geohashim

Reliability in production

Latency, limits, alerts and flaky tests, measured instead of assumed.

Seven repos that measure how software behaves in production instead of assuming it: request time budgets across services, rate limiters under skewed clocks and restarts, trace diffs that catch N+1 queries, alert rules replayed against past incidents, regression tests for noisy systems, flaky tests diagnosed by cause, and a real-time dashboard measured in the browser.

See the seven repos On the landing route
Repos
7
Passing tests
499, one dot each, run in October 2026 the way each repo's CI runs them
Languages
Java, Go, Python, TypeScript

    The seven repos

    Open a repo for its project page: what it does stage by stage, with real runs.

    • deadline

      49 passing tests

      Request time budgets across HTTP services

      A Java 21 library that passes a request's remaining time budget from service to service over HTTP, so each hop stops work nobody is waiting for, skips retries that can't finish, and reports where the time went.

      Language
      Java
      Stack
      Java 21 standard library only (JDK HttpClient and HttpServer), JUnit 5, Maven
      Counted with
      ./mvnw -B verify (39 test methods plus 10 parameterised cases)

      GitHub: deadline

    • rate-proof

      33 passing tests

      Rate limiters under clock skew and restarts

      Measures what six rate-limiting algorithms let through once deployed across nodes with skewed clocks, a remote store and restarts, and checks its simulator against a real Redis, decision for decision.

      Language
      Go
      Stack
      Go standard library only, Redis and Lua
      Counted with
      go test -count=1 -json ./... (2 of them need a real Redis, as in CI)

      GitHub: rate-proof

    • trace-diff

      37 passing tests

      Before-and-after OpenTelemetry trace diffs

      Compares OpenTelemetry traces from before and after a change and reports the N+1 queries, retry storms and new blocking calls that tests and latency dashboards miss, with trace ids to open.

      Language
      Go
      Stack
      Go, OTLP protobuf types; OpenTelemetry Collector and Postgres demo in Docker
      Counted with
      go test -race -count=1 -json ./... and make demo-test (21 tests plus 12 subtests, and 4 demo tests, one of which needs Postgres as in CI's demo job)

      GitHub: trace-diff

    • slo-proof

      71 passing tests

      Alert rules replayed against past incidents

      Replays past SLI data through a proposed alert rule and reports which incidents it would have caught, how late, what budget burned first, and how often it would have paged for nothing.

      Language
      Python
      Stack
      Python standard library only, Hypothesis, Prometheus range-query import
      Counted with
      pytest

      GitHub: slo-proof

    • noisefloor

      202 passing tests

      Regression tests for noisy systems

      A regression harness for non-deterministic systems that measures run-to-run noise before calling a change a regression.

      Language
      Python
      Stack
      Pydantic, PyYAML
      Counted with
      pytest

      GitHub: noisefloor

    • flaky-hunt

      47 passing tests

      Why each flaky test flakes

      Finds flaky pytest tests and the one factor that makes each fail, confirms it with a significance test in fresh processes, and hands back a plain pytest command that reproduces it.

      Language
      Python
      Stack
      pytest plugin, time-machine
      Counted with
      pytest -m "not slow" and pytest -m slow (42 fast tests plus 5 slow ones that run whole hunts, as CI splits them)

      GitHub: flaky-hunt

    • ops-console

      60 passing tests

      A real-time dashboard, measured in the browser

      A real-time Angular 22 dashboard that streams 5,000 services at 10,000 events a second, measures its frames, input delay and memory in Chromium, and works from the keyboard alone.

      Language
      TypeScript
      Stack
      Angular 22 zoneless with signals, no UI or chart library, Playwright, axe-core
      Counted with
      npm test and npx playwright test (32 unit tests plus 28 Playwright tests)

      GitHub: ops-console