Branch | Build Status |
---|---|
main | |
val | |
production |
MFP is the CMCS MDCT application for collecting state data related to the Money Follows the Person (MFP) program. The collected data assists CMCS in monitoring and managing grantee progress and identifying challenges and improvement opportunities.
The MFP demonstration supports state efforts for rebalancing their long-term services and supports system so that individuals have a choice of where they live and receive services. From the start of the program in 2008 through the end of 2020, states have transitioned over 107,000 people to community living under MFP.
Team members are encouraged to setup all MDCT Products using the script located in the MDCT Tools Repository. Please refer to the README for instructions running the MDCT Workspace Setup. After Running workspace setup team members can refer to the Running the project locally section below to proceed with running the application.
If you have run the MDCT Setup Script this section can be skipped
Before starting the project we're going to install some tools. We recommend having Homebrew installed if you haven't already to install other dependencies. Open up terminal on your mac and run the following:
-
(Optional) Install Homebrew:
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
-
Install nvm:
curl -o- https://raw.githubusercontent.com/nvm-sh/nvm/v0.37.2/install.sh | bash
-
Install specified version of node. We enforce using a specific version of node, specified in the file
.nvmrc
. This version matches the Lambda runtime. We recommend managing node versions using NVM:nvm install
, thennvm use
-
Install Serverless:
npm install -g serverless
-
Install yarn:
brew install yarn
-
Install pre-commit on your machine with either:
pip install pre-commit
orbrew install pre-commit
If you have run the MDCT Setup Script this section can be skipped
- Clone the repo:
git clone https://github.com/Enterprise-CMCS/macpro-mdct-mfp.git
- Ensure you either have a 1Password account and have 1Password CLI installed. Alternatively, reach out to the team for an example of .env files
- In the root directory run
pre-commit install
In the root of the project run ./run local --update-env
to pull in values from 1Password and run the project. Alternatively, if you do not have a 1Password account you will need to reach out to an MDCT team member for values for your .env
. Then you can run ./run local
to use a static manually populated .env
file.
(Make sure you've finished setting up the project locally above before moving on to this step!)
Once you've run ./run local
you'll find yourself on a login page at localhost:3000. For local development there is a list of users that can be found at services/ui-auth/libs/users.json. That's where you can grab an email to fill in.
For a password to that user, please ask a fellow developer.
In order to run DynamoDB locally you will need to have java installed on your system. The MDCT Workspace Setup script installs that for you.
If you cannot run the MDCT Workspace setup script see below for manual instructions:
M1 Mac users can download java from azul. Note that you'll need the x86 architecture Java for this to work. You can verify the installation with java --version
. Otherwise install java from here.
To view your database after the application is up and running you can install the dynamodb-admin tool.
- Install and run
DYNAMO_ENDPOINT=http://localhost:8000 dynamodb-admin
in a new terminal window
If you're getting an error such as inaccessible host: 'localhost' at port '8000'
, some steps to try:
- confirm that you're on the right Java version -- if you have an M1 mac, you need an x86 install
- delete your
services/database/.dynamodb
directory and then rundev local
in your terminal
Local dev is configured as a Typescript project. The entrypoint in ./src/run.ts
manages running the moving pieces locally: the API, database, filestore, and frontend.
Local dev is built around the Serverless plugin serverless-offline. serverless-offline
runs an API Gateway locally configured by ./services/app-api/serverless.yml
and hot reloads your Lambdas on every save. The plugins serverless-dynamodb-local and serverless-s3-local stand up the local database and s3 in a similar fashion.
Local authorization bypasses Cognito. The frontend mimics login in local storage with a mock user and sends an id in the cognito-identity-id
header on every request. serverless-offline
expects that and sets it as the cognitoId in the requestContext for your lambdas, just like Cognito would in AWS.
The postman folder contains a full API collection and environment for this application. See the README for more information.
We use Jest for unit tests.
To run all frontend unit tests:
cd services/ui-src/
yarn test
To run all backend unit tests:
cd services/app-api/
yarn test
In either of these directories you can also check code coverage with
yarn coverage
Live reload all tests
yarn test --watch
We use Cypress for integration tests. See additional info here in the Cypress readme
We use axe and pa11y for primary accessibility testing.
Unit tests can use jest-axe, pa11y, and HTML Code Sniffer.
Integration tests can use cypress-axe and cypress-audit/pa11y.
We use Prettier to format all code. This runs as part of a Git Hook and invalid formats in changed files will cause the deploy to fail. If you followed the instructions above this is already installed and configured.
Most IDEs have a Prettier plugin that can be configured to run on file save. You can also run the format check manually from the IDE or by invoking Prettier on the command line.
npx prettier --write "**/*.tsx"
All changed files will also be checked for formatting via the pre-commit hook.
ESLint works in a similar manner for all code linting.
On a push to the repository or opening a pull request the deploy.yml file runs. This script sets up and does a number of things. For a simple push it's mostly checking code coverage.
Upon opening a pull request into the main branch the scripts will also trigger a Cypress E2E and an A11y step to ensure that the code quality is still passing the End-to-End and accessibility tests.
This application is built and deployed via GitHub Actions.
Please Note: Do Not Squash Your Merge Into Val Or Prod When Submitting Your Pull Request.
We have 3 main branches that we work out of:
- Main (Pointed to https://mdctmfpdev.cms.gov/) is our development branch
- Val (Pointed to https://mdctmfpval.cms.gov/) is our beta branch
- Production (Pointed to http://mdctmfp.cms.gov/) is our release branch
When a pull request is approved and merged into main the deploy script will spin up and upon completion will deploy to https://mdctmfpdev.cms.gov/. If a user wants to deploy to val they simply need to create a pull request where Main is being merged into Val. Once that pull request is approved, the deploy script will run again and upon completion will deploy to https://mdctmfpval.cms.gov/. So to quickly break it down:
- Submit pull request of your code to Main.
- Approve pull request and merge into main.
- Deploy script runs and will deploy to https://mdctmfpdev.cms.gov/.
- Submit pull request pointing Main into Val.
- Approve pull request and DO NOT SQUASH YOUR MERGE, just merge it into Val
- Deploy script runs and will deploy to https://mdctmfpval.cms.gov/.
- Submit pull request pointing Val into Production.
- Approve pull request and DO NOT SQUASH YOUR MERGE, just merge it into Production
- Deploy script runs and will deploy to http://mdctmfp.cms.gov/.
If you have a PR that needs Product/Design input, the easiest way to get it to them is to use the cloudfront site from Github. Go to your PR and the Checks
tab, then Deploy
tab. click on deploy
, then click to exapnd the deploy
section on the right. Search for Application endpoint
and click on the generated site.
MFP pipes updates from fieldData and the report object tables to BigMac for downstream consumption. To add a topic for a new report type, update the following locations:
services/app-api/serverless.yaml
- Add table streams to postKafkaData's event triggers
- Declare another lambda to listen to events from the relevant s3 buckets. The same handler file can be used, but serverless has a limitation of 1 existing bucket per lambda.
services/app-api/handlers/kafka/post/postKafkaData.ts
- Add the bucket and table names into the appropriate arrays. They will be parsed with their event types accordingly.services/topics/createTopics.js
- Declare the new topic names. Both the stream name for the bucket and table should be added here.
General Structure - React frontend that renders a form for a user to fill out, Node backend that uses S3 and Dynamo to store and validate forms.
Custom JSON & form field creation engine (formFieldFactory) - Each report has a custom JSON object, stored in a JSON file, written using a custom schema. This JSON object is referred to as the form template and it is the blueprint from which report form fields are created. It is also used to create routes and navigation elements throughout the app. When provided form fields from this template, the formFieldFactory renders the appropriate form fields. A similar process occurs when a report is exported in PDF preview format.
Page and Form Structure Each page has a name, path, and pageType, for example the first page a user sees in the form will be have ‘pageType: standard’ with a ‘verbiage’ object that includes all of the text that precedes the form fields. The the ‘form’ object follows with a unique id and ‘fields’ array that holds one or more objects that represent the individual questions in a form. There are different types of forms as well. If there is a "pageType": "modalDrawer", then instead of a ‘form’ object, it will have a ‘modalForm’ object. Here is an example of a standard page with one field:
{
"name": "Standard Page",
"path": "/standard-page",
"pageType": "standard",
"verbiage": {
"intro": {
"section": "Section I: Standard Page",
}
},
"form": {
"id": "abc",
"fields": [
{
"id": "textFieldId",
"type": "text",
"validation": "text",
"props": {
"label": "field label",
"hint": "Field hint.",
}
},
]
}
},
Storage and retrieval of fieldData When a report is created, the fieldData is stored alongside it in an S3 bucket and reference to that fieldData’s location is stored in report metadata in Dynamo. FieldData is a large object whose structure has all of the non-entity-related data (fields that apply to the entire report) stored at the root level and all entity-related data (fields that are answered once per entity) is stored in an array of entity data objects, as shown below.
const fieldData = {
// non-entity-related data
textFieldId: "textFieldValue",
...
// entity-related data
entityName: [
{
id: "entity1Id",
name: "entity1",
otherField: "otherFieldValue",
...
},
{
id: "entity2Id",
name: "entity2",
otherField: "otherFieldValue",
...
},
]
}
Dropdown and dynamic fields are not currently supported as nested child fields. All other field types are.
Storage and retrieval of the form template When a report is created, the form template is stored alongside it in an S3 bucket and reference to that form template’s location is stored in report metadata in Dynamo. This ensures that future changes to the form template do not break existing forms. However, it also means that changes to the form template are generally only forward looking unless an ETL operation is undertaken.
Field ids Field ids are immutable, or should at least be treated that way. They should be descriptive of the data captured and should never change so that they can be relied on and referenced by downstream data analysts.
Choice ids Fields which accept a list of choices (radio, checkbox, dropdown) require choices with unique, immutable ids. These ids must remain immutable even across versions of the form template to ensure they can be relied on and referenced by downstream data analysts. We have chosen to manually generate these ids.
Nuanced behaviors like the “-otherText” flag on a question’s id Most of the structure of the form template schema is captured in the types contained in types/index.tsx however there are some behaviors like the otherText flag that are not. For example, when a report is exported to PDF, subquestions like nested optional text area fields for the purpose of providing additional information must have ids that end in -otherText or they will not render the entered answer correctly.
Form We use react-hook-form for form state management. The formFieldFactory renders individual field inputs and registers them with RHF which exposes an onSubmit callback hook that is used to check error states and display inline validation messaging.
Form Hydration Any time data is stored in Dynamo or S3 we also pull the latest field data and update the DOM with it through the reportProvider/reportContext. This uses the form hydration engine to ensure that the latest data is shown to the user whether that data comes from the database or the user’s entered but as-of-yet unsaved input.
Validation We use yup for data schema validation. In the form template each field is assigned a validation type corresponding to a custom validation type defined using yup as a baseline. A version of this validation schema exists on the frontend and the backend. While not identical, they are similar and updates to one should often be made to the other. Frontend validation schema is primarily used for inline validation and backend validation schema is primarily used for pre-submission validation. When a form field’s validation type is read, it is matched to the appropriate validation schema.
Server-side validation Anytime an API call to write data is triggered, the unvalidated payload is first validated using a custom yup validation method. The schema used for validation varies depending on the data being written. If the data being written is field data, the validation schema is retrieved from the associated fetched form template. Other metadata has a locally stored longterm validation schema that is used. If the data is valid, the operation continues; if the data is invalid, the operation fails and returns an error.
CustomHTML parser - function checks if element is a string, if so then the element will be passed in the function “sanitize” from "dompurify", and then the result from that process gets passed into the function “parse” from "html-react-parser" and the result gets returned. If the element is not a string, then the elements are treated as an array and get mapped over returning a key, as, and spread the props. The last check is in this else block, checking whether the element is ‘html’, in which case the content will get passed through ‘sanitize’ and ‘parse’ and the ‘as’ prop gets deleted before returning the modified element type, element props, and content.
Report metadata table in Dynamo vs field data in S3 - When a user creates a report, metadata related to the report is stored in Dynamo, along with a unique ID corresponding to the associated form template (stored in another Dynamo table) and another unique ID corresponding to the associated field data (stored in S3 bucket). We decided to store the programs in S3 because these data can get so large that we can’t reliably store it all in Dynamo, nor search through them without the app breaking.
This repository uses 3 webhooks to publish to 3 different channels all in CMS Slack.
-
SLACK_WEBHOOK: This pubishes to the
macpro-mdct-mfp-alerts
channel. Alerts published there are for deploy or test failures to themain
,val
, orproduction
branches. -
INTEGRATIONS_SLACK_WEBHOOK: This is used to publish new pull requests to the
mdct-integrations-channel
-
PROD_RELEASE_SLACK_WEBHOOK: This is used to publish to the
mdct-prod-releases
channel upon successful release of MFP to production.- Webhooks are created by CMS tickets, populated into GitHub Secrets
- Secrets are added to GitHub secrets by GitHub Admins
- Development secrets are maintained in a 1Password vault
See LICENSE for full details.
As a work of the United States Government, this project is
in the public domain within the United States.
Additionally, we waive copyright and related rights in the
work worldwide through the CC0 1.0 Universal public domain dedication.