note 001 · planned
Reading skew from the Spark UI is a scavenger hunt. Here’s the plan-level shortcut.
Why task-duration histograms lie, what AQE already knows, and the three fields in the event log that make skew detection boring.
ENGINEERING NOTES
Nothing published yet. On purpose.
We would rather ship no posts than launch-week filler. These are the first three on the bench:
note 002 · planned
The real cost of a DBU: a worked attribution model
From job_run to team rollup with every intermediate step shown — including the ugly parts: spot interruptions, shared clusters, retries.
note 003 · planned
Small files: how 1.9 MB averages quietly 4×’d a customer’s runtime
A 90-day forensic on OPTIMIZE debt, with the Delta log statistics that would have caught it in week two.
want these when they land? → hello@telemetria.ai