Kill Three Pillars

Nope! Traces are the most valuable of the three data types. Why? They are an almost ideal candidate for context aware, wide events.

The #kill3pill movement isn't about killing the data types of traces, logs and metrics. They're still a valuable part of the observability discipline.

It's about killing the idea that they're pillars. Pillars are stand alone. So we end up with three silos, which makes for expensive tooling and a fragmented user experience for engineers.

Well... maybe. Structured logs are still valuable. They can be your "wide events" - they're just not great at holding context.

They have some advantages over traces though - they're cheaper, have more tooling around them and they're more widely understood. Debatably...

Logs can also be derived from traces. The #kill3pill movement is not about removing the data types. It's about removing the idea that they're pillars, which leads to siloed data.

Not quite. Sending metrics directly to any observability tool isn't a great idea.

They tell you the what - "this system is not performing well..."

But they don't tell you the why - "...because the queue has been flooded with bad data."

That's fine - it's not their role. But derive metrics from deeper more rich data sources such as logs or traces. Then you'll have way more flexibility in your metrics and you'll be able to understand the why behind incidents.