What Is This OpenTelemetry Thing? • Martin Thwaites • GOTO 2024

OpenTelemetry has become the de facto standard for observability, overtaking other projects in the CNCF and unifying previously competing standards (OpenTracing and OpenCensus)

Key components include:

• Traces: Groups of spans showing request flow and causality
• Metrics: Time series data with labels for aggregation
• Logs: Point-in-time structured data
• Collector: Central component for processing and routing telemetry data

The OpenTelemetry Collector provides critical benefits:

• Centralized configuration
• Security and access control
• Data filtering and redaction
• Vendor-agnostic data routing
• Reduced egress costs

Auto-instrumentation libraries provide quick startup with minimal code changes, while manual instrumentation allows deeper customization and context

Sampling is essential for cost control:

• Head sampling: Simple but loses context
• Tail sampling: More complex but preserves full trace context
• Must balance storage costs with debugging needs

Semantic conventions provide standardized naming and attributes across different languages and frameworks

The protocol and data model are vendor-agnostic, allowing easy switching between backends and multi-vendor setups

Key challenges include:

• Documentation needs improvement
• Development can be slow due to committee processes
• Running in async/serverless environments
• Managing high cardinality data

Benefits include:

• Reduced vendor lock-in
• Standardized telemetry across systems
• Better debugging capabilities
• Cost control through centralized management