Comparison
Idle computevsWarehouse sizing
Idle compute
the cluster stayed up all weekend waiting for queries that never came, and the invoice noticed.
Compute that is running and billing while doing nothing. It is the most embarrassing line on a warehouse bill and the easiest to remove, with a suspend timeout that is short enough to matter and long enough not to cold-start every dashboard tile. The tension with cold starts is real, which is why the setting is per-warehouse and why interactive and scheduled workloads usually want different ones.
Full entry →Warehouse sizing
you doubled the warehouse size, the query took half as long, and the bill was identical.
Choosing how much compute a workload gets, on a platform where cost is usually size multiplied by time. For a query that parallelises cleanly, a bigger warehouse is cost-neutral and finishes sooner; for one bottlenecked on a single skewed task it is pure waste. Sizing is therefore diagnostic — if doubling the size does not halve the time, the problem is skew or spill, and no amount of compute will fix it.
Full entry →