Senders of large email volumes rely on the concept of multiple virtual queues for efficient delivery; separating email from different customers, to different destinations. This guarantees that for example one customer with a gigantic queue, or a destination which is tarpitting, doesn’t affect delivery of other traffic. Email transactions can take some time to complete, often as a result of content filtering at the destination. It is therefore necessary to support many simultaneous (concurrent) connections in each virtual queue, to support the throughput required for timely delivery. If each message takes an average of 1 second to deliver, 50 concurrent connections are needed to deliver 50 email per second.
The total maximum concurrency needed is the number of virtual queues multiplied with the desired concurrency for each queue. Traditional MTAs uses a process- or thread-based connection model. Those models scales poorly, and consequentially struggles to support more than a few thousand concurrent connections per server. This limitation becomes a problem as the system is scaled up, leading to bottlenecks or low customer density per MTA instance.
The Halon MTA uses a modern event-based connection model, sometimes called asynchronous or non-blocking. It scales exceptionally well; supporting tens of thousands of concurrent connections. Multiple worker threads are used to leverage all available CPU cores. This eliminates bottlenecks and can reduce the number of instances needed, which translates to lowered costs and simplified management.
Being asynchronous have benefits for receiving and processing inbound traffic as well. Consider the scenario where thousands of slow, or even idle, clients are connected to an MTA. Traditional MTAs would have to defer new connections from potentially legitimate senders, disturbing the service. Asynchronous MTAs like Halon are on the other hand be able to cope with a significantly large number of concurrent connections, and would handle the scenario above without breaking a sweat.
If you’re interested in learning more, don’t hesitate to reach out.
Since the first Halon MTA release in 2008, we’ve had a text-based queue query syntax called HQL (a play on SQL) as part of our SOAP API. While it has served us well during all those years, it was time to move on to something more modern. The new 5.3 release (codename “buffy”) comes with a Protocol Buffers and JSON API which introduces a programmatic approach to queue operations.
The request and response body schemas are available on our Github page. For your convenience, the QueueList, QueueGroupBy, QueueUnload and QueueUpdate requests all have the same Condition argument. Those API calls can operate on both the active and defer queues, as well as messages on hold. That is why the condition argument both contains things like retry count, as well as resolved remote MX and IP. You can specify as many conditions as you like, and create logic-or expressions by specifying multiple conditions of the same type. There are exact matching, regular expressions, and intervals for integers and date/time. Needless to say; incredibly powerful.
The QueueGroupBy call returns the distribution based on the grouping parameters and intervals you choose; such as number of messages in various age buckets, grouped by recipient domain. This is useful for getting an overview of a large queue. Queries are blazingly fast, even with very large queues. All queue metadata (essentially the fields available as conditions) is loaded into memory, in order for the virtual sub-queues to work.
Halon 5.3 also comes with a new CLI called halonctl. It happens to be very useful when working with our API, as it can output the API request and response bodies for the command you run in JSON format. As you can se in the example below, the request body is printed first:
The CLI covers all the functionality of the product, and is a great complement to the web administration. Its configuration management sub-commands are useful for integrating Halon MTA instances into provisioning, deployment and CI/CD toolchains such as Puppet or Chef, where running commands is easier than making API calls.
Halon 5.3 comes with many other great improvements; such as connection pooling, a more efficient queue quota function, a new on-disk queue format and an Iconv() class for internationalisation conversion. Please see the release notes for a complete list of changes. We hope that you will enjoy this release as much as we do! If you are new to Halon, don’t hesitate to contact us, or dig into all our documentation that is available publicly on our website.
We’re very proud to announce the upcoming 5.2 release “polly” which introduces a powerful queue policy engine. First and foremost, the queue and SMTP client’s network layer is now asynchronous. This allows an instance to handle tens of thousands of parallel connections. In combination with the reworked connection concurrency limits, this allows dynamic creation of a virtually unlimited number of independent sub-queues. This is useful for senders that need to separate email streams so that those that move slowly or get stuck don’t block others.
As usual, we made it flexible enough to fit any email service provider’s needs. Rather than having a fixed set of parameters and rollup/grouping options for establishing the sub-queues (with their respective thresholds), we allow you to define what constitutes a unique entry. You can choose any combination of fields, and group/rollup entries using regular expressions or wildcard. In the example below, we limit the concurrent per source IP and remote MX, and also rollup all Google’s MX entries into the same entry. The default concurrency is 5, except Google that gets 10.
Sometimes rollup per MX doesn’t cut it. There are several Microsoft Office365 locations (clusters), but the customer MX doesn’t reveal which they are on. To set a certain threshold for Office365 locations, we can rollup and match per MX, but limit per IP, as per the example below. Note that there’s no default threshold; it only affects Office365.
Thresholds and suspensions can be modified on the fly without reloading the configuration via API, CLI, web administration or the MTA itself through this Halon script:
// If we have more than 10 failures per minute, lower rate for 5 minutes
$mx = $arguments["attempt"]["connection"]["remotemx"];
$code = $arguments["attempt"]["result"]["code"];
if ($mx and $code >= 400 and !rate("mx-fail", $mx, 10, 60, ["sync" => false]))
cache ["ttl" => 300]
["remotemx" => $mx],
["rate" => [10, 60]], 300);
The reworked queue naturally comes with many new tools and APIs for interacting with the new functionality. This includes more subtle improvements, like the ability to view the queue’s shape by message age. By pressing an interval, you can dig into the specific messages, which are grouped by fields of your choice.
The new shared memory script functions and API opens up several possibilities. You can script statistic counters, which can then be read periodically over the API. Another use case is pre-loading data into the MTA over the API, rather than fetching and caching from within the script.
The upcoming Halon 5.1 release introduces a new SMTP server proxy script. It’s configured to be executed before specific (or all) SMTP commands, even command which isn’t recognised by the SMTP server. In this blog we’ll describe how to implement our proposed SMTP LANG extension using this new script hook. First of all, we announce the LANG extension and the languages supported in the HELO script:
We celebrate the new year with news on the upcoming release, which bundles many exciting features.
First and foremost, the new RESTful API with an OpenAPI specification makes integration into various development and deployment toolchains much more enjoyable. Since most of our customers already integrate Halon into their directories and control panels by making REST queries from Halon, it makes perfect sense that Halon can be provisioned in the same way.
Secondly, we’re introducing a new end-of-DATA script that’s executed once per message, as opposed to the per-recipient DATA script. Whereas the per-recipient version is convenient when you want to treat each recipient individually and let the Halon software take care of queueing and consolidating the respective actions into an SMTP response, the per-message version gives you maximum flexibility and control over execution. The $transaction variable is populated gradually during the SMTP conversation with sender information, and an array with recipients accepted by each RCPT TO command. To then relay a message to its recipients, you call Queue() for each $transaction["recipients"] and then Accept(). Making per-recipient message modifications using the MIME() class is now easier thanks to the new snapshot() and restore() methods.
The code editor’s built-in CSV editor now supports custom form controls, defined like a “schema” on a per-file basis using a JSON format. You can use checkboxes for booleans, select controllers for enumerated types, and input fields with validation for things like dates, email addresses or any regular expression you like. It makes it much more convenient and safe to create and edit lists and settings that you want to have in your Halon configuration file.
There’s a new LDAP() class that replaces the previous ldap_ functions and LDAP settings in the configuration. It provides greater flexibility, and an improved usage pattern using an iterable LDAP result object.
Finally, there are massive under-the-hood improvements. There’s a new on-disk YAML configurations with JSON schemas and Protobuf control sockets, which is used by the componentised Linux package’s new Visual Studio Code plugin and command line tools. The integrated package is built on FreeBSD 12, which ships with OpenSSL 1.1 and thus TLS 1.3 support. It was published as a standard by IETF in August last year, and is much anticipated as it contains many security improvements over previous TLS versions. The queue database is now using the latest and greatest PostgreSQL version 11.1, and the queue is automatically migrated on boot as usual.
We have that you’ll like this new release as much as we do! Check out the full changelog on GitHub for more information, and familiarise yourself with the important changes outlined in the release notes document before upgrading.
Halon 4.0 introduced a feature we call “live staging” where you can deploy multiple running configurations at the same time, with per-connection conditions. It allows you to reliably rollout changes or new features to a production system for only a few testing IPs, or a select percentage of the traffic. With Halon 4.7, we proudly present “live debugging” using which you can add logpoints to your scripts. It enables you to inspect the full context of SMTP transactions in real-time, using the live staging conditions as connection selector.
Those points are added directly to the Monaco-based IDE, and results are inspected on a per-connection basis. You can create multiple points, triggered by multiple messages, and jump back and forth between them.
Authenticated Received Chain (ARC) is a proposed standard that have been developed to help address issues with DMARC and certain forwarders, such as mailing lists. It defines a standard for how to pass authentication results from one intermediary to another, making this information available to the recipient system. It works even in the case of multiple intermediaries, a.k.a. a chain
DMARC verifies the sender authenticity, as specified by the RFC5322.From header domain name, using SPF and DKIM. Certain indirect email flows such as mailing lists break this by altering the message, while maintaining the original From header. It causes issues for both senders that publish a DMARC policy, and receivers that verify DMARC. The two large mailbox providers AOL and Yahoo published a p=reject DMARC policy for their domains in 2014, causing some disruption for senders on those domains. It occurred when emailing recipients on mailbox services that verifies DMARC via for example mailing lists. This was, and still is, remedied by ad-hoc solutions.
ARC in itself isn’t a reputation system. The specification doesn’t define how the reputation of intermediates should be tracked, nor how public lists should be operated. In other words, as a recipient mailbox provider you still have to operate such systems in order to make use of the information that ARC provides. DMARC.org announced ARC at a M3AAWG meeting in Atlanta, 2015, where it’s been a frequent topic ever since.
$chain = ARC::chainValidate();
if ($chain["status"] == "pass" or $chain["status"] == "none")
"201805", "example.com", "pki:arc",
->SPF(["smtp.client-ip" => $senderip])
->addMethod("arc", $chain["status"], ["header.oldest-pass" => $chain["oldestpass"] ?? "0"])
We have just released an implementation for ARC (draft 14) on Github, which supports both verification and (re)sealing. It’s written in Halon script, and we’re using it on our own domain to start with. If you’re interested in taking it for a spin, just let us know.
M3AAWG meetings are an exceptional opportunity to discuss the latest in messaging security with other professionals in a focused environment of working sessions and educational panels. This time we meet in Munich, Germany. Leading industry experts, researchers and public policy officials address such diverse topics as bot mitigation practices, social networking abuse, mobile abuse and pending legislation.
As an official supporter member, we will of course participate in the Munich meeting on June 4th-7th. If you want to meet up, just get in touch!
You probably know from before that Halon’s scriptable SMTP server enable email providers to avoid blacklisting and increase deliverability. The 4.6 release, “curry”, contains Cyren’s outbound anti-spam (OAS). In combination with our cluster-synchronised rate limit function, it provides incredibly effective and accurate abuse prevention. Just like their Cyren’s inbound anti-spam, OAS uses a hash-sharing technology called recurrent pattern detection (RPD) that identifies outbreak patterns. It’s designed to detect spam from internal sources rather than external, and doesn’t report/contribute any signatures since it could blacklist your own infrastructure.
With the flexibility of scripting you can determine customer/sender identities accurately even in mixed traffic. This is used as identifier for rate limits based on classifiers such as Cyren’s OAS, delivery failure rate, queue size, etc. By using IP source hashing and alternative IPs for suspicious traffic, deferring obvious abuse and controlling connection concurrency, you can achieve high deliverability with minimal administration.
The 4.6 release comes with many additional features and improvements. It adds SNI support to the TLS functions. The Monaco-based code editor now have additional code completion, built-in documentation, tabs, and a mini-map.
In the beginning, everything was ASCII and English. Since then, we’ve seen Unicode (international character sets) and IDN (international domains names) become widely adopted. Last year we implemented SMTPUTF8 that enables international mailboxes.
So why not support other languages in text-based protocols? We give to you “The SMTP Service Extension for Protocol Internationalization” RFC draft, introducing the EHLO keyword LANG. It will be the first SMTP software to support our to-be submitted RFC draft. Initially it will support Swedish, Spanish and Australian, and will default to Swedish when talking to supported systems.
250-LANG SE ES AU
BREV TILL:<hå[email protected]än.se>
250 Togs emot
250 Vi ses!
If you made it this far, April fool! We will publish information on the upcoming 4.6 release some time after the 1st of April.
The Halon MTA is a flexible email operations and security platform.
It enables organisations that operate large-scale email services to offer competitive features by rapid implementation
and to lower maintenance costs through reliable deployment and reduced complexity.