Data and Object Management · Topic 2 of 11

Multi-Cluster Warehouses - Use Cases, Scaling Policy and Monitoring

Manage multi-cluster warehouses: when they help, Maximized versus Auto-scale mode, Standard and Economy scaling policy, cluster limits and monitoring.

  • 5 min
  • Includes a diagram
In a nutshell
  • Multi-cluster warehouses fix concurrency and need Enterprise Edition or higher.
  • Maximized: min equals max. Auto-scale: max greater than min.
  • Scaling policy applies to Auto-scale only; STANDARD is the default and ECONOMY keeps clusters full (6 minutes of work).

What it is

A multi-cluster warehouse is one warehouse that can run several clusters of the same size at the same time. Snowflake adds clusters when queries queue and removes them when load drops.

It solves concurrency: many users or many short queries at once. It does not make one large query faster. It needs Enterprise Edition or higher.

Many concurrent queriesQueue formsExtra cluster startsup to MAX_CLUSTER_COUNTQueue clearsIdle cluster stops

How it works

Use cases and benefits

SituationMulti-cluster?Why
Hundreds of BI users at peak hoursYesQueries run in parallel instead of waiting
One huge nightly transformNoA larger size helps, more clusters do not
Unpredictable user loadYes, Auto-scalePay only for clusters that run
Load that is always highYes, MaximizedAll clusters are ready, no ramp up delay

Benefits are steady response time at high concurrency and cost that follows demand. The warehouse still bills per second for each running cluster.

Configuration and modes

Cluster count is set with MIN_CLUSTER_COUNT and MAX_CLUSTER_COUNT. Max must be greater than 1 for a multi-cluster warehouse, and min must be equal to or less than max.

ModeSettingBehavior
Maximizedmin equals max (greater than 1)All clusters start when the warehouse starts
Auto-scalemax greater than minSnowflake starts and stops clusters based on load
CREATE WAREHOUSE reporting_wh
  WAREHOUSE_SIZE = 'MEDIUM'
  MIN_CLUSTER_COUNT = 1
  MAX_CLUSTER_COUNT = 4
  SCALING_POLICY = 'STANDARD';

The maximum allowed depends on the size: 300 for X-Small, Small and Medium, falling to 10 for 4X-Large and above. Snowsight allows at most 10; higher values need SQL. You can change min and max at any time, even while the warehouse is running. Scaling down removes the extra clusters only after they finish their statements.

Scaling policy (Auto-scale mode only)

!A multi-cluster warehouse runs in Maximized mode, where min equals max and every cluster starts, or Auto-scale mode, where max is greater than min. Only Auto-scale uses a scaling policy: STANDARD (default) prevents queuing and ECONOMY saves credits by starting a cluster only if it has at least 6 minutes of work.

PolicyGoalStarts a new clusterShuts a cluster down
STANDARD (default)Prevent queuingSoon after queries start to queueAfter a sustained period of low load
ECONOMYSave credits, keep clusters fullOnly if there is enough work to keep it busy for at least 6 minutesWhen less than 6 minutes of work remains for it

Economy accepts some queuing in return for fewer started clusters. Choose it for workloads where a short wait is fine and credits matter. Choose Standard when response time matters.

To change the policy: ALTER WAREHOUSE reporting_wh SET SCALING_POLICY = 'ECONOMY';. The policy has no effect in Maximized mode, because every cluster is already running.

Exam tip

A scenario that says "minimize cost, queries may wait a little" points to Economy. "Users must never wait" points to Standard, or to Maximized if the load is always high.

Credit use

The most a multi-cluster warehouse can use in an hour is the credit rate of its size multiplied by the maximum cluster count. A Medium warehouse with 3 clusters can use 12 credits in a full hour. Actual use depends on how many clusters ran and for how long.

Docs recommend Auto-scale for most cases, a minimum of 1 unless high availability needs more, and a maximum as high as your cost limits allow.

Monitoring

WhereWhat you see
Snowsight Compute, WarehousesMin, max and running cluster counts, running and queued queries
SHOW WAREHOUSESmin_cluster_count, max_cluster_count, started_clusters
Query History in SnowsightCluster number for each query when filtered by warehouse
WAREHOUSE_LOAD_HISTORYRunning and queued load per interval
WAREHOUSE_EVENTS_HISTORYCluster start and stop events
WAREHOUSE_METERING_HISTORYCredits used by the warehouse

If the warehouse is always at its maximum cluster count and queries still queue, raise the maximum or split the workload. If it never goes above the minimum, the maximum is too high or the minimum is wasting credits.

Queuing seenWAREHOUSE_LOAD_HISTORYCheck started clustersSHOW WAREHOUSESAt max?Raise MAX_CLUSTER_COUNTor change policy

Common mistakes

Using multi-cluster for one slow query. Clusters do not split a single query. Use a larger size.

Expecting Economy to behave like Standard. Economy waits for 6 minutes of work before adding a cluster, so queuing is normal.

Choosing Maximized for spiky load. All clusters run and bill from the start. Auto-scale fits spikes.

Forgetting the edition. On Standard Edition, multi-cluster warehouses are not available.

What to remember for the exam

  • Multi-cluster warehouses fix concurrency and need Enterprise Edition or higher.
  • Maximized: min equals max. Auto-scale: max greater than min.
  • Scaling policy applies to Auto-scale only; STANDARD is the default and ECONOMY keeps clusters full (6 minutes of work).
  • Maximum hourly credits are rate times maximum cluster count.
  • Monitor with Snowsight, SHOW WAREHOUSES, load history and metering history.

Check yourself

  • Which mode starts every cluster when the warehouse starts?

    Maximized, where min and max cluster counts are equal and greater than 1.

  • A team wants the fewest credits and can tolerate some queuing. Which scaling policy?

    ECONOMY.

  • A Large warehouse (8 credits per hour) is set to max 3 clusters. What is the most it can use in an hour?

    24 credits, the size rate times the maximum cluster count.

Topic 2 of 11 in Data and Object Management

Test yourself on Data and Object Management

The full Snowflake ADA-C02 question bank has 814 questions, with timed practice exams and an explanation for every answer.

This study guide is independent and is not affiliated with, endorsed by, or sponsored by Snowflake. Vendor names, certification names and exam codes are trademarks of their respective owners and are used for identification only.