# Scheduling Azure Data Explorer (Kusto) Cluster

> Kusto cluster schedules issue stop only when the cluster reports Running, and anything else is soft-skipped. While stopped, ingestion halts, so Event Hub sources must be resumed within their retention window or events age out; starts take well over 10 minutes, and the hot cache rebuilds from scratch afterward.

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

---

## The state gate comes first

Stop and start are only attempted against a cluster in the Running state. Mid-scale, updating
or already-stopped clusters are skipped rather than forced, so the schedule cannot corrupt an
in-progress operation, but a skipped stop also means a night at full price. Watch the first
week's runs to confirm the cluster is actually cycling rather than perpetually busy at the
stop hour.

## Ingestion stops, retention clocks keep running

A stopped cluster ingests nothing. Data sitting in upstream buffers waits, and each source has
its own patience: Event Hubs hold events only for their configured retention, typically 1-7
days. An overnight window is comfortably inside that; a stop that stretches across a long
weekend plus a holiday starts flirting with the limit, and events that age out are gone
without an error on the Kusto side. Confirm the source retention exceeds the longest scheduled
gap by a margin.

## Cold cache, slow first queries

Kusto's interactive speed comes from the hot cache on local SSD, and a started cluster begins
with it empty. Queries fall back to storage until the caching policy repopulates, so first-hour
dashboards run visibly slower than their afternoon selves. Warm-up queries after the start
cron, touching the hot tables in rough priority order, turn a rough first hour into a rough
first few minutes.

## Starts are measured in tens of minutes

Bringing a Kusto cluster up is heavyweight: expect well over 10 minutes from start to
queryable, longer for large clusters. The start cron belongs an hour before the first analyst,
not fifteen minutes.

## What stays on the bill

Compute engine and data-management nodes stop charging. The storage layer underneath, where
all ingested data lives, bills continuously, as do any follower-related resources outside the
cluster. The saving is the compute line, which for interactive analytics clusters is
fortunately most of the total.