Skip to content

comparison.md

Latest commit

 

History

History
35 lines (26 loc) · 2.85 KB

comparison.md

File metadata and controls

35 lines (26 loc) · 2.85 KB
Raw

traceperf compared with other tools

An honest comparison: where traceperf is the better choice, and where something else is.

At a glance

traceperf clinic.js / 0x node --cpu-prof OpenTelemetry SDK APM agents (Datadog, New Relic, ...)
What it shows call trees and stats for tracked calls, plus causes flame graphs, event-loop and GC diagnosis a CPU profile distributed traces traces, metrics, dashboards
Setup --require traceperf/auto, or one import run the app under the tool a flag SDK, exporter, collector, backend agent plus account
Async-aware request trees yes no no yes yes
Explains why something was slow yes (explain) yes (Doctor) no no partly
Line-level CPU detail for slow calls (profileSlowCalls) yes yes no some (profiling add-on)
Meant for production yes (sampling, kill switch, bounded memory) no no yes yes
Across services no (exports to OTel for that) no no yes yes
Runs in browsers, Bun, Deno, edge yes no no partly partly
Cost free, no dependencies free free free, plus a backend paid

Choose traceperf when

  • You want to know why a request or function is slow, and want the answer in your terminal in minutes, not after setting up a tracing backend.
  • You want the same tool in development, CI and production: trees while you develop, traceperf compare --fail in CI, stats/Prometheus in production.
  • You care about specific operations (a handler, a query, a job) and want per-operation p50/p95/p99 without a backend.
  • You run code in several runtimes (Node, Bun, Deno, Workers, browsers) and want one tool.

Choose something else when

  • You have no idea where the time goes and it's CPU-bound: start with node --cpu-prof or 0x. traceperf's profileSlowCalls covers this only around calls that were already slow.
  • The slowness is GC or memory pressure: clinic.js Doctor/HeapProfiler and Chrome DevTools heap snapshots are better. traceperf's profileAllocations is a narrow tool for one function.
  • You need traces across many services: use OpenTelemetry. traceperf can feed it (openTelemetryExporter), but it does not propagate context between processes.
  • You want hosted dashboards, alerting and retention without running anything: an APM product does that; traceperf gives you Prometheus metrics and files.

How it differs from OpenTelemetry

OpenTelemetry is a standard for collecting telemetry and sending it somewhere. traceperf is a tool that answers a question on the spot. The two work together: traceperf's call trees and stats are useful without a backend, and when you have one, openTelemetryExporter sends the same calls as spans. A traceperf span costs about a sixth of an OpenTelemetry SDK span (1.5 µs vs ~10 µs, see benchmarks).