bitblit
Epsilon
A tiny library to simplify serving consistent apis from Lambda with OpenAPI
You may wish to read the changelog
TODO:
- Discuss pros/cons of single lambda for http/batch in this document
- path/query var checking against open api doc
- check compression handling
How better than just using straight Node?
- Uses typescript instead of the Godforsaken straight javascript
- Handles route mapping (multiple ends, single lambda)
- Uses Promises and has a top level .catch to convert to 500
- Adds compression
- Adds CORS
- Adds JWT handling
- Consistent error handling
- Can serve static content as well
- Kinda-persistent objects allow for optimistic caching
- Built in support for Non-HTTP (Batch) processing via SaltMint, SNS, SQS, Email, etc
How better than using Express?
- Doesn't have req/res architecture to fake so much easier to test
- OpenAPI first design
- Much lighter
- Support for backgrounding, CRON, Inter-API messaging, AWS Batch out of the box
Other services
- Environmental service
- Simple redirects
CDK Automatic Construction
Introduction
Prerequisites
- Any time you are doing something automated like this, I HIGHLY RECOMMEND setting up a billing alert first. Don't blame me if you skip this step and then accidentally spend $1,000 next month. (https://docs.aws.amazon.com/AmazonCloudWatch/latest/monitoring/monitor_estimated_charges_with_cloudwatch.html)
- CDK V2 must be bootstrapped in your account (https://docs.aws.amazon.com/cdk/v2/guide/bootstrapping.html) -- e.g. (npx cdk bootstrap aws://1234567890/us-east-1)
- The user running the CDK deploy needs -LOTS- of AWS privs (I'll list them all later. For now, assume a bunch)
- Policy arn:aws:iam::aws:policy/service-role/AWSLambdaVPCAccessExecutionRole must exist (this should be there by default)
- Policy arn:aws:iam::aws:policy/service-role/AWSBatchServiceRole must exist (You may have to create it - see https://docs.aws.amazon.com/batch/latest/userguide/service_IAM_role.html)
- You should have a VPC that your Lambda's will deploy into, and you should know its id (e.g., vpc-123456789)
- That VPC should have one or more subnets, and you should know their ids (e.g., 05966bfadca940a88)
- That VPC should have a security group, which probably should allow outbound traffic if you want your lambda to talk to the internet. You should know its id (e.g., 02a89a55b0f2cb4ae - leave off the sg- prefix)
Api Setup
- Your api layer, in addition to depending on Epsilon, will need: * "aws-cdk-lib": (Peer version) ** "constructs": (Peer version) \* "walk": (Peer version)
Docker setup
.dockerignore
In general the docker file needs to be in a folder above both your api and cdk folders, but it must ignore the cdk folder in that case or it will infinitely recurse
# Need this here to prevent infinite recusion of cdk folder at least
.github
.idea
.yarn
.git
modules/cdk
node_modulesDockerFile
FROM public.ecr.aws/lambda/nodejs:14
COPY modules/api/package.json modules/api/gql-codegen.yml modules/api/tsconfig.json ${LAMBDA_TASK_ROOT}/
COPY modules/api/src ${LAMBDA_TASK_ROOT}/src
COPY lambda-bootstrap-shell.sh ${LAMBDA_TASK_ROOT}
RUN npm install -g yarn
RUN yarn install
RUN yarn clean-compile
ENTRYPOINT ["sh","/var/task/lambda-bootstrap-shell.sh"]
CMD [ "dist/lambda.handler" ]lambda-bootstrap-shell.sh
#!/bin/sh
# Based on the default script in public.ecr.aws/lambda/nodejs:14, which is copyright 2020 Amazon.com, Inc. or its affiliates. All Rights Reserved.
# Modifications copyright Christopher Weiss, 2022
if [ -z "${EPSILON_RUNNING_IN_AWS_BATCH}" ]; then
if [ $# -lt 1 ]; then
echo "entrypoint requires the handler name to be the first argument" 1>&2
exit 142
fi
export _HANDLER="$1"
RUNTIME_ENTRYPOINT=/var/runtime/bootstrap
if [ -z "${AWS_LAMBDA_RUNTIME_API}" ]; then
exec /usr/local/bin/aws-lambda-rie $RUNTIME_ENTRYPOINT
else
exec $RUNTIME_ENTRYPOINT
fi
else
echo "Running Epsilon inside AWS batch - triggering direct $1 $2"
exec node dist/aws-batch-epsilon-cli.ts --process $1 --data $2
fiGraphQL Support (v0.1.x and above)
If you are just doing straight GraphQL then you don't really need to use Epsilon at all (I'd recommend just going with straight https://www.npmjs.com/package/apollo-server-lambda). However, if you want to start messing with GraphQL while maintaining your existing OpenAPI 3.0 endpoints, Epsilon allows you to designate a regular expression for which all matching requests are delegated to a supplied ApolloServer, bypassing Epsilon.
To do this, you must include the following libraries (They aren't marked as dependencies of Epsilon since they aren't required if you don't support GraphQL)
"apollo-server-lambda": "2.8.1",
"graphql": "14.4.2",Then, in your router-config, you must set an ApolloServer and an Apollo Regex:
const typeDefs = gql`
type Query {
hello: String
}
`;
// Provide resolver functions for your schema fields
const resolvers = {
Query: {
hello: () => 'Hello world!',
},
};
const server: ApolloServer = new ApolloServer({ typeDefs, resolvers });
// ...
const cfg: RouterConfig = RouterUtil.openApiYamlToRouterConfig(yamlString, handlers, authorizers, options);
// ...
cfg.apolloServer = server;
cfg.apolloCreateHandlerOptions = {
origin: '*',
credentials: true,
} as CreateHandlerOptions;
cfg.apolloRegex = new RegExp('.*graphql.*');Usage
Using WebHandler to simplify the Lambda
You will configure a RouterConfig, and then create a WebHandler from that. Your lambda function should look like:
const handler: Handler = (event: APIGatewayEvent, context: Context, callback: Callback) => {
const routerConfig: RouterConfig = getMyRouterConfig(); // Implement this function
const commonHandler: WebHandler = new WebHandler(routerConfig);
commonHandler.lambdaHandler(event, context, callback);
};
export {handler};
Using auth/AuthHandler to simplify a JWT token based auth
Your auth lambda should look like this (I here assume you are storing your encryption key in AWS System Manager so you can keep it encrypted at rest, which you definitely should be doing):
import {AuthHandler} from '@bitblit/epsilon/dist/auth/auth-handler';
import {Callback, Context, CustomAuthorizerEvent, Handler} from 'aws-lambda';
import {EnvironmentService} from '@bitblit/ratchet/dist/aws/environment-service';
import 'reflect-metadata';
const handler: Handler = (event: CustomAuthorizerEvent, context: Context, callback: Callback) => {
EnvironmentService.getConfig('MyConfigurationName').then(cfg => {
const commonAuth: AuthHandler = new AuthHandler('api.mycompany.com', cfg['encryptionKey']);
commonAuth.lambdaHandler(event, context, callback);
});
};
export {handler};
This will pass through anyone with a valid JWT token. Note that Epsilon doesn't yet support role based filtering in this version.
To create valid JWT tokens, your authentication endpoint can use the auth/WebTokenManipulator class like so (after you have verified the users principal/credentials pair) :
// Other authentication stuff happens up here.
const email: string = 'user-email@test.com';
const roles: string[] = ['USER','NOT-AN-ADMIN'];
const userData: any = {'other': 'stuff'};
const myConfig: any = await EnvironmentService.getConfig('MyConfigurationName'); // same as above
const encryptionKey: string = cfg['encryptionKey'];
const adminUser: any = null; // Set this if the user is an admin doing run-as (this is the admin user)
const expSec: number = 3600; // How long until this token expires in seconds
const tokenHandler: WebTokenManipulator = new WebTokenManipulator(encryptionKey, 'api.mycompany.com');
const token: string = tokenHandler.createJWTString(email, userData, roles, expSec, admin);
Notes on adding a new gateway/stage
You'll need to auth the gateway to hit the lambda (yes, as of 2018-10-13 this is still ugly) :
aws lambda add-permission --function-name "arn:aws:lambda:us-east-1:{accountId}:function:{lambda-function-name}"
--source-arn "arn:aws:execute-api:us-east-1:{account number}:{api id}/*/*/*"
--principal apigateway.amazonaws.com
--statement-id b57d8a0f-08e5-407c-9093-47d7e8e840bc
--action lambda:InvokeFunction
And you'll need to remember to go to IAM / Keys and authorize the new stack user to use your KMS key (if you are using KMS to encrypt your config via SystemManager, which you should be doing.)