df20

Overview

This is the code that handles the email ingest/processing for the Hey Email Research Labs (HERL) Dumpsterfire. All ingest, scanning, filtering, processing, and moderation are handled here.

You will need:

1. AWS account
2. Domain names
3. A Raspberry Pi with node-red
4. Familiarity with Terraform, AWS Lambda, SES, SQS, ruby, node.js, node-red, bash, and jq.
5. A lot of patience.

Details

1. Customer emails dumpsterfire@hey.com.

2. AWS SES receives email.

   1. SES checks for spam/dkim/spf
   2. SES sends headers to email_filter js lambda.
   3. SES drops email into s3 bucket s3://dumpsterfire-bucket/inbound/

3. S3 fires off SNS notification when email hits /inbound.

4. SNS triggers processor ruby lambda. This moves the email between 4 states, triggering a S3->SNS notification each time:

   1. /inbound -> lambda screens email for content size (<5MB)
   2. /screened -> lambda formats email body for printing
   3. /print -> lambda publishes message to SQS screener queue with job data, and another SQS queue for initial marketing response.
   4. /completed, but that happens in step 8.

5. SQS triggers screener ruby lambda. This reads the s3://dumpsterfire-rules/rules.json file and filters jobs accordingly, dumping the ones that pass into the print queue.

6. Raspberry Pi pulls message off SQS print queue for approval; either sending to the moderated queue or deleting the job from the queue.

7. The print loop pulls messages off the moderated queues (Normal, VIP, or Special), and prints/burns them.

8. Rasberry Pi puts message on SQS Queue when complete.

9. SQS triggers cleanup lambda. This puts a message on another SQS queue for final email response, and moves the file to s3://dumpsterfire-bucket/completed/.

10. S3 fires off SNS notification when email hits /completed 9 SNS triggers processor ruby lambda, which updates s3://dumpsterfire-cloudfront/stats.json.

Infrastructure

Terraform bits are under terraform, ruby lambda code is under lambdas/processor, and node-red flows are under node-red.

Appropriately named dumpster profile needed in your ~/.aws/config/~/.aws/credentials for this to work.

Generate a new lambda package:

for lambda in processor cleanup screener; do
    pushd lambdas/$lambda
    rake package
    mv *.zip ../../terraform/production/
    popd
done
pushd terraform/production
terraform apply -auto-approve
popd

Email Filter

It's a javascript lambda because that's the only lambda runtime that supports callbacks, which are required for SES Actions to work correctly. (AFAICT)

pushd lambdas/email_filter
zip -r email_filter.zip index.js index-async.js node_modules package.json package-lock.json
mv email_filter.zip ../../terraform/production/
popd
pushd terraform/production
terraform apply -auto-approve
popd

Hey Marketing Auto-reply

This is a systemd service running on a node authorized to send email from the hey domain. The service file is /etc/systemd/system/reply.service, and it runs /path/to/hey-marketing-reply/reply.sh. The script relies on the aws commands and credentials in my home directory. 😬

To update, copy the new script into /path/to/hey-marketing-reply/reply.sh, fix ownership if needed, and run sudo systemctl restart reply.service.

Logs are available via sudo journalctl -f -u reply.service.

Same for the hey-marketing-complete-reply mailer.