# Scheduling Instance Pool

> Pool schedules set minimum idle instances to 0, releasing the standby VMs a pool holds warm. Those machines bill cloud VM rates around the clock while earning no DBUs. Nothing is blocked overnight: clusters using the pool still start, they just cold-provision, and the saved minimum returns at the scheduled resume.

Source: https://zop.dev/integrations/databricks/databricks-pool/schedule
Updated: 2026-08-19

---

## The idle floor is pure standby spend

An instance pool's minimum idle count keeps that many VMs warm at all times so clusters can
grab them instantly. Those idle machines bill their cloud VM rate every hour. Databricks
charges no DBUs for idle pooled instances, but Azure or AWS charges for the VMs all the same.
A minimum of 8 mid-size instances held warm overnight is 8 VMs' worth of rent for latency
nobody is there to enjoy. The schedule drops that floor to 0 in the evening and restores the
saved value in the morning.

## Nothing stops working at night

Zeroing the floor blocks nothing. A cluster that requests instances from the pool at 2am still
gets them, freshly provisioned from the cloud rather than handed over warm. The trade is
start latency, minutes instead of seconds, paid only by whoever actually shows up at 2am. Of
all the schedules on this directory, this one is closest to free of failure modes: no
evictions, no lost state, no refused work.

## Warm mornings are part of the contract

The resume half restores the minimum ahead of business hours, so the pool re-fills with warm
instances before the first notebook attaches. Set the resume 10-15 minutes early, since the pool
needs time to provision its floor, and the day's first cluster start is as fast as it ever
was.

## Max capacity is untouched

Only the minimum moves. The pool's maximum stays where it was, so burst behavior during the
night window is identical to daytime; the schedule shapes the floor, never the ceiling.

## Pair it with the cluster and warehouse schedules

Pools rarely idle alone; the same workspace usually has abandoned interactive clusters and
always-on warehouses. The pool schedule removes the quietest of the three wastes, the one no
dashboard flags, because idle pooled VMs look intentional. Auditing the floor against actual
overnight cluster starts is usually the fastest few hundred dollars in the workspace.