"We do push the event model as an integration abstraction between logs, traces and metrics." 🙌📈🌈
~ @rictomm, studying observability from a scientific point of view with his student @mario_scrock (and officially making my day).
~ @rictomm, studying observability from a scientific point of view with his student @mario_scrock (and officially making my day).
"As an integration abstraction", events as a bridge. That is a fascinating way of unifying the worldviews and it makes perfect sense.
🤔
🤔
Metrics are shit because you have discarded all the connective tissue of the event, so you can always derive metrics and aggregates from events and never vice versa.
Logs are just primordial events, begging for structure.
Tracing is just a particular visualization for events.
Logs are just primordial events, begging for structure.
Tracing is just a particular visualization for events.
The rich, structured data blob describing the extended context of what executed, when, and what it's environment looked like as possible -- it's so clearly the correct and irreducible way we should be storing bits on disk.
Events uber alles.
Events uber alles.
Looking forward to talking more about their research this week while I'm in London, and arming them with loads of real world examples. ☺️🌷
P.S. dynamic weighted sampling is an absolutely critical corollary skill. Almost *everything* that happens on a busy system is boring. It's interesting only in aggregate. Except when it isn't.
So keep more of the events that are more likely to be "interesting".
So keep more of the events that are more likely to be "interesting".
Don't be like the jokers who are like "SAMPLING DROPS DATA" while they truck in aggregates and discard all context. 🙄
If you can't drill down to raw events, it's not observability, because you're only asking canned questions for predictable problems.
If you can't drill down to raw events, it's not observability, because you're only asking canned questions for predictable problems.
