#performance
Latency and throughput
- Connection pooling vs read replicas for a 50k-member forum
We used to raise `max_connections` whenever p99 login latency spiked. After about 4_000 concurrent sessions that stopped working — the wait moved to row locks on `topic_subscript...
- Should we store markdown, HTML, or both?
TalkMoro stores `body` and `body_html`. Re-rendering on every read is a non-starter. Re-rendering on every edit is cheap. The expensive case is a renderer change (heading ids, tas...
- What we learned from 90 days of community SLOs
SLOs we actually page on: - API 5xx under 0.5% - Search p95 under 120ms - Upload 2xx under 2s for files ≤ 5 MB (5_000_000 bytes) We do not page on posts per minute. That is a vanity chart. On-call is for 5xx, not for unread comments.
- Anonymous read of public topics without a session cookie
Some campus proxies strip cookies. Anonymous HTML should not depend on a session id for the first paint. This is also the AI-crawler path.