0x74696d.com · Saved

0x74696d | Analytics on the Cheap

0x74696d.com/posts/analytics-on-the-cheap

From the page

Saved with the link. Notes from 2007 to 2016 are mostly the page's own words, so this is probably an excerpt.

The technique I’m describing has one important constraint: messages have to be either idempotent or have a built-in ordering (ex. time series data), because there’s no way for the ingest servers to handle the ordering of messages otherwise. You won’t be able to come up with a “true” total ordering here either, but we’re okay with that on a statistical level.