How To Build an Email Service on SES

Foundations

Amazon Simple Email Service (SES) handles hundreds of billions of email messages every month. While many are outbound, one of the fastest-growing parts of the business is for inbound traffic. Customers send and receive email via SES using a combination of public SMTP interfaces and the SES SDK. Traditionally, most customers used SES alongside their existing corporate mail systems, but did you know it’s possible to build a complete email service with SES at its core? In fact, it’s already been done – it’s known as Amazon WorkMail, and it provides mailbox and calendar services to tens of thousands of customers (and millions of mailboxes) around the world.

Ingredients for Success

Email transport depends on a few core components. First of all, you have to be a reputable sender, or the receiving email systems are going to reject anything you try to send. You also have to be insulated against spurious reports of abuse, so that one bad apple can’t take down the entire service for everyone. The solution for both of those issues is the same: have an enormous number of public Mail Transfer Agents (MTAs), and manage their IP reputations actively. If someone reports spam coming from one of those IPs, and it gets added to a block list somewhere on the internet, you have to have a rapid response mechanism to engage with the block list operator and take their prescribed steps to clean up the entry.

The Highest Standards of Security

Similarly, you have to consult those same block lists when mail is sent to your own systems from anywhere on the internet. Inbound email is subjected to a variety of authentication steps before it’s released for delivery to a destination. Quality providers will leverage checks called SPF (Sender Policy Framework) and DKIM (Domain Keys Identified Mail). SPF is designed to prevent malicious senders from masquerading as other domains, and DKIM enables a receiving system to validate the authenticity of the sender and to confirm it hasn’t been manipulated while in transit. If either of these checks fail, a receiving system may take action ranging from dropping the message entirely, to flagging it as suspicious but still delivering it to the user’s inbox. A third security control, DMARC (Domain-based Message Authentication, Reporting, and Conformance) takes SPF and DKIM outputs and generates a series of instructions for receiving mailbox providers about what to do with questionable mail. Any serious provider will support these mechanisms and provide visibility into their actual performance on your email.

Amazon WorkMail’s Interface with SES

Once you’ve got clean email and reputable senders or recipients, you have to be able to figure out where to deliver the message itself. SES Inbound has a specific internal action when used with WorkMail, where the message is routed to WorkMail’s own infrastructure for matching against a known user’s inbox and performing the indexing and storage operations necessary to make it show up in your desktop, web, or mobile mail client. There are a number of options which may take place while that message is in transit, however, and the SES framework supports those with its flexible routing options. For example, a very popular choice is for customers to trigger a transport rule powered by AWS Lambda for inbound and/or outbound messages. Some of these are simple – they append a standard banner to the message if it is inbound from an external source, for example – but there is really no limit to what programmatic steps can be taken. You could submit message content to a large language model (LLM) for training or inspection. You could examine its use of language with AWS Bedrock to train a foundational model in generative AI about how to write emails itself. WorkMail and SES support and encourage these kind of big ideas for working with your message content.

Managing Spikes and Growth

Another critical advantage SES provides is the ability to absorb huge spikes in inbound traffic, and to sustain very large permitted volumes of outbound traffic as well. Email’s underlying standards and protocols offer administrators some degree of control over delays in transit, by implementing retry intervals to buffer messages if they can’t be delivered immediately. The classic on-premise enterprise use case, however, still runs the risk of overwhelming the capacity of the (single) mail server, either due to a malicious action by a sender or a huge increase in usage over a very short period of time. SES absorbs those spikes automatically and has orders of magnitude more capacity than any typical on-premise deployment, meaning that your mail enjoys multiple tiers of buffering only when required, and with no introduced latency if buffering is unnecessary.

Putting it All Together

So how does it all work together? The inbound use case is our main focus. When a message arrives via SMTP, SES first interrogates a back-end directory to confirm that the message is destined for an SES customer. If so, it looks up how the customer’s domain is configured, or if it is a WorkMail customer domain. From there the message passes through the SES message scanner, where its content is evaluated for spam or malware, and a scoring indicator is added to the message headers. That score may result in the message being dropped altogether, or it may result in the message ultimately being delivered to a Junk Mail folder in a WorkMail mailbox. Once scored, the message is either stored in the customer’s S3 storage, or delivered to WorkMail for further processing, such as being put in a specific folder, or redirected to another recipient. Once it’s stored somewhere, the customer can interact with it either using SES APIs, or via standard mail clients interacting with a WorkMail mailbox. In practice a mailbox is a structured object format also within S3, but without raw S3 access because the storage is managed as a system resource within WorkMail instead of being owned by an end customer.

The Customer Experience

When a WorkMail customer wants to send a message, they compose it in a mail client and then click ‘Send’ to send it via SMTP. In the outbound case WorkMail relays the message to SES internet-facing mail relays, which in turn look up the recipient domain information for details on how to route it. SES mail relays also perform the necessary security and authentication checks to ensure that the message is sent by a valid user (either SES native or WorkMail) and that the content is cryptographically signed so a receiving system can verify it hasn’t been manipulated in transit, using the DKIM mechanism described previously. When those steps are complete, the message is handed off to the next mail relay on the internet, and SES has no further role in its future unless a receiving system flags it as abusive. In that case the feedback is delivered to SES automatically and a series of containment actions are considered based on the nature and history of abuse reports. Thus the feedback loop to IP reputation is maintained even in the case of a rogue actor sending bad mail.

Robust Tooling Makes Email Look Easy

The bottom line is that SES enables these flows, and a customer wanting to build a comprehensive mail system could do so themselves if they didn’t want to use WorkMail or another existing email service provider. We’ve seen a tremendous range of creative solution-building from customers when they combine SES inbound and outbound mail, a subset of WorkMail mailboxes and their own rules and organization policies, the use of AWS Lambdas, and inline email security gateways. The flexibility to build whatever you need, without being tied to a single product vendor, is what makes SES so popular with its customers, and ensures that WorkMail – as a turnkey mail service – works so reliably for those customers who just need their mail and calendar to work.