Thread by @mipsytipsy

This. Your systems exist in a continuous state of partial degradation. Unless -- and only unless -- they are completely down.

https://twitter.com/tleam/status/1096334126918053890

An under-explored reality that undergirds @honeycombio's existence is this: errors can only be
_managed_ in the aggregate, and can only be _investigated_ in the particular.

Think about that for just a sec.

You probably have a monitoring system that stores metrics, tags and aggregates in a time series db, and alerts on the health of the system.

And you probably have a bunch of logs and exception trackers that describe individual errors in detail.

And what connects the two?

For most shops, the answer is "human eyeballs". Ops, usually. It's hard to express how much signal is lost by hopping back and forth between systems mid-question.

And now people are adding a third hop: tracing. 😖 Painful. And did I mention how much this all COSTS? 💸💸💸

Debugging stops being painful and yields much more interesting results once you can switch modalities without switching systems:

- find interesting trend
- look at examples
- what do they have in common?
- who is affected?
- is the set dominated by a single user/other dimension?

See how many times I flip back and forth between modes while exploring a single question? This is why @honeycombio users go nuts over us. They don't have to understand all the underlying theory, they just know:

"debugging just got way the fuck easier"

We have a free tier for personal use: take it for a spin. All you need to do is add a beeline lib to your app. E.g. docs.honeycomb.io/beeline/go/ 🐝

If that's too hard for a Saturday, check out honeycomb.io/play -- we have our last outage in a sandbox for you to play in.