# Scheduling Azure SQL Database

> Azure SQL Database pauses only on the Serverless compute tier. Provisioned DTU and vCore databases have no pause at all. A calendar pause beats the native auto-pause whenever 1 lingering connection keeps the idle detector from ever firing, and a paused database bills storage while compute drops to zero.

Source: https://zop.dev/integrations/azure/sqldatabase/schedule
Updated: 2026-08-19

---

## The tier decides whether pause exists

Pause is a Serverless-tier capability, full stop. Provisioned databases, whether DTU or vCore, run
continuously by design and no schedule can pause them; their cost lever is scaling down, a
different operation with different latency. The first eligibility check on any Azure SQL
schedule is therefore the compute tier, and a fleet audit usually finds a mix.

Databases inside elastic pools are excluded for a different reason: the pool bills as a unit, and
pools have no pause operation at all.

## Auto-pause versus the calendar

Serverless already auto-pauses after an idle delay of at least 1 hour. The catch is the
definition of idle. A connection pool minimum of 1, a monitoring probe or an abandoned SSMS tab
is enough to reset the detector forever, and the database bills compute all night
while technically "auto-pausing". The scheduled pause is deterministic where the detector is
hopeful. Databases that reliably auto-pause need no schedule; databases that never quite
manage it are the target audience.

## Resume happens on first touch

A paused database wakes when a connection arrives, with a warm-up delay around a minute, and
the first connection attempts fail while it climbs. Applications need retry logic on connect,
which they should have anyway. A scheduled morning resume ahead of business hours converts
that user-facing stumble into an invisible one.

## Paused still bills storage

Compute billing reaches zero during a pause; storage billing does not, covering data and
backups at their normal rates. The saving on a lightly-used serverless database is the compute
line, which for a database idle 16 hours a day is the majority, though never the whole bill.

## Configurations that block pausing

Certain features pin a serverless database awake; long-term geo-replication arrangements are
the classic case. If the pause never seems to stick, check the feature list before blaming
the schedule; the database may be contractually unpausable in its current shape.