benchmarking – Page 3 – Türker Yıldırım

Mutation testing for one directory, and what it found

Deep diveOctober 24, 202310 min read

94% line coverage on the pricing code, and a bug that shipped anyway. A test that called the method and asserted it did not throw.

The alert that fired for two years and meant nothing

Deep diveOctober 18, 202310 min read

Thirty-one alert rules, a channel muted by everybody, and 412 alerts in a month of which four corresponded to a real problem.

A benchmark that measured the benchmark

NoteAugust 24, 20232 min read

A microbenchmark showing one implementation four times faster than another, where the difference was the loop. Three separate errors in six lines: an unused result, a shared warm…

The N+1 that only appeared in production

Deep diveJuly 29, 202310 min read

An endpoint at 40 ms locally and 2.1 seconds in production. Nine hundred queries on one request, and three on the same request in a test.

Partitioning by month, and the query that scanned all of them

Deep diveJune 27, 202310 min read

A 400 million row events table and a delete that could not finish. Dropping a partition takes two seconds; getting there takes a schema change.

A suite that runs in four minutes because it stopped touching the database

Deep diveMay 28, 202310 min read

1,400 tests, fourteen minutes, and 1,100 of them opening a transaction against MySQL for no reason anybody could name.

PHPUnit 10 is faster, and the number is smaller than claimed

NoteApril 20, 20232 min read

The release notes talk about a significant speed improvement, and on a suite dominated by database work the improvement is in the part that was never slow. The…

A container we replaced with four functions

Deep diveApril 12, 20238 min read

A package with a dependency injection container, for eleven services, spending 90 ms compiling on a request that does 12 ms of work.

Percentiles over averages, again, with a graph

NoteMarch 29, 20232 min read

An average response time of 180 ms describing a distribution where a fifth of requests take over a second. The correlation between the slow tail and the valuable…

A read replica, and the six queries that could not use it

Deep diveMarch 16, 202314 min read

A primary at 80% CPU and reports running during the working day. Setting up the replica takes an afternoon; the consistency model takes a month.