Anti-Spam Server with Rspamd on Offshore VPS

SpamAssassin was the standard open-source spam filter for two decades, but it is showing its age. Perl-based, single-threaded per message, and slow under contemporary mail volumes. Rspamd was built from scratch in C and event-driven, with a Lua scripting layer for custom rules and a Redis backend for shared state across multiple Rspamd workers. In practice, Rspamd scans messages in single-digit milliseconds where SpamAssassin takes hundreds. The accuracy is at least as good thanks to built-in DKIM and SPF verification, fuzzy hashing against known spam, Bayesian autolearn, and DNS blacklists. For a fresh self-hosted mail deployment on Anubiz Host offshore VPS, Rspamd is the recommended choice over SpamAssassin in almost every scenario. The main reason to choose SpamAssassin is legacy compatibility with an existing ruleset you do not want to migrate.

On an Anubiz Host offshore VPS running Debian 12 or Ubuntu 24.04, install Rspamd from the official repository at rspamd.com, install Redis from the distribution package, and wire Postfix to Rspamd via the milter protocol. The default Rspamd configuration is sensible out of the box for most setups. Enable the web UI bound to localhost only, then expose it via SSH tunnel for administrative access. Never expose the Rspamd UI directly to the public internet. Configure DKIM signing through Rspamd's dkim_signing module rather than running a separate OpenDKIM daemon, which simplifies operations. For Bayesian learning, feed Rspamd a few hundred ham and spam samples to bootstrap the classifier. Autolearn then continues training from real traffic. Most deployments reach excellent accuracy within a week of normal operation.

Rspamd in production benefits from a few discipline items. First, tune the action thresholds based on observed score distributions. The default thresholds, greylist around 4, add header around 6, reject around 15, are reasonable but every mail flow has different characteristics. Second, integrate Rspamd reputation modules carefully. The ratelimit module catches outbound abuse, the asn module penalizes mail from low-reputation networks, and the surbl module checks URIs in message bodies against known spam URI lists. All three are valuable but can also produce false positives on legitimate transactional mail if misconfigured. Third, monitor the Rspamd statistics endpoint with Prometheus and Grafana. Watching scan counts, action distribution, and Bayesian accuracy over time surfaces drift before users complain. Anubiz Host customers running Rspamd typically settle into a low-maintenance steady state within a few weeks of initial deployment.