In AWS Cognito, the default method of sending emails and SMS messages is AWS' own

services: SES and SNS correspondingly.

Usually it makes sense, you're already in the AWS ecosystem anyway... But what if you have
a requirement to use a 3rd party ESP (Email Service Provider), like Twilio Sendgrid or
Mailgun/Sendinblue/Mailchimp? Some of the reasons could be:
Designers need easy access to the email templates in the ESP UI
Users need a way to unsubscribe from the emails they receive (subscription status tracked
by the ESP)
Your company relies on analytics tools provided by ESPs which you wouldn't have with SES
The email template is large (html and css-wise) and it doesn't fit into Cognito's 20k
characters limit
You do not want to (additionally) authenticate SES origin (SPF, DKIM, DMARC)
You want to reuse high-reputation IP addresses managed by the ESP
There's a Custom message Lambda trigger (e.g. invoked
by  CustomMessage_ForgotPassword  user action) but it will only allow you to customize the
email subject and body, not change the underlying transport.
Sometime late 2020, AWS added a new type of Lambda trigger to Cognito: Custom
Sender Lambda Triggers.
Cognito documentation is a bit lacking... The code is not copypastable and some steps in
the instructions are missing.
Furthermore it doesn't show how to configure custom email and SMS senders with
infrastructure as code.
Below is a guide to deploying and using these new Cognito Lambda triggers. 
We will see how to deploy them with Cloudformation and Terraform.
Custom sender Lambda triggers
More specifically, there are 2 new triggers:
CustomEmailSender  to override the default (SES) way of sending emails
CustomSMSSender  when you need to use an external service for text messages instead of
SNS
In this post we will focus on the custom email sender Lambda, but the process for custom
SMS sender is identical.
The parameters these Lambda triggers receive from Cognito are a bit different from what
Custom Message Trigger gets. With  CustomMessage_*  triggers no secrets are passed to the
code. Instead, you get a placeholder string that you put in the right place in the email or
SMS to be sent. That placeholder gets replaced with the code by Cognito right before the
notification is sent.
Custom sender triggers on the other hand receive encrypted notification code. Thus we
will need to set up a KMS key - for Cognito to encrypt the codes and for us to decrypt
them in the function code.
How are new triggers supported by the tooling?
Cognito Console doesn't let you configure the triggers yet
Cognito documentation suggests using AWS CLI to configure triggers
CloudFormation docs say the feature is not yet supported. From my tests it worked,
probably the docs are not updated yet
Terraform does not yet support it but there's a workaround
Send Cognito emails with Twilio Sendgrid
For the demo, we will register a new user in Cognito and will make sure the email is sent
with Sendgrid.
Prerequisites to follow along:
AWS account
Sendgrid account (or whichever ESP you're using)
Docker
That's right, no AWS CLI, no Node.js, Docker is the only dev dependency. What a beautiful
time to be a developer in!
Since this is not a tutorial on how to deploy Cognito, I will focus only on parts relevant to
the custom sender triggers.
You can find full code with all resources defined in CloudFormation and Terraform in
the demo repository.
Create KMS key
As mentioned above we need a key to encrypt/decrypt the notification code. The key
policy matches the default permissions for a new KMS key when it is created with AWS
Console.
CloudFormation
Parameters:
...
CallingUserArn:
Description: Calling user ARN
Type: String

Resources:
...
KmsKey:
Type: AWS::KMS::Key
Properties:
Enabled: true
KeyPolicy:
Version: '2012-10-17'
Statement:
- Effect: Allow
Principal:
AWS: !Sub 'arn:aws:iam::${AWS::AccountId}:root'
Action: 'kms:*'
Resource: '*'
 - Effect: Allow
Principal:
AWS: !Ref CallingUserArn
Action:
- "kms:Create*"
- "kms:Describe*"
- "kms:Enable*"
- "kms:List*"
- "kms:Put*"
- "kms:Update*"
- "kms:Revoke*"
- "kms:Disable*"
- "kms:Get*"
- "kms:Delete*"
- "kms:TagResource"
- "kms:UntagResource"
- "kms:ScheduleKeyDeletion"
- "kms:CancelKeyDeletion"
Resource: '*'
CallingUserArn  parameter is a little trick to pass calling IAM user's ARN to CloudFormation:
aws cloudformation deploy ... --parameter-overrides CallingUserArn="$(aws sts get-caller-identity --query Arn --ou

Terraform
data "aws_caller_identity" "current" {}

resource "aws_kms_key" "kms_key" {

description = "KMS key for Cognito Lambda trigger"
policy = <<EOF
{
"Version": "2012-10-17",
"Statement": [
{
"Effect": "Allow",
"Principal": {
"AWS": "arn:aws:iam::${data.aws_caller_identity.current.account_id}:root"
},
"Action": "kms:*",
"Resource": "*"
},
{
"Effect": "Allow",
"Principal": {
"AWS": "${data.aws_caller_identity.current.arn}"
},
"Action": [
"kms:Create*",
"kms:Describe*",
"kms:Enable*",
 "kms:List*",
"kms:Put*",
"kms:Update*",
"kms:Revoke*",
"kms:Disable*",
"kms:Get*",
"kms:Delete*",
"kms:TagResource",
"kms:UntagResource",
"kms:ScheduleKeyDeletion",
"kms:CancelKeyDeletion"
],
"Resource": "*"
}
]
}
EOF
}

Create Lambda IAM role

Besides the standard  AWSLambdaBasicExecutionRole  managed policy, we need to grant Lambda
access to decrypt our KMS key.
Note you probably want to replace  AWSLambdaBasicExecutionRole  with a fine-grained policy so
that it has the least required privileges.
For both CF and TF scripts, the outcome is the same: a role for the Lambda trigger with 2
policies attached.
CloudFormation
Resources:
...
LambdaTriggerRole:
Type: AWS::IAM::Role
Properties:
AssumeRolePolicyDocument:
Version: "2012-10-17"
Statement:
- Effect: Allow
Action: "sts:AssumeRole"
Principal:
Service: "lambda.amazonaws.com"
ManagedPolicyArns:
- arn:aws:iam::aws:policy/service-role/AWSLambdaBasicExecutionRole

LambdaTriggerRoleKmsPolicy:
Type: AWS::IAM::Policy
Properties:
PolicyDocument:
Version: "2012-10-17"
Statement:
 - Effect: "Allow"
Action:
- "kms:Decrypt"
Resource: !GetAtt KmsKey.Arn
PolicyName: "LambdaKmsPolicy"
Roles:
- !Ref LambdaTriggerRole
Terraform
data "aws_iam_policy_document" "AWSLambdaTrustPolicy" {
version = "2012-10-17"
statement {
actions = ["sts:AssumeRole"]
effect = "Allow"
principals {
type = "Service"
identifiers = ["lambda.amazonaws.com"]
}
}
}

resource "aws_iam_role" "iam_role" {

assume_role_policy = data.aws_iam_policy_document.AWSLambdaTrustPolicy.json
name = "${var.project}-iam-role-lambda-trigger"
}

resource "aws_iam_role_policy_attachment" "iam_role_policy_attachment_lambda_basic_execution" {

role = aws_iam_role.iam_role.name
policy_arn = "arn:aws:iam::aws:policy/service-role/AWSLambdaBasicExecutionRole"
}

data "aws_iam_policy_document" "iam_policy_document_lambda_kms" {

version = "2012-10-17"
statement {
actions = ["kms:Decrypt"]
effect = "Allow"
resources = [
aws_kms_key.kms_key.arn
]
}
}

resource "aws_iam_role_policy" "iam_role_policy_lambda_kms" {

name = "${var.project}-iam-role-policy-lambda-kms"
role = aws_iam_role.iam_role.name
policy = data.aws_iam_policy_document.iam_policy_document_lambda_kms.json
}

Create Lambda function (Node.js code)

Unlike the AWS SDK itself,  @aws-crypto/client-node  encryption library has to be packaged
and deployed with the code. If you don't have Node.js installed locally, you can install
dependencies with Docker. Assuming you're in the cloned repo:
cd lambda/

docker run -it --rm -v $(pwd):/var/app node:12 bash

npm i
The function code:
const AWS = require('aws-sdk')
const b64 = require('base64-js')
const encryptionSdk = require('@aws-crypto/client-node')
const sgMail = require("@sendgrid/mail")

const { decrypt } = encryptionSdk.buildClient(encryptionSdk.CommitmentPolicy.REQUIRE_ENCRYPT_ALLOW_DECRY

const keyIds = [process.env.KEY_ID];
const keyring = new encryptionSdk.KmsKeyringNode({ keyIds })

sgMail.setApiKey(process.env.SENDGRID_API_KEY)

exports.handler = async(event) => {

let plainTextCode
if (event.request.code) {
const { plaintext, messageHeader } = await decrypt(keyring, b64.toByteArray(event.request.code))
plainTextCode = plaintext
}

const msg = {
to: event.request.userAttributes.email,
from: "[email protected]",
subject: "Your Cognito code",
text: `Your code: ${plainTextCode.toString()}`,
}

await sgMail.send(msg)
}

It expects the KMS key ARN and ESP API key to be passed as environment variables. The
notification code is decrypted and can be used in the message body sent to the email
provider API for delivery.
Example  event  object passed to the function:
{
"version": "1",
"triggerSource": "CustomEmailSender_ForgotPassword",
"region": "us-east-1",
"userPoolId": "us-east-1_LnS...",
"userName": "54cf7eb7-0b96-4304-...",
 "callerContext": {
"awsSdkVersion": "aws-sdk-nodejs-2.856.0",
"clientId": "6u7c9vr3pkstoog..."
},
"request": {
"type": "customEmailSenderRequestV1",
"code": "AYADeILxywKhhaq8Ys4mh0aHutYAgQACABVhd3MtY3J5c...",
"clientMetadata": null,
"userAttributes": {
"sub": "54cf7eb7-0b96-4304-8d6b-...",
"email_verified": "true",
"cognito:user_status": "CONFIRMED",
"cognito:email_alias": "[email protected]",
"phone_number_verified": "false",
"phone_number": "...",
"given_name": "Max",
"family_name": "Ivanov",
"email": "[email protected]"
}
}
}

Create Lambda function (infra)

We will define a Node.js Lambda that will be triggered by Cognito each time an email
should be sent. It's deployed as a ZIP file. The only part that is unique to the custom
senders trigger is the environment variables.
We need a variable for the KMS key ID and another for the email provider API key.
CloudFormation
Parameters:
...
SendgridApiKey:
Description: Sendgrid API key
Type: String

Resources:
...
LambdaTrigger:
Type: AWS::Lambda::Function
Properties:
Code: "../lambda"
Environment:
Variables:
KEY_ID: !GetAtt KmsKey.Arn
SENDGRID_API_KEY: !Ref SendgridApiKey
FunctionName: !Sub ${ProjectName}-lambda-custom-email-sender-trigger
PackageType: Zip
Role: !GetAtt LambdaTriggerRole.Arn
 Runtime: nodejs12.x
Handler: index.handler
If you're familiar with CloudFormation, there shouldn't be any surprises.
Terraform
data "archive_file" "lambda" {
type = "zip"
source_dir = "../lambda"
output_path = "lambda.zip"
}

resource "aws_lambda_function" "lambda_function_trigger" {

environment {
variables = {
KEY_ID = aws_kms_key.kms_key.arn
SENDGRID_API_KEY = var.sendgrid_api_key
}
}
code_signing_config_arn = ""
description = ""
filename = data.archive_file.lambda.output_path
function_name = "${var.project}-lambda-function-trigger"
role = aws_iam_role.iam_role.arn
handler = "index.handler"
runtime = "nodejs12.x"
source_code_hash = filebase64sha256(data.archive_file.lambda.output_path)
}
Thanks to the  source_code_hash , each time the function code is modified, the resource will be
marked as changed and the code will get redeployed.
Create Cognito User Pool
This is the most confusing part. We need to set the  LambdaConfig  setting of the User Pool. It
is an object storing configuration of Lambda triggers invoked by Cognito. 2 new options
we're interested in are:
CustomEmailSender: { LambdaArn: "...", LambdaVersion: "..." }
KMSKeyID: "..."
The process is straightforward with CloudFormation but requires a workaround in
Terraform. Details below.
CloudFormation
The official docs say setting the custom sender options is Not currently supported by AWS
CloudFormation. But lucky we, that's not the case. It worked perfectly in multiple tests I ran.
Resources:
UserPool:
Type: AWS::Cognito::UserPool
Properties:
AccountRecoverySetting:
RecoveryMechanisms:
 - Name: verified_email
Priority: 1
AutoVerifiedAttributes:
- email
LambdaConfig:
CustomEmailSender:
LambdaArn: !GetAtt LambdaTrigger.Arn
LambdaVersion: "V1_0"
KMSKeyID: !GetAtt KmsKey.Arn
UsernameConfiguration:
CaseSensitive: false
UserPoolName: !Sub ${ProjectName}-user-pool
UsernameAttributes:
- email
Policies:
PasswordPolicy:
MinimumLength: 10
Schema:
- Name: name
AttributeDataType: String
Mutable: true
Required: true
- Name: email
AttributeDataType: String
Mutable: true
Required: true
LambdaConfig:  is the part of the most interest above.
Terraform
There's an open issue to track the feature status in Terraform. But what can we do now,
before it's available?  null_resource  can help to pull this off, but still there are some gotchas.
Unlinke the CloudFormation definition, we don't add anything  CustomEmailSender -related in
the resource definition, so that's your good old Cognito User Pool in Terraform:
resource "aws_cognito_user_pool" "cognito_user_pool" {
name = "${var.project}-cognito-user-pool"

account_recovery_setting {
recovery_mechanism {
name = "verified_email"
priority = 1
}
}

auto_verified_attributes = ["email"]

password_policy {
minimum_length = 10
temporary_password_validity_days = 7
 require_lowercase = false
require_numbers = false
require_symbols = false
require_uppercase = false
}

schema {
attribute_data_type = "String"
developer_only_attribute = false
mutable = true
name = "email"
required = true

string_attribute_constraints {
max_length = "2048"
min_length = "0"
}
}

schema {
attribute_data_type = "String"
developer_only_attribute = false
mutable = true
name = "name"
required = true

string_attribute_constraints {
max_length = "2048"
min_length = "0"
}
}

username_attributes = ["email"]
username_configuration {
case_sensitive = false
}
}
In order to set the Lambda configuration in the user pool, we will use the  aws cognito-idp
update-user-pool --lambda-config "CustomEmailSender={LambdaVersion=V1_0,LambdaArn=... AWS CLI
command.
The problem is, if you don't pass all the other relevant pool options to this command, they
will be reset to the default values. Suggested solution:
1. Deploy Terraform stack without setting the lambda config
2. Generate a skeleton of the input variables expected by the  update-user-pool command:
aws cognito-idp update-user-pool --user-pool-id us-east-1_evzTb... --generate-cli-skeleton input
3. Fetch the current configuration of the user pool:
aws cognito-idp describe-user-pool --user-pool-id us-east-1_evzTb... --query UserPool > input.json
4. From the fetched config, remove the keys that are not listed in the skeleton. Only
configuration options accepted by the  update-user-pool  must be left. One can probably
come up with a script to do this automatically... but I edited it manually.
5. Add and deploy the new  null_resource  with Terraform.
locals {
update_user_pool_command = "aws cognito-idp update-user-pool --user-pool-id ${aws_cognito_user_pool.cog
}

resource "null_resource" "cognito_user_pool_lambda_config" {

provisioner "local-exec" {
command = local.update_user_pool_command
}
depends_on = [local.update_user_pool_command]
triggers = {
input_json = filemd5(var.update_user_pool_config_file)
update_user_pool_command = local.update_user_pool_command
}
}

A quick comment on what's happenning here. We define a local value with the  update-user-
pool  command. It accepts the user pool ID, the JSON file with current user pool
configuration prepared in step  4. , and the lambda config. Terraform  null  resource executes
the command the first time you run  apply  and every time the command or the config file
are updated.
If you get the "Error parsing parameter 'cli-input-json': Invalid JSON received." error, make
sure the path to the input parameters json is correct and is prefixed with  file:// . I.e.  --cli-
input-json file://${var.update_user_pool_config_file} .
If you get the "Parameter validation failed: Unknown parameter in input: "Id", ..." error, make
sure you removed all keys not supported by the  update-user-pool  from the parameters file.
If you get the "An error occurred (InvalidParameterException) when calling the
UpdateUserPool operation: Please use TemporaryPasswordValidityDays in PasswordPolicy
instead of UnusedAccountValidityDays" error, remove
the  AdminCreateUserConfig.UnusedAccountValidityDays  setting. It is replaced
by  Policies.PasswordPolicy.TemporaryPasswordValidityDays .
Make sure it works
Once all the resources are deployed we can register a new user to make sure the email
with a code is sent by the ESP.
With CloudFormation you can find out the Cognito User Pool Client ID with
aws cloudformation describe-stacks --stack-name cognito-custom-email-sender-cf-stack --query "Stacks[0].Outpu

With Terraform, it will be listed in the outputs.

Register a new user:
aws cognito-idp sign-up --client-id <CLIENT_ID> --username [email protected] --password <PASSOWORD> --u
{
"UserConfirmed": false,
"CodeDeliveryDetails": {
"Destination": "h***@m***.io",
"DeliveryMedium": "EMAIL",
"AttributeName": "email"
},
"UserSub": "51c9045e-2f3e-4..."
}

It worked!

If you get index.handler is undefined or not exported error in CloudWatch, make sure you

zipped only the function files, and not the containing folder.
If you get KMS key arn must be a string. or Unable to decrypt data key and one or more KMS
CMKs had an error. error in CloudWatch, make sure you're passing the KMS key in the
environment variables to the Lambda and the value is ARN of the KMS key and not its ID.
Cleanup
At least destroying resources is much easier!
CloudFormation
aws cloudformation delete-stack --stack-name cognito-custom-email-sender-cf-stack
Terraform
terraform destroy