Can Fluent Bit Replace the OpenTelemetry Collector? v4 Benchmark & Review

Deep Dive Welcome to another episode of Is it Observable?—your go-to series for everything OpenTelemetry and Kubernetes!
In this episode, we explore the exciting new features of Fluent Bit v4, announced at KubeCon Europe 2025. Fluent Bit has long been a powerhouse for log processing, but with v4, it’s stepping into the world of distributed traces and metrics like never before.


🔍 What You’ll Learn:
All the major updates in Fluent Bit v4
How to use the new conditional processors for smarter log pipelines
Deep dive into trace sampling processors: probabilistic vs. tail sampling
Real-world benchmarking results comparing Fluent Bit and the OpenTelemetry Collector
Limitations and missing features you should know before going all-in

💡 Whether you're building observability pipelines, optimizing resource usage, or just curious about Fluent Bit’s evolution—this episode is packed with insights, demos, and practical advice.

📊 Benchmark Highlights:
Fluent Bit’s tail sampling can consume up to 1GB of memory—why this might be a red flag
CPU usage increases by 250% with sampling—but still within acceptable limits
OpenTelemetry Collector outperforms Fluent Bit in sampling stability and telemetry

⚠️ Key Takeaways:
Fluent Bit v4 is a big leap forward, but still lacks:
Kubernetes metadata enrichment for traces/metrics
Metric conversion capabilities
Trace-aware load balancing for horizontal scaling
Telemetry visibility into sampling decisions


👍 Like, Subscribe & Hit the Bell to stay updated on all things observability!

Consider joining the channel:    / isitobservable  

▬▬▬▬▬▬ 🔗 Additional Info 🔗 ▬▬▬▬▬▬
🔗Dynatrace Trial: https://bit.ly/3KxWDvY
🔗Github repo for the tutorial: https://github.com/isItObservable/flu...
🔗fluentbit : https://fluentbit.io/
🔗Blog : https://isitobservable.io

▬▬▬▬▬▬ 👋 Contact me 👋 ▬▬▬▬▬▬
➡ Twitter:   / isitobservable  
➡ LinkedIn:   / isitobservable  

▬▬▬▬▬▬ ⏱ Timecodes ⏱ ▬▬▬▬▬▬
00:00 Intro
00:13 Welcome
02:20 What’s New
06:49 Conditional processor
11:24 Sampling processor
20:38 Benchmark
24:10 Conclusion