Engineering

Apdex compresses thousands of request latencies into one 0-to-1 number. The math, how to pick T, what counts as good, and how TracePath computes it.

Turning minified stack traces back into source, fast: a flat pointer-free cache format, mmap instead of a heap cache, and a 22,000x parse-to-lookup gap.

Same $16.49/month box, same binary, two embedded databases. DuckDB wrote 4x to 15x faster than SQLite and moved every dashboard read cliff by 100x.

Embedded telemetry stores fail slowly, not loudly: WAL growth, read cliffs, retention, and backpressure, plus the settings that decide each one.

A single Go binary writing OTLP metrics into SQLite on a 2 vCPU, 8 GB box sustained 58k metric points a second. Here is the method, and where it stops.

A minified stack trace points into compiled output. Symbolication maps it back to your source. Here is how source maps, VLQ mappings and debug IDs work.

Running your own observability stack is a real option with a real bill. An honest framework for choosing, from a vendor that only sells the managed one.

net/http recovers handler panics by default. That keeps the process up, not the service correct. How to pick a recovery boundary, and when to exit 1.

A collector buffers, redacts and samples telemetry between your apps and your backend. Here is a minimal OTLP config that forwards to TracePath.

TracePath speaks MCP: a remote server with OAuth built in and a local stdio one on the CLI. The tools, the three credentials, and when to skip it.

An obfuscated Flutter trace is bare instruction offsets. Here is how to resolve one to a function and a line by reading the DWARF tree yourself.

How TracePath records the ten seconds before a Flutter exception: a RepaintBoundary ring buffer, MP4 encoded only on error, and widget-level masking.

Web replay serializes DOM mutations. Flutter has no DOM. Here is the design space that leaves, what TracePath picked, and what the approach cannot do.

A source map recovers the exact file, line and column of a minified stack frame — but not the function name. Here is why, decoded by hand, and the fix.

An embedded row store ingests telemetry surprisingly well. The dashboard is what breaks first — and which query shape breaks tells you exactly why.

An honest map of the open-source observability landscape: what each stack is shaped like, what running it actually costs, and how to pick one.

A working loop: a production error opens a GitHub issue, an agent reads the telemetry behind it over MCP, and CI turns the result into a PR or a comment.