On-demand vs Editions, idle reservation baselines, slot contention and jobs queued for slots. We split 'wrong SQL' from 'wrong capacity', so you fix the meter when the meter is the waste.
The compute pricing model finding compares on-demand bytes to Editions slot-hours per project, minute by minute, counting idle baseline as free, and offers a daily schedule when busy and quiet hours want different models. Single dbt models and service accounts get their own pricing-model findings. Under-utilized reservations carry a list-price $/mo. Contention and queuing stay advisories.
BigQuery productComing soon: warehouse auto-suspend that is too long, XL size a dashboard never needed, and multi-cluster min that never scales to zero.
Snowflake productComing soon: serverless SQL warehouse auto-stop, classic warehouses left running, and DBU meters that do not match the query load.
Databricks productTimelines fail soft if the reservation admin project is out of reach.
Wrong meter? Floor too high? Pinned to the ceiling? Waiting to start?
Recommendations are capacity-side. The Cognis AI DBA can resize a reservation, move a project to the cheaper pricing model (once or on a daily schedule), or open a dbt pull request that moves one model, only after an owner approves. It never buys capacity.
system.query.history and system.compute.warehouses rank full scans, idle warehouses, and shuffle-heavy jobs. SlashPeak never needs SELECT on your tables.
Photon speeds up a shuffle join; it does not shrink it. Adaptive execution helps at the edges, but an unclustered join key still burns DBUs every run.
ZORDER and liquid clustering write min/max stats so Photon can skip files. The wrong columns, or a stale table, means it still reads everything.
A serverless warehouse with a 10-minute auto-stop bills 10 minutes after the last query. Oversizing is a separate leak, and system tables show both.
Understand, reduce and govern your cloud data spend with SlashPeak — starting today.