Blog

Notes from building Runnev

Engineering writing about event streaming over HTTP: the transport decisions, the operational realities of long-lived connections, and the delivery guarantees we actually offer.

• Sizing ring buffers: the retention math

  2026-05-077 min read

  Bytes per event, eviction, and how to pick max_bytes and retention_seconds so the two bounds tell the same story, with a worked calculation.

• Cursors, sequences, and the delivery guarantee we actually offer

  2025-11-186 min read

  Why we say at-least-once with idempotent writes instead of exactly-once, and how committing your cursor after the side effect gets you exactly-once processing anyway.

• What a 40-hour HTTP response taught us about intermediaries

  2025-06-128 min read

  The buffers and idle timeouts between a client and us, and the specific settings in nginx, Caddy, Cloudflare, and an AWS ALB we had to change before very long responses stopped resetting.

• Raw batch mode: shipping protobuf without a schema registry

  2025-01-225 min read

  Why we stopped parsing customer payloads, what the octet-stream ingest path looks like now, and the throughput we got back by not doing a parse we never needed.

• Why we did not build this on WebSockets

  2024-10-087 min read

  The upgrade handshake failing behind corporate proxies and some CDNs, serverless runtimes that cannot hold a socket, mobile radios dropping idle sockets, and what HTTP/2 gives you instead.