-forwarded-port is 443, x-forwarded-for and requestContext.http.sourceIp are 127.0.0.1, and x-amzn-trace-id carries an X-Ray-shaped id that no trace exists for.

The stage’s variables, when it has any, reach the handler as event.stageVariables.

A handler returning an object with a statusCode produces that HTTP response. Its headers are sent as headers, its cookies become set-cookie headers, and an empty body is sent as no body, so a 204 stays a valid response.

A handler returning anything else produces a 200 whose body is that value as JSON. That includes an object with no statusCode in it, so a handler returning { body: "hi" } produces a 200 whose body is the JSON {"body":"hi"} with content-type: application/json, which is what real API Gateway does with it.

GetApisCommand, GetIntegrationsCommand, GetRoutesCommand and GetStagesCommand list what an API has. Each answers in full, since paging is not simulated.

/**
* Listing what a simulated HTTP API has.
*/
import {
CreateApiCommand,
CreateStageCommand,
GetApisCommand,
GetStagesCommand,
} from "@aws-sdk/client-apigatewayv2";
import { SimAws } from "@kensio/yulin";
const simAws = new SimAws();
const apiGateway = simAws.apiGatewayV2();
const { ApiId } = await apiGateway.createApi(
new CreateApiCommand({ Name: "orders", ProtocolType: "HTTP" }),
);
await apiGateway.createStage(
new CreateStageCommand({
ApiId,
StageName: "$default",
AutoDeploy: true,
StageVariables: { catalogue: "v2" },
}),
);
const apis = await apiGateway.getApis(new GetApisCommand({}));
console.log(apis.Items.map((api) => api.Name));
const stages = await apiGateway.getStages(new GetStagesCommand({ ApiId }));
console.log(stages.Items[0]?.StageVariables);

DeleteRouteCommand, DeleteIntegrationCommand and DeleteStageCommand each take one resource off an API and leave the rest of it in place. A deleted route stops matching, so a request that used to reach it is answered the way any unmatched request is. A deleted stage stops resolving, so a request addressed to it finds nothing, while the routes it served are still served by the API’s other stages.

An integration a route still points at cannot be deleted. That is a BadRequestException naming the routes in the way, as it is on real AWS, so an API comes apart routes first and then the integrations behind them. Deleting a route does not delete its integration, since an integration outlives the routes pointing at it.

/**
* Deleting a route, an integration and a stage from a simulated HTTP API.
*/
import {
CreateApiCommand,
CreateIntegrationCommand,
CreateRouteCommand,
CreateStageCommand,
DeleteIntegrationCommand,
DeleteRouteCommand,
DeleteStageCommand,
GetIntegrationsCommand,
GetRoutesCommand,
} from "@aws-sdk/client-apigatewayv2";
import { SimAws } from "@kensio/yulin";
const simAws = new SimAws();
const apiGateway = simAws.apiGatewayV2();
const { ApiId } = await apiGateway.createApi(
new CreateApiCommand({ Name: "orders", ProtocolType: "HTTP" }),
);
const { IntegrationId } = await apiGateway.createIntegration(
new CreateIntegrationCommand({
ApiId,
IntegrationType: "AWS_PROXY",
IntegrationUri: "arn:aws:lambda:eu-west-2:111111111111:function:orders",
PayloadFormatVersion: "2.0",
}),
);
const { RouteId } = await apiGateway.createRoute(
new CreateRouteCommand({
ApiId,
RouteKey: "GET /orders",
Target: `integrations/${IntegrationId}`,
}),
);
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "dev", AutoDeploy: true }),
);
// The route comes off first, since the integration cannot be deleted while
// anything still targets it.
await apiGateway.deleteRoute(new DeleteRouteCommand({ ApiId, RouteId }));
await apiGateway.deleteIntegration(
new DeleteIntegrationCommand({ ApiId, IntegrationId }),
);
await apiGateway.deleteStage(
new DeleteStageCommand({ ApiId, StageName: "dev" }),
);
const routes = await apiGateway.getRoutes(new GetRoutesCommand({ ApiId }));
const integrations = await apiGateway.getIntegrations(
new GetIntegrationsCommand({ ApiId }),
);
console.log(routes.Items.length, integrations.Items.length);

DeleteApiCommand deletes the API and everything under it, so taking a whole API away needs none of these.

DisableExecuteApiEndpoint: true stops the generated endpoint serving, which is how an API reachable only through a custom domain is configured. A request to it is answered with a 403 and {"message":"Forbidden"}. AWS publishes neither the status nor the body for that case, so both are what a disabled endpoint was observed to answer rather than something documented.

ImportApiCommand takes a serialised OpenAPI 3.0 document and creates the API, one route and one integration per operation, and one authorizer per security scheme an operation names. The route key is the operation key uppercased and the path taken verbatim, since OpenAPI path templating is already API Gateway’s path parameter syntax.

An import creates no stage. CreateStageCommand or an AWS::ApiGatewayV2::Stage is still declared separately, and an imported API with no stage answers 404.

/**
* Creating a simulated HTTP API from an OpenAPI 3 definition.
*/
import {
CreateStageCommand,
GetRoutesCommand,
ImportApiCommand,
} from "@aws-sdk/client-apigatewayv2";
import {
AddPermissionCommand,
CreateFunctionCommand,
} from "@aws-sdk/client-lambda";
import { SimAws } from "@kensio/yulin";
import type { SimPayload2Event } from "@kensio/yulin/apigatewayv2";
import { makeLambdaZipFileInput } from "@kensio/yulin/lambda";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const { FunctionArn } = await simAws.lambda().createFunction(
new CreateFunctionCommand({
FunctionName: "orders",
Role: "arn:aws:iam::111111111111:role/OrdersRole",
Code: {
ZipFile: makeLambdaZipFileInput((event: SimPayload2Event) => ({
statusCode: 200,
headers: { "content-type": "text/plain" },
body: `order ${event.pathParameters?.["orderId"] ?? "none"}`,
})),
},
}),
);
const openApi = {
openapi: "3.0.1",
info: { title: "orders", version: "1.0" },
paths: {
"/orders/{orderId}": {
get: {
// Ignored, as on AWS: HTTP APIs do no request validation.
responses: { "200": { description: "200 response" } },
"x-amazon-apigateway-integration": {
type: "aws_proxy",
httpMethod: "POST",
uri:
`arn:aws:apigateway:us-east-1:lambda:path/2015-03-31/functions/` +
`${FunctionArn}/invocations`,
payloadFormatVersion: "2.0",
},
},
},
},
};
const apiGateway = simAws.apiGatewayV2();
const { ApiId, ApiEndpoint } = await apiGateway.importApi(
new ImportApiCommand({ Body: JSON.stringify(openApi) }),
);
const routes = await apiGateway.getRoutes(new GetRoutesCommand({ ApiId }));
console.log(routes.Items[0]?.RouteKey); // "GET /orders/{orderId}"
// An import creates no stage, so the API answers 404 until one is created.
await apiGateway.createStage(
new CreateStageCommand({ ApiId, StageName: "$default", AutoDeploy: true }),
);
await simAws.lambda().addPermission(
new AddPermissionCommand({
FunctionName: "orders",
StatementId: "api-gateway-invoke",
Action: "lambda:InvokeFunction",
Principal: "apigateway.amazonaws.com",
SourceArn: `arn:aws:execute-api:us-east-1:888888888888:${ApiId}/*/*`,
}),
);
const srv = await serveSimAws({ simAws });
const response = await fetch(srv.localUrl(`${ApiEndpoint}/orders/42`));
console.log(await response.text()); // "order 42"
await srv.close();

The API is named by info.title. uri is read as either the long arn:aws:apigateway:<region>:lambda:path/2015-03-31/functions/<function-arn>/invocations form above or as the bare function ARN.

AWS sorts what an import finds into three categories, and the third is valid OpenAPI that HTTP APIs do not support. It is ignored silently, and so is it here: requestBody, the content schemas under responses, components.schemas, and an operation’s parameters, summary, description and tags. HTTP APIs perform no request validation, so a request whose body contradicts a declared schema still reaches the handler.

Everything else the document carries and this simulation cannot apply is refused, naming the JSON pointer of the member, such as #/paths/~1orders~1{orderId}/get/x-amazon-apigateway-integration/payloadFormatVersion.

An operation’s x-amazon-apigateway-integration may be a reference into components.x-amazon-apigateway-integrations:

{
"get": {
"x-amazon-apigateway-integration": {
"$ref": "#/components/x-amazon-apigateway-integrations/orders"
}
}
}

The referenced definition is created once and shared by every operation naming it. A $ref anywhere else is refused, naming the pointer it holds.

A security scheme of type oauth2 carrying an x-amazon-apigateway-authorizer with type: "jwt" becomes a JWT authorizer. The scheme key is the authorizer’s name, one authorizer is created per scheme, and an operation naming it gets AuthorizationType: "JWT" with the requirement’s scope list as its AuthorizationScopes. An operation with no security is open.

{
"components": {
"securitySchemes": {
"pool-authorizer": {
"type": "oauth2",
"x-amazon-apigateway-authorizer": {
"type": "jwt",
"identitySource": "$request.header.Authorization",
"jwtConfiguration": {
"issuer": "https://cognito-idp.us-east-1.amazonaws.com/us-east-1_abc",
"audience": ["3n4b5..."]
}
}
}
}
}
}

A security scheme of type apiKey carrying an x-amazon-apigateway-authorizer with type: "request" becomes a Lambda REQUEST authorizer, and an operation naming it gets AuthorizationType: "CUSTOM". The scheme’s own name and in are read by AWS for the OpenAPI document to be valid, and identitySource is what decides where the values come from, here as there.

{
"components": {
"securitySchemes": {
"session-authorizer": {
"type": "apiKey",
"name": "cookie",
"in": "header",
"x-amazon-apigateway-authorizer": {
"type": "request",
"identitySource": "$request.header.cookie",
"authorizerUri": "arn:aws:lambda:us-east-1:111111111111:function:session-authorizer",
"authorizerPayloadFormatVersion": "2.0",
"enableSimpleResponses": true,
"authorizerResultTtlInSeconds": 300
}
}
}
}
}

A scheme whose type and whose authorizer disagree, such as a jwt authorizer under apiKey, is refused naming the extension’s own type.

identitySource is the comma-separated string a document writes. A JWT authorizer takes one entry and a value carrying more is refused, as more than one IdentitySource is on CreateAuthorizer. A REQUEST authorizer takes every entry the string names, and requires a request to carry all of them.

Simulated CloudFormation deploys AWS::ApiGatewayV2::Api, AWS::ApiGatewayV2::Authorizer, AWS::ApiGatewayV2::Integration, AWS::ApiGatewayV2::Route and AWS::ApiGatewayV2::Stage, so a synthesized or hand-written template produces an API that serves requests.

Ref and Fn::GetAtt return what real CloudFormation returns for each type:

Resource type Ref Fn::GetAtt
Api the API id ApiId, ApiEndpoint
Authorizer the authorizer id AuthorizerId
Integration the integration id IntegrationId
Route the route id RouteId
Stage the stage name none, as AWS documents none

Fn::GetAtt: ["Api", "ApiEndpoint"] is the generated endpoint with no trailing slash and no stage segment, on the real amazonaws.com hostname. CDK’s httpApi.url is built from AWS::URLSuffix instead, which resolves to the local sim-aws.localhost form. Both reach the same served API.

An integration’s IntegrationUri is accepted as the bare Lambda function ARN CDK emits, and as the arn:aws:apigateway:<region>:lambda:path/2015-03-31/functions/<function-arn>/invocations form. A route’s Target is the integrations/<integration-id> string, which CDK builds with Fn::Join over a Ref to the integration.

/**
* Deploying a simulated HTTP API from a CloudFormation template.
*/
import { SimAws } from "@kensio/yulin";
import { serveSimAws } from "@kensio/yulin/serve";
const simAws = new SimAws();
const stack = await simAws.cloudFormation().deployTemplate({
stackName: "orders-stack",
template: {
Resources: {
HandlerRole: {
Type: "AWS::IAM::Role",
Properties: {
RoleName: "orders-role",
AssumeRolePolicyDocument: {
Version: "2012-10-17",
Statement: [
{
Effect: "Allow",
Principal: { Service: "lambda.amazonaws.com" },
Action: "sts:AssumeRole",
},
],
},
},
},
Handler: {
Type: "AWS::Lambda::Function",
Properties: {
FunctionName: "orders",
Role: { "Fn::GetAtt": ["HandlerRole", "Arn"] },
Handler: "index.handler",
Runtime: "nodejs20.x",
Code: {
ZipFile:
"exports.handler = async () => ({ statusCode: 200, body: 'orders' });",
},
},
},
HandlerPermission: {
Type: "AWS::Lambda::Permission",
Properties: {
Action: "lambda:InvokeFunction",
FunctionName: { "Fn::GetAtt": ["Handler", "Arn"] },
Principal: "apigateway.amazonaws.com",
SourceArn: {
"Fn::Join": [
"",
[
"arn:aws:execute-api:us-east-1:111111111111:",
{ Ref: "Api" },
"/*/*",
],
],
},
},
},
Api: {
Type: "AWS::ApiGatewayV2::Api",
Properties: { Name: "orders", ProtocolType: "HTTP" },
},
Stage: {
Type: "AWS::ApiGatewayV2::Stage",
Properties: {
ApiId: { Ref: "Api" },
StageName: "$default",
AutoDeploy: true,
},
},
Integration: {
Type: "AWS::ApiGatewayV2::Integration",
Properties: {
ApiId: { Ref: "Api" },
IntegrationType: "AWS_PROXY",
IntegrationUri: { "Fn::GetAtt": ["Handler", "Arn"] },
PayloadFormatVersion: "2.0",
},
},
Route: {
Type: "AWS::ApiGatewayV2::Route",
Properties: {
ApiId: { Ref: "Api" },
RouteKey: "GET /orders",
AuthorizationType: "NONE",
Target: {
"Fn::Join": ["", ["integrations/", { Ref: "Integration" }]],
},
},
},
},
Outputs: {
ApiEndpoint: { Value: { "Fn::GetAtt": ["Api", "ApiEndpoint"] } },
},
},
});
await stack.waitForDeployComplete();
// https://<api-id>.execute-api.us-east-1.amazonaws.com
const apiEndpoint = stack.outputs.get("ApiEndpoint")?.value as string;
const srv = await serveSimAws({ simAws });
const response = await fetch(srv.localUrl(`${apiEndpoint}/orders`));
console.log(response.status);
console.log(await response.text());
await srv.close();

Every property outside the simulated set is left out of what is created and recorded in stack.ignoredProperties, naming the Resource type, the logical id and the ones this can act on instead. The API, authorizer, integration, route or stage is created either way, so the stack deploys and the record says which of its parts behaves differently to the template. The simulated properties are:

  • Api: Name, ProtocolType, Description, DisableExecuteApiEndpoint, Body, FailOnWarnings
  • Authorizer: ApiId, Name, AuthorizerType, IdentitySource, JwtConfiguration, AuthorizerUri, AuthorizerPayloadFormatVersion, EnableSimpleResponses, AuthorizerResultTtlInSeconds
  • Integration: ApiId, IntegrationType, IntegrationUri, PayloadFormatVersion, Description
  • Route: ApiId, RouteKey, Target, AuthorizationType, AuthorizerId, AuthorizationScopes
  • Stage: ApiId, StageName, AutoDeploy, StageVariables, Description

CDK’s HttpIamAuthorizer deploys too. It emits no AWS::ApiGatewayV2::Authorizer and no AuthorizerId, only AuthorizationType: "AWS_IAM" on the Route, and the deployed route then requires IAM authorization when it is served.

CDK’s HttpJwtAuthorizer and HttpUserPoolAuthorizer both deploy. HttpUserPoolAuthorizer builds its issuer from Fn::GetAtt <UserPool>.ProviderURL, which resolves to the same string the pool’s tokens name as their issuer, so a CDK-declared authorizer and a CDK-deployed pool agree with nothing to configure. Pass the app client explicitly through userPoolClients, because otherwise the authorizer adds a client of its own with CDK’s defaults, and those emit the OAuth properties simulated Cognito refuses.

An Authorizer with AuthorizerType: "REQUEST" deploys as a Lambda authorizer, and a Route with AuthorizationType: "CUSTOM" and a Ref to it is decided by that authorizer’s function. AuthorizerUri is read the same way an integration’s URI is, so the bare function ARN and the arn:aws:apigateway:<region>:lambda:path/2015-03-31/functions/<function-arn>/invocations form both work. The AWS::Lambda::Permission alongside it needs a SourceArn of arn:aws:execute-api:<region>:<account>:<api-id>/authorizers/<authorizer-id>, which is what CDK writes and what the authorizer is invoked under. A function used both as an integration and as an authorizer needs two permissions, as it does on AWS.

CDK’s HttpLambdaAuthorizer deploys when it is given responseTypes: [HttpLambdaResponseType.SIMPLE]. Its default is IAM, which sets AuthorizerPayloadFormatVersion to 1.0, and that authorizer event is not simulated.

const authorizer = new HttpLambdaAuthorizer("SessionAuthorizer", authorizerFn, {
responseTypes: [HttpLambdaResponseType.SIMPLE],
identitySource: ["$request.header.cookie"],
});
httpApi.addRoutes({
path: "/account",
methods: [HttpMethod.GET],
integration: new HttpLambdaIntegration("Account", accountFn),
authorizer,
});

A Route with AuthorizationType: "JWT" or "CUSTOM" whose AuthorizerId does not resolve to an authorizer of that API, or resolves to one of the other kind, fails the stack, naming both. That covers a Ref to a Resource this simulation skipped: a skipped Resource resolves to its own logical ID, and no authorizer has that id.

Api.Body carries an OpenAPI document as an inline JSON object, which CloudFormation resolves Ref and Fn::GetAtt inside as it does anywhere else, so an operation’s integration URI can be an Fn::GetAtt on a function the same stack deploys. The document goes through the same ImportApi translator an SDK caller reaches, so the two produce the same API.

Api:
Type: AWS::ApiGatewayV2::Api
Properties:
Body:
openapi: "3.0.1"
info: { title: orders, version: "1.0" }
paths:
/orders/{orderId}:
get:
x-amazon-apigateway-integration:
type: aws_proxy
httpMethod: POST
uri: !GetAtt Handler.Arn
payloadFormatVersion: "2.0"

Name and ProtocolType are both optional alongside a Body, which is what AWS documents. A ProtocolType that is present has to be HTTP, and a Name that is present names the API instead of the document’s info.title. Description and DisableExecuteApiEndpoint are refused alongside a Body, because ImportApi does not take them and nothing here changes an API after it is created. Each is recorded rather than applied, so the API is created from the document without them.

A template combining an Api with a Body and a separate Route, Integration or Authorizer Resource for that same API fails the stack, naming both logical IDs. The document already declares the API’s parts, and which of the two AWS would keep is not established.

An Api carrying a Policy property is refused with its own message rather than the generic one. AWS has no such property on this Resource type, because an HTTP API has no resource policy, so a template carrying one was written for a REST API.

AWS::ApiGatewayV2::Deployment, DomainName, ApiMapping, VpcLink and the WebSocket-only Model, RouteResponse and IntegrationResponse create nothing, so a template carrying one has that resource skipped rather than deployed.

Every command is authorized by simulated IAM. API Gateway is unusual in what it asks for: the action is the HTTP method of the underlying REST call rather than a name matching the SDK operation, and the resource is the request path rather than an ARN naming a resource type. Creating a route on API a1b2c3d4e5 asks whether the caller may apigateway:POST on arn:aws:apigateway:<region>::/apis/a1b2c3d4e5/routes. Those ARNs carry no Account id, as API Gateway control-plane ARNs leave the Account segment empty.

A policy written the way policies for other services are written matches nothing here, as it matches nothing on real AWS.

An ApiGatewayV2Client can be intercepted, so application code that builds its own client reaches the simulation without being given one. See the SDK interception docs.

  • CreateApi, GetApi, GetApis and DeleteApi, with the API id, the generated endpoint, and the Account and Region scoping a real API has
  • CreateIntegration, GetIntegrations and DeleteIntegration for an AWS_PROXY integration naming a Lambda function, with an integration a route still targets refused rather than deleted
  • CreateRoute, GetRoutes and DeleteRoute, with route keys parsed and validated at creation, requests matched to a route by method, literal segment, {name} parameter, {proxy+} parameter and $default, and a deleted route no longer matching anything
  • Path parameters captured by the matched route, reaching the handler as event.pathParameters
  • CreateAuthorizer, GetAuthorizers and DeleteAuthorizer for a JWT authorizer, and routes protected by one with AuthorizationType: "JWT" and AuthorizationScopes
  • Real RS256 verification of a token against the keys its issuer publishes, with the claims checked against the simulation’s clock, and the accepted claims reaching the handler as event.requestContext.authorizer.jwt
  • Routes protected with AuthorizationType: "AWS_IAM", evaluating execute-api:Invoke against the execute-api ARN of the route being called, for a caller resolved from a SigV4 signature or an x-sim-aws-caller header, and reaching the handler as event.requestContext.authorizer.iam
  • CreateAuthorizer, GetAuthorizers and DeleteAuthorizer for a Lambda REQUEST authorizer, and routes protected by one with AuthorizationType: "CUSTOM", invoking the authorizer’s function with the payload format 2.0 authorizer event, reading a simple response or an IAM policy response, and passing the returned context to the handler as event.requestContext.authorizer.lambda
  • AuthorizerResultTtlInSeconds, holding a decision against the identity source values it was made for and expiring it against the simulation’s clock, with $context.routeKey as an identity source to hold one per route
  • CreateStage, GetStages and DeleteStage for the $default stage and for named stages served under their own path segment, including stage variables
  • Serving the generated endpoint through serveSimAws, invoking the integrated function with a payload format 2.0 event and turning its result back into an HTTP response
  • The invoke permission of an integration’s function and of an authorizer’s, each evaluated against that function’s resource policy with its own AWS:SourceArn
  • DisableExecuteApiEndpoint, refusing requests to the generated endpoint
  • ImportApi for an OpenAPI 3.0 document, creating one route and one integration per operation and one JWT or Lambda REQUEST authorizer per security scheme an operation names
  • Deployment of AWS::ApiGatewayV2::Api, Authorizer, Integration, Route and Stage from a CloudFormation template, including one synthesized by CDK from an HttpApi, and an Api declared as an OpenAPI document through Body
  • Authorization of every command by simulated IAM, against the HTTP method and resource path real API Gateway uses
  • SDK interception of an ApiGatewayV2Client

Current documented limitations:

  • HTTP APIs only. ProtocolType: "WEBSOCKET" is refused.
  • Two of the route selection rules, and the place of the method comparison in the order, are observed rather than published by AWS. See Which route serves a request.
  • A route path naming the same parameter twice, such as GET /pets/{id}/toys/{id}, is refused. This is stricter than AWS is known to be: it was refused because only one of the two captures could reach the handler, not because real API Gateway was seen to refuse it.
  • Deployments are not simulated, so CreateStage requires AutoDeploy: true. A stage without it serves whichever Deployment it was given, which on real AWS is nothing until one is created.
  • RouteSettings, DefaultRouteSettings and AccessLogSettings on a stage are refused, as is any other option CreateStage takes and this one does not.
  • AWS_PROXY is the only integration type, and its URI must name an unqualified Lambda function ARN, written either as that ARN or as the arn:aws:apigateway:<region>:lambda:path/2015-03-31/functions/<function-arn>/invocations form. Both reach the same function, and GetIntegrations answers with the function ARN whichever was written. A version or alias qualifier is refused, since simulated Lambda has no versions. HTTP proxy integrations and AWS service integrations are not simulated.
  • Payload format 1.0 is refused. A handler written for 1.0 reads event fields a 2.0 event does not have, so treating one as the other would pass here and fail on AWS.
  • AuthorizationScopes on an AWS_IAM or CUSTOM route is refused. This is stricter than AWS, which documents route scopes as meaningful only for JWT and ignores them here. Accepting one would let a test assert on a scope restriction that nothing applies.
  • The method and path segments of the ARN an AWS_IAM route is authorized against are inferred rather than documented, and the two callers of that ARN builder fill them differently: this one names the request’s own method and path, while the integration’s invoke permission names the route key template. AWS documents one format for both. See Protecting a route with IAM.
  • A * in a policy resource is uniformly greedy here, so it crosses / boundaries. AWS distinguishes * from */* in some ARN path positions, which makes the simulator more permissive than AWS for a policy relying on that distinction.
  • Only execute-api:Invoke is evaluated. Nothing constrains the action string a policy may name, so a policy can be written with execute-api:ManageConnections or execute-api:InvalidateCache and nothing will ever ask about it.
  • accessKey in the iam block is empty, callerId and userId carry the caller ARN rather than the AIDA/AROA unique id real AWS puts there, and cognitoIdentity and principalOrgId are always null. None of those is available at the simulator’s request boundary.
  • A request signed for another service is refused by the serving boundary before route matching, even on a NONE route, where real API Gateway would ignore the signature. That boundary answers {"Message":"Forbidden"} with a capital M, unlike the API’s own refusals.
  • An unsigned request to an AWS_IAM route is a 403 rather than the 403 with an x-amzn-ErrorType of IncompleteSignatureException real API Gateway was observed to send. The simulator resolves such a request to an anonymous caller and refuses it by ordinary IAM evaluation.
  • HTTP API resource policies are not simulated, because AWS does not have them. A caller from another Account is therefore always refused, since a cross-Account request needs an Allow from the resource side. Assume a Role in the API’s Account instead, which is the route through on AWS as well.
  • AuthorizerPayloadFormatVersion: "2.0" is required on a REQUEST authorizer, and stating nothing is refused too. AWS defaults it to 1.0, which builds a different event and answers a policy against a method ARN, and none of that is built here. CDK defaults HttpLambdaAuthorizer to responseTypes: [HttpLambdaResponseType.IAM], which sets the format to 1.0, so a default new HttpLambdaAuthorizer(...) fails the stack. Pass responseTypes: [HttpLambdaResponseType.SIMPLE] to get an authorizer that deploys.
  • A REQUEST authorizer’s event carries no body and no isBase64Encoded. AWS’s published example of that event carries neither, so an authorizer cannot read the request body.
  • An authorizer function that throws is a 500 rather than a 401. Real Lambda turns a thrown error into a payload carrying errorMessage, and simulated Lambda rejects with the error itself, so returning { "errorMessage": "Unauthorized" } is how an authorizer asks for a 401 here.
  • AuthorizerCredentialsArn is refused on CreateAuthorizer, as is authorizerCredentials in an imported document. On an AWS::ApiGatewayV2::Authorizer it is recorded instead, and the authorizer is created without it. It names a Role API Gateway assumes to invoke the authorizer, and the function’s own resource policy is the whole decision here.
  • AuthorizerResultTtlInSeconds is accepted between 0 and 3600, which is the range AWS accepts, and is refused on an authorizer with no IdentitySource, since there would be nothing to key the held decision on.
  • A held decision is the whole answer, so a policy response is not re-evaluated per route on a cache hit. That is what AWS’s own warning about caching describes, and $context.routeKey as an identity source is the documented way to separate routes.
  • An authorizer that could not answer at all, by throwing or by replying in neither format, is not held. There is no answer to hold, and the next request asks the function again.
  • A REQUEST authorizer requires at least one IdentitySource. An authorizer with none is invoked for every request on AWS, including one carrying nothing, and that is not simulated.
  • A REQUEST authorizer’s identity source is $request.header.<name>, $request.querystring.<name> or $context.routeKey. The rest of $context and all of $stageVariables, which a REQUEST authorizer may also name on AWS, are refused rather than read from nowhere. A JWT authorizer takes one identity source naming something the client sent, so $context.routeKey is refused for it, and a second source is refused rather than partly read.
  • An identity source naming nothing after its prefix is refused, and so is one naming a header that is not an HTTP field name. Either would find nothing on every request, and an invalid header name would fail at request time rather than at the command that configured it.
  • JwtConfiguration.Audience is required. What real API Gateway does with an authorizer that has an empty audience list is not documented, so this is stricter than AWS may be, in the direction that cannot quietly admit an app client.
  • A token with no exp claim is refused. Real Cognito always sets one, and admitting a token that nothing can expire is the divergence worth failing on.
  • token_use is not checked, which is what real API Gateway does, so an ID token passes an authorizer that configures only an audience. See Route scopes, and access tokens versus ID tokens.
  • aws.cognito.signin.user.admin is the only scope any simulated Cognito flow issues, so it is the only satisfiable route scope. Resource servers, custom scopes and the client credentials grant are not simulated.
  • A token invalidated by GlobalSignOut still passes. Real API Gateway knows nothing about the pool’s issued tokens, so consulting them here would refuse a token AWS would accept.
  • The pool’s JWKS is read in process rather than fetched. A pool’s published OpenID configuration names the localhost origin it is served from while its tokens name the real AWS URL, so a discovery client would reject its own issuer’s tokens.
  • One signing key per issuer, no key rotation and no JWKS caching. Real Cognito publishes two keys and rotates between them, so code assuming a single entry passes here and is still wrong on AWS.
  • The 403 body for an unmet route scope, the string rendering of claim values, and scopes being null rather than [] are all what the real endpoint was observed to send rather than anything AWS publishes. Only the one error_description AWS documents is ever sent; every other refusal names the scheme and nothing else. How AWS lays out the www-authenticate parameters around that description is not published either, so they are comma-separated as RFC 6750 writes them.
  • Deleting an authorizer a route still points at leaves that route refusing every request. What real API Gateway does with such a route is not established, so it stays closed rather than falling open.
  • The method and path segments of the source ARN are inferred rather than documented. See Granting the API permission to invoke the function.
  • An integration CredentialsArn, the IAM Role alternative to a resource policy grant, is refused by CreateIntegration. A permission on the function is the only way to admit the invocation.
  • No Update* commands. A route, integration or stage is changed by deleting it and creating it again. DeleteApi deletes everything under the API, as it does on AWS.
  • Custom domain names and API mappings are not simulated. They change what rawPath holds, so an API reached through one behaves differently on AWS from what is served here.
  • No paging. MaxResults and NextToken are refused rather than ignored, and every list command answers in full.
  • CorsConfiguration and Tags are refused, as is the RouteKey/Target quick-create shorthand on CreateApi. Anything else the real commands accept and this one does not is refused by name rather than dropped.
  • AWS::ApiGatewayV2::Api records CorsConfiguration rather than applying it. CORS request handling is not simulated, so a CDK stack using corsPreflight deploys and its API answers preflight requests here differently from AWS. The record is where a test checks that.
  • Only OpenAPI 3.0.x is imported. A swagger: "2.0" document and an openapi: "3.1.0" one are both refused by version, and only JSON is parsed, not YAML.
  • FailOnWarnings is honoured only in its strict sense. Everything an import cannot apply is refused rather than warned about, so true is accepted and has no further effect and false is refused by name. What AWS defaults the property to is not established, so nothing here relies on a default.
  • Basepath on ImportApi, BasePath on AWS::ApiGatewayV2::Api and servers in the document are all refused. A base path changes the path every route matches on, which belongs with custom domain names.
  • ReimportApi is not simulated, as no Update* command is. Delete the API and import again.
  • An imported operationId is dropped rather than stored. AWS maps it to the route’s OperationName, and no command here takes one.
  • A trace operation, a path item $ref, x-amazon-apigateway-any-method and a document-level security are each refused by name. None of the four is established for HTTP APIs by the research behind this, so each is refused rather than turned into a route the API may not have on AWS.
  • An operation carrying more than one security requirement is refused, as is a requirement naming more than one scheme. A route has one authorizer.
  • A security scheme that is not oauth2 with an explicit jwtConfiguration.issuer is refused. That includes openIdConnect, where AWS reads the issuer out of the discovery document at openIdConnectUrl: nothing here is fetched over HTTP, and taking the URL as the issuer would mismatch every token’s iss and answer a silent 401.
  • http_proxy integrations, integrationMethod, integrationSubtype, requestParameters, credentials, tlsConfig, responseTransferMode, connectionId and connectionType in an imported integration are all refused by name, as their AWS::ApiGatewayV2::Integration counterparts are.
  • x-amazon-apigateway-cors is refused, alongside the CorsConfiguration refusal above.
  • A referenced x-amazon-apigateway-integrations definition becomes one shared integration, so two operations naming it produce one entry in GetIntegrations. Whether AWS shares one or creates one per use is not established; this is what a reusable definition reads as.
  • A Name on an AWS::ApiGatewayV2::Api with a Body names the API rather than the document’s info.title. Which of the two AWS takes when both are present is not established, and it affects only the name GetApi reports.
  • A terminal {proxy+} in an imported path reaches CreateRoute unchanged. Greedy segments are established for route keys rather than for OpenAPI path templating.
  • AWS::ApiGatewayV2::Api records BodyS3Location rather than reading it, so the API is created with no routes at all. Reading a document out of a simulated S3 bucket adds a fetch path and nothing about OpenAPI.
  • AWS::ApiGatewayV2::Api refuses Policy with a message of its own saying an HTTP API has no resource policy. There is no such property on the real Resource type, so a template carrying one was written for a REST API rather than hitting a gap here.
  • A stack update replaces a changed resource of these types rather than updating it in place, as it does for any other type. See the CloudFormation limitations.
  • Access logging, throttling, usage plans and API keys are not simulated.
  • The response an API Gateway endpoint returns itself uses a lower-case message field, as a real HTTP API does. A Lambda Function URL uses Message for the same thing, so the two are not interchangeable.

Documenting Yulin v1.10.2