The MicroApps project enables rapidly deploying many web apps to AWS on a single shared host name, fronted by a CloudFront Distribution, serving static assets from an S3 Bucket, and routing application requests via API Gateway. MicroApps is delivered as a CDK Construct for deployment, although alternative deployment methods can be used if desired and implemented.
MicroApps allows many versions of an application to be deployed either as ephemeral deploys (e.g. for pull request builds) or as semi-permanent deploys. The microapps-router
Lambda function handled routing requests to apps to the current version targeted for a particular application start request using rules as complex as one is interested in implementing (e.g. A/B testing integration, canary releases, per-user rules for logged in users, per-group, per-deparment, and default rules).
Users start applications via a URL such as [/{prefix}]/{appname}/
, which hits the microapps-router
Lambda@Edge OriginRequest handler that looks up the version of the application to be run, and either forwards the request to the target Lambda Function URL (--startupType direct
invoke mode) or returns a transparent iframe
(--startupType iframe
) with a link to that version. direct
mode works with frameworks, like Next.js, that can return pages that have build-time computed relative URLs to static resources and API calls. iframe
mode works with frameworks that do not write computed relative URLs at build time and/or that do not use URLs that are completely relative to wherever the applications is rooted at runtime; this mode is primarily for quick prototyping as it has other complications (such as indirect access to query strings). The URL seen by the user in the browser (and available for bookmarking) has no version in it, so subsequent launches (e.g. the next day or just in another tab) will lookup the version again. All relative URL API requests (e.g. some/api/path
) will go to the corresponding API version that matches the version of the loaded static files, eliminating issues of incompatibility between static files and API deployments.
For development / testing purposes only, each version of an applicaton can be accessed directly via a URL of the patterns [/{prefix}]/{appname}?appver={semver}
for direct
mode or [/{prefix}]/{appname}/{semver}/
for iframe
mode. These "versioned" URLs are not intended to be advertised to end users as they would cause a user to be stuck on a particular version of the app if the URL was bookmarked. Note that the system does not limit access to particular versions of an application, as of 2023-03-04, but that can be added as a feature.
- Overview
- Why MicroApps
- Request Routing for Static Assets / App - Diagram
- Request Dispatch Model for Multi-Account Deployments
- Video Preview of the Deploying CDK Construct
- Installation / CDK Constructs
- Tutorial - Bootstrapping a Deploy
- Limitations / Future Development
- Related Projects / Components
- Why Lambda @ Origin and Not Lambda @ Edge for Apps
- Architecure Diagram
- Project Layout
- Creating a MicroApp Using Zip Lambda Functions
- Creating a MicroApp Using Docker Lambda Functions
- Troubleshooting
MicroApps are like micro services, but for Web UIs. A MicroApp allows a single functional site to be developed by many independent teams within an organization. Teams must coordinate deployments and agree upon one implementation technology and framework when building a monolithic, or even a monorepo, web application.
Teams using MicroApps can deploy independently of each other with coordination being required only at points of intentional integration (e.g. adding a feature to pass context from one MicroApp to another or coordination of a major feature release to users) and sharing UI styles, if desired (it is possible to build styles that look the same across many different UI frameworks).
MicroApps also allow each team to use a UI framework and backend language that is most appropriate for their solving their business problem. Not every app has to use React or Next.js or even Node on the backend, but instead they can use whatever framework they want and Java, Go, C#, Python, etc. for UI API calls.
For internal sites, or logged-in-customer sites, different tools or products can be hosted in entirely independent MicroApps. A menuing system / toolbar application can be created as a MicroApp and that menu app can open the apps in the system within a transparent iframe. For externally facing sites, such as for an e-commerce site, it is possible to have a MicroApp serving /product/...
, another serving /search/...
, another serving /
, etc.
Note: requests can also be dispatched into the same account, but this model is more likely to be used by organizations with many AWS accounts.
npm i --save-dev @pwrdrvr/microapps-cdk
- Add
MicroApps
construct to your stack - The
MicroApps
construct does a "turn-key" deployment complete with the Release app -
Construct Hub
- CDK API docs
- Python, DotNet, Java, JS/TS installation instructions
-
git clone https://github.com/pwrdrvr/microapps-core.git
- Note: the repo is only being for the example CDK Stack, it is not necessary to clone the repo when used in a custom CDK Stack
cd microapps-core
-
npm i -g aws-cdk
- Install AWS CDK v2 CLI
-
asp [my-sso-profile-name]
- Using the
aws
plugin fromoh-my-zsh
for AWS SSO - Of course, there are other methods of setting env vars
- Using the
-
aws sso login
- Establish an AWS SSO session
-
export AWS_REGION=us-east-2
- Region needs to be set for the Lambda invoke - This can be done other ways in
~/.aws/config
as well
- Region needs to be set for the Lambda invoke - This can be done other ways in
-
./deploy.sh
- Deploys the CDK Stack
- Essentially runs two commands along with extraction of outputs:
npx cdk deploy --context @pwrdrvr/microapps:deployReleaseApp=true microapps-basic
npx pwrdrvr publish --app-name release --new-version ${RELEASE_APP_PACKAGE_VERSION} --deployer-lambda-name ${DEPLOYER_LAMBDA_NAME} --app-lambda-name ${RELEASE_APP_LAMBDA_NAME} --static-assets-path node_modules/@pwrdrvr/microapps-app-release-cdk/lib/static_files/release/${RELEASE_APP_PACKAGE_VERSION}/ --overwrite --no-cache
- URL will be printed as last output
- AWS Only
- For the time being this has only been implemented for AWS technologies and APIs
- It is possible that Azure and GCP have sufficient support to enable porting the framework
- CDK would have to be replaced as well (unless it's made available for Azure and GCP in the near future)
- Release Rules
- Currently only a Default rule is supported
- Need to evaluate if a generic implementation can be made, possibly allowing plugins or webhooks to support arbitrary rules
- If not possible to make it perfectly generic, consider providing a more complete reference implementation of examples
- Release App
- The Release app is an initial, rudimentary, release control console for setting the default version of an application
- Built with Next.js
- pwrdrvr/microapps-app-release
- Next.js Demo App
- The Next.js Tutorial application deployed as a MicroApp
- pwrdrvr/serverless-nextjs-demo
Calling resources (DBs and other services) and waiting for a long synchronous response is an anti-pattern in Lambda as the Lambda function will be billed for the time spent waiting for the response. This is especially true for Lambda@Edge as the cost is 3x higher than Lambda at the origin.
With Lambda@Edge (even with Origin Requests) the cost is 3x higher per GB-second and the time spent waiting for a 1 ms service response from an origin that is 250 ms away is 750x higher (250 ms / 1ms * 3x higher cost) than making that same request within the region where the resource resides.
- Lambda@Edge is at least 3x more expensive than Lambda at the origin:
- In US East 1, the price per GB-Second is $0.00005001 for Lambda@Edge
- Source: https://aws.amazon.com/lambda/pricing/ (bottom of page)
- Updated: 2023-03-04
- In US East 1, the price per GB-Second is $0.0000166667 for Lambda at the origin on x86
- Source: https://aws.amazon.com/lambda/pricing/
- Updated: 2023-03-04
- Ratio
- Lambda@Edge / Lambda@Origin = $0.00005001 / $0.0000166667 = 3.0006x
- In US East 1, the price per GB-Second is $0.00005001 for Lambda@Edge
- Any DB or services calls from Lambda@Edge back to the origin will pay that 3x higher per GB-Second cost for any time spent waiting to send the request and get a response. Example:
- Lambda@Edge
- 0.250s Round Trip Time (RTT) for EU-zone edge request to hit US-East 1 Origin
- 0.200s DB lookup time
- 0.050s CPU usage to process the DB response
- 0.500s total billed time @ $0.00005001 @ 128 MB
- $0.000003125625 total charge
- Lambda at Origin
- RTT does not apply (it's effectively 1-2 ms to hit a DB in the same region)
- 0.200s DB lookup time
- 0.050s CPU usage to process the DB response
- 0.250s total billed time @ $0.0000166667 @ 128 MB
- Half the billed time of running on Lambda@Edge
- 1/6th the cost of running on Lambda@Edge:
- $0.000000520834375 total charge (assuming no CPU time to process the response)
- $0.000003125625 / $0.000000520834375 = 6x more expensive in Lambda@Edge
- Lambda@Edge
-
packages/cdk
- Example CDK Stack
- Deploys MicroApps CDK stack for the GitHub Workflows
- Can be used as an example of how to use the MicroApps CDK Construct
-
packages/demo-app
- Example app with static resources and a Lambda function
- Does not use any Web UI framework at all
-
packages/microapps-cdk
- MicroApps
- "Turn key" CDK Construct that creates all assets needed for a working MicroApps deployment
- MicroAppsAPIGwy
- Create APIGateway HTTP API
- Creates domain names to point to the edge (Cloudfront) and origin (API Gateway)
- MicroAppsCF
- Creates Cloudfront distribution
- MicroAppsS3
- Creates S3 buckets
- MicroAppsSvcs
- Create DynamoDB table
- Create Deployer Lambda function
- Create Router Lambda function
- MicroApps
-
packages/microapps-datalib
- Installed from
npm
:npm i -g @pwrdrvr/microapps-datalib
- APIs for access to the DynamoDB Table used by the
pwrdrvr
CLI,microapps-deployer
, and@pwrdrvr/microapps-app-release-cdk
- Installed from
-
packages/microapps-deployer
- Lambda service invoked by the
pwrdrvr
CLI to record new app/version in the DynamoDB table, create API Gateway integrations, copy S3 assets from staging to prod bucket, etc. - Returns a temporary S3 token with restricted access to the staging S3 bucket for upload of the static files for one app/semver
- Lambda service invoked by the
-
packages/pwrdrvr
- Installed from
npm
:npm i -g pwrdrvr
- Node executable that updates versions in config files, deploys static assets to the S3 staging bucket, optionally compiles and deploys a new Lambda function version, and invokes
microapps-deployer
- AWS IAM permissions required:
lambda:InvokeFunction
- Installed from
-
packages/microapps-router
- Lambda function that determines which version of an app to point a user to on a particular invocation
[TBC]
Docker Lambdas are great for large applications. These used to be slower to cold start but as of early 2023 that appears to no longer be an issue.
Create a Next.js app then follow the steps in this section to set it up for publishing to AWS Lambda @ Origin as a MicroApp. To publish new versions of the app use npx pwrdrvr --new-version x.y.z
when logged in to the target AWS account.
Replace the version with 0.0.0
so it can be modified by the pwrdrvr
CLI.
npm i --save-dev pwrdrvr
FIXME: Out of date 2023-03-04
Add this file to the root of the app.
FROM node:15-slim as base
WORKDIR /app
# Download the sharp libs once to save time
# Do this before copying anything else in
RUN mkdir -p image-lambda-npms && \
cd image-lambda-npms && npm i sharp && \
rm -rf node_modules/sharp/vendor/*/include/
# Copy in the build output from `npx serverless`
COPY .serverless_nextjs .
COPY config.json .
# Move the sharp libs into place
RUN rm -rf image-lambda/node_modules/ && \
mv image-lambda-npms/node_modules image-labmda/ && \
rm -rf image-lambda-npms
FROM public.ecr.aws/lambda/nodejs:14 AS final
# Copy in the munged code
COPY --from=base /app .
CMD [ "./index.handler" ]
FIXME: Out of date 2023-03-04
Add this file to the root of the app.
Replace appname
with your URL path-compatible application name.
const appRoot = '/appname/0.0.0';
// eslint-disable-next-line no-undef
module.exports = {
target: 'serverless',
webpack: (config, _options) => {
return config;
},
basePath: appRoot,
publicRuntimeConfig: {
// Will be available on both server and client
staticFolder: appRoot,
},
};
Requests to the API Gateway origin can be rejected with a 403 Forbidden error if the signed request headers are not sent to the origin by CloudFront.
The error in the API Gateway CloudWatch logs will show up as:
"authorizerError": "The request for the IAM Authorizer doesn't match the format that API Gateway expects."
This can be simulated by simply running curl [api-gateway-url]
, with no headers.
To confirm that API Gateway is allowing signed requests when the IAM Authorizer is configured, establish credentials as a user that is allowed to execute the API Gateay, install awscurl
with pip3 install awscurl
, then then use awscurl --service execute-api --region [api-gateway-region] [api-gateway-url]
.
Signature headers will not be sent from CloudFront to API Gateway unless the OriginRequestPolicy
is set to specifically include those headers on requests to the origin, or the headersBehavior
is set to cfront.OriginRequestHeaderBehavior.all()
.
Similarly, if presign
is used, the OriginRequestPolicy
must be set to cfront.OriginRequestQueryStringBehavior.all()
or to specifically forward the query string parameters used by the presigned URL.
authorization
x-amz-date
x-amz-security-token
x-amz-content-sha256