Telemetry Showdown: Fluent Bit vs. OpenTelemetry Collector - A Comprehensive Benchma...
summary
Benchmarks run with logs only, logs+traces, and logs+traces+metrics show Fluent Bit using 15–20 MB vs OpenTelemetry Collector 80–120 MB for simple log pipelines, but adding Prometheus scrapes and cumulative→delta conversion on a daemonset makes the Collector hit memory spike to ~10 GB due to cumulative→delta conversion; moving metrics/traces to a stateful collector flattens usage. Tail sampling raises costs: Fluent Bit sampling ~800 MB vs Collector ~2.3 GB (about 1.5 GB higher), so pick Fluent Bit for cheap log processing, Collector for richer metric/trace conversions.
AI-written summary · about
chapters (13)
- 0:00Telemetry agents comparison
- 1:47Benchmark purpose
- 2:09Fluent Bit vs OpenTelemetry
- 3:35Design differences
- 8:13Logging features
- 10:25Metrics support
- 12:11Traces ingestion
- 13:14Benchmark setup
- 15:00Performance results
- 15:42Memory leak fixes
- 16:25Sampling performance
- 17:31Recommendations
- 18:13Q&A
related talks
24mTelemetry Showdown: Fluent Bit Vs. OpenTelemetry Collector - A Comprehensive Benchma...
25mTelemetry Showdown: Fluent Bit Vs. OpenTelemetry Collector - a Comprehensive...
57mTelemetry Showdown: Fluent Bit vs. OpenTelemetry Collector - A Comprehensive Benchmark Analysis
07mFluent Bit vs. OpenTelemetry Collector - Why Not Both?
15mJourney from Fluent Bit, Fluentd and Prometheus to Open...
27mLog Support in OpenTelemetry
23mControlling Data Overhead with the OpenTelemetry Collector
36mFluent Bit: Logs, OpenMetrics, and OpenTelemetry all-in-one