Discord Feature Trails

Feature trailActive

Cost Attribution

Discord is developing capabilities to understand and attribute hosting costs to specific product features and API endpoints. This involves extending profiling tooling to allocate deployment costs based on the time spent on feature-related code, aiming to provide granular insights into spending for features like messaging, streaming, and gifting. The initiative now includes detailed cost attribution for API endpoints and background tasks, enabling engineers to understand the cost impact of changes. The latest iteration focuses on attributing costs at the API endpoint and background task level, allowing for more granular understanding of hosting spend for specific features.

1 post · 2026

Timeline

Feature trailActive

App Data Access Controls

Discord is implementing stricter controls and review processes for how applications access user data within servers. This initiative aims to enhance user privacy and data security by introducing annual reviews, higher access thresholds, and a phased approach for developers to adapt to new data access requirements. The changes impact apps that rely on sensitive user information like member lists, presence, and message content, encouraging the adoption of more secure in-app interactions like slash commands. Developers are given 90 days to adapt to new requirements.

3 posts · 2026

Timeline

Feature trailActive

Edge Voice Infrastructure

Discord is expanding its reach and integration into new platforms, with a significant focus on virtual reality (VR) environments and game development. This initiative aims to provide a seamless Discord experience within VR headsets and games, allowing users to communicate with their servers and friends without leaving the VR environment or game. Key developments include direct app availability on VR app stores (Meta Quest), full feature parity with the desktop/mobile app, integration with VR har. Discord has successfully migrated a significant portion of its voice and video traffic to Cloudflare's edge network, deploying in over 300 cities. This migration has led to measurable improvements in call quality, with 70% of regions showing year-over-year improvements, and specific reductions in ping averages and packet loss in areas like Frankfurt. The initiative involved building new infrastructure to support this edge deployment and investigating quality issues.

1 post · 2026

Timeline

Feature trailActive

End-to-End Encryption for Voice and Video

Discord has successfully implemented end-to-end encryption (E2EE) for all voice and video calls across all platforms, including desktop, mobile, web browsers, gaming consoles, support for Discord bots/apps, and its Social SDK. This was achieved by developing and deploying the DAVE protocol, an open and audited E2EE protocol for audio and video. The migration ensures that all calls, except those in stage channels, are now E2EE by default, enhancing user privacy and security without compromising t. This post details the final rollout to browsers and consoles, including technical challenges like WebAssembly performance and Web Worker architecture, and the deprecation of non-E2EE calls.

2 posts · 2025–2026

Timeline

20252026

Feature trailActive

ScyllaDB Cluster Automation

Discord has developed an automated system for provisioning, configuring, and managing ScyllaDB clusters at scale. This system significantly reduces the time and effort required to set up replica clusters for testing and validation, enabling faster release cycles and more robust pre-production validation. The automation now includes provisioning, configuration, joining nodes, replication validation, and dual-write pipeline setup, reducing the process from a day and a half to under two hours. This post details the further refinement of this automation, focusing on reducing the time to stand up a full replica cluster for release validation from a day and a half to under two hours.

1 post · 2026

Timeline

Feature trailActive

Session Management and Realtime Infrastructure Resilience

Discord's session management and realtime infrastructure have been enhanced to improve resilience against cascading failures. Following a significant outage caused by a configuration update that led to a large number of session management servers shutting down, the team has analyzed system behavior under sudden load. Improvements focus on preventing similar widespread degradation by strengthening downstream systems and identifying and mitigating new bottlenecks that emerge during load spikes. This post details how a 17% loss of session management servers overwhelmed a routing service, leading to voice and video degradation, and the subsequent analysis and improvements made to prevent recurrence.

1 post · 2026

Timeline

Feature trailActive

Experiment Metrics Optimization

Discord is optimizing its experiment metrics by reducing the number of metrics used to improve the quality of insights and reduce false positives. This involves a shift from a broad 'Default Metric List' to a curated set of high-quality metrics that capture distinct concepts, addressing issues related to p-values, false positives, and recall in experiment analysis. The team is moving towards using fewer, higher-quality metrics that capture distinct concepts to mitigate issues with false positives and recall.

1 post · 2026

Timeline

Feature trailActive

Distributed Tracing for Elixir Systems

Discord has implemented distributed tracing for its Elixir systems to provide deeper insights into the performance and reliability of its services. This initiative allows engineers to trace operations across different parts of the system, helping to diagnose performance bottlenecks and outages more effectively. The implementation involved building custom solutions to propagate tracing context within Elixir's communication mechanisms, enabling end-to-end visibility without impacting system availability. The system now provides detailed end-to-end user experience insights beyond metrics and logs.

1 post · 2026

Timeline

Feature trailCooling

Distributed ML Platform

Discord has developed a platform around Ray to simplify distributed ML for engineers. This platform includes custom CLI tooling, orchestration with Dagster and KubeRay, and an observability layer called X-Ray. The focus is on improving developer experience for distributed ML, enabling complex models and scaling training jobs across multi-GPU clusters. The platform has evolved to support multi-GPU clusters and ease the transition from single-node experiments to production-ready distributed comput

1 post · 2025

Timeline

20252026

Feature trailCooling

Safety Rules Engine

Discord has open-sourced Osprey, its safety rules engine, to help companies build stronger safety measures. Osprey allows teams to investigate real-time activities and deploy dynamic rules to address emerging threats with minimal engineering overhead. ROOST has now announced Osprey as a free, open-source tool for the AI era, making it accessible to organizations of all sizes. The open-sourcing of Osprey by ROOST signifies a commitment to shared safety technologies and accelerating innovation in online safety.

3 posts · 2025

Timeline

20252026

Feature trailCooling

Message Indexing and Search

Discord's message search infrastructure has evolved significantly since its initial design in 2017. Initially, messages were indexed lazily into Elasticsearch, sharded by guild or direct message across two clusters. As Discord grew, this system began to show cracks, necessitating further development and optimization of message indexing and search capabilities to handle trillions of messages. This post details the evolution from indexing billions to trillions of messages, highlighting the challenges and solutions in scaling Elasticsearch for massive message volumes, including strategies for sharding, indexing, and querying.

1 post · 2025

Timeline

20252026

Feature trailCooling

dbt Data Transformation at Scale

Discord has developed custom solutions to scale dbt for processing petabytes of data and supporting over 100 developers working on 2,500+ models. This addresses limitations in re-compilation times, incremental materialization, and developer collaboration, creating a robust and scalable data transformation platform. The latest iteration focuses on optimizing dbt's performance for massive datasets, reducing compile times from 20+ minutes to rapid development cycles, and automating complex backfill operations. The post details specific custom solutions implemented to enhance developer productivity, prevent breaking changes, and streamline complex calculations, extending dbt's core functionality for massive scale.

1 post · 2025

Timeline

20252026