Datadog Engineering

How we built an async-aware Python profiler

This article details how Datadog overhauled its Python profiler to support asynchronous code by reconstructing relationships between asyncio tasks. The authors explain the challenges of tracking execution flow across non-linear task dependencies and describe the optimizations implemented to reduce profiling overhead by 60 percent.

AI summary

02

Datadog EngineeringAug 11, 202614 readsenior

How we improved APM Java startup by encoding a prefix trie as a JVM constant

The Datadog APM team optimized Java agent startup time by 30% by improving class-matching efficiency during the JVM's early execution phase. The approach involved encoding class-name prefix matches as JVM constants and developing a lightweight class-file parser to reduce the computational overhead of identifying classes for instrumentation.

AI summary

03

Datadog EngineeringJul 1, 202619 readsenior

How we measure data completeness at scale

Datadog developed a distributed observability system to verify data completeness across billions of ingestion payloads per second. By modeling their ingestion pipelines as segments and tracking telemetry through each stage, the team created a framework that identifies where and why data drops occur. This approach enables real-time diagnostic capabilities and provides reliable signals for both human operators and automated decision-making agents.

AI summary

04

Datadog EngineeringJun 23, 202615 readsenior

How we migrated a live routing system using AI-assisted refactoring

Datadog engineers refactored their Stream Router control plane by migrating to a new storage architecture while maintaining live production traffic. They leveraged AI tools to accelerate the implementation of method-level refactoring by feeding models existing code, new schemas, and failing tests. The success of the migration relied on a robust test-driven development process to validate AI-generated code against a high-stakes, large-scale metrics pipeline.

AI summary