Databuff
GitHub
Skip to content

中文  |  English

Architecture · Application Performance

Read Telemetry Pipeline and Storage first.

Design Intent

APM should not be a burden for ops teams — minimal architecture, full features, works out of the box.


Three Core Components

ComponentResponsibilityWhy this design
IngestReceive traces and metrics from applicationsData entry only — lightweight and efficient
StorageUnified storage for all observability dataOne engine — no multiple databases
PlatformQuery, visualization, alerting, AIAll capabilities in one service

vs traditional APM: Often Elasticsearch + Kafka + many microservices + complex config. DataBuff runs everything with 3 containers.


Full Data Pipeline

Key Design Choices

DesignValue
OTel standard ingestionNot tied to a specific agent; ecosystem-friendly
Automatic trace assemblyDistributed fragments become full chains — no app changes
Metrics derived from tracesOne data source, two uses — lower collection cost
Minute-level pre-aggregationFast queries, efficient storage, efficient alert evaluation

Value of Minimal Architecture

Traditional APMDataBuff
Components10+3
Minimum resources16GB+ RAM8GB workable
Time to valueDaysMinutes
Ops staffDedicated teamDev self-service

Minimal does not mean bare-bones — troubleshooting, tracing, service metrics, topology, and alert evaluation are all covered in Phase 1.


Relationship with AI

Another benefit of minimal APM architecture: AI's data path is minimal too.

No multi-source stitching or cross-system queries — AI experts reach metrics, traces, topology, and alerts through one storage entry. That is the engineering foundation of AI-native OpenTelemetry APM.