Organizations often experience highly variable workload demand throughout the day. Sizing a virtual network data gateway for peak demand can increase costs, while sizing for average demand can lead to congestion during spikes. Autoscaling for Virtual Network Data Gateway (Preview) automatically adjusts gateway capacity based on workload demand while helping balance performance and cost.
VNET Data gateways allow Microsoft Fabric workloads (data factory, data flows, copy job etc.) to securely connect to data sources inside a customer-managed virtual network. As usage grows, gateway demand can change throughout the day — from quiet periods with little activity to sudden spikes from refresh operations, Direct Query workloads, pipelines, or copy jobs. Until now, Gateway admins had to plan gateway capacity more statically, which could lead to either under-provisioning during busy periods or unnecessary cost during idle periods.
With autoscaling, the gateway can dynamically adjust the number of active gateway nodes within administrator-defined minimum and maximum limits based on workload demands.
VNET Data Gateway Autoscaling is designed to:
- Handle workload spikes without manual intervention.
- Improve reliability and reduce failures caused by resource constraints.
- Optimize costs by scaling down during periods of low or no activity.
- Align gateway capacity with actual usage while maintaining VNet isolation and security controls.
Autoscaling signals
Autoscaling for Virtual Network Data Gateway intelligently adjusts gateway capacity based on workload demand, helping ensure consistent performance while optimizing resource utilization. The service monitors workload activity and automatically scales gateway instances when sustained load increases and scales them down when demand remains low. Different workload types are evaluated using workload-specific signals to ensure responsive scaling behavior, particularly for data movement and query-intensive scenarios.
The autoscaler samples a few key metrics and signals every five minutes. When sustained values cross the threshold, the gateway scales out. When sustained values fall below the scale-in threshold, the gateway scales in. The scaling frequency cannot be tuned.
Autoscaling always stays within the minimum and maximum node caps configured by the gateway administrator; node caps depend on the IP’s availability in the subnet.
Lifecycle controls
Autoscaling also works alongside lifecycle controls that help gateway admins manage gateway availability and cost.
Gateway Admins can:
- Start, stop, or restart the gateway cluster.
- Set minimum and maximum node limits.
- Configure auto-shutdown after an idle period.
- Monitor active node usage and scaling behavior.
Autoscaling manages the number of running gateway instances only after the gateway is online. If the gateway is offline, select More Actions → Start to bring it online. A dataplane request can also automatically start the gateway. After startup, the gateway may initially start with the configured max cluster size and then automatically scale down or up based on workload, subject to the configured TTL settings.
Monitoring
Gateway administrators can monitor active gateway instances directly from the Manage Connections and Gateways experience. Diagnostics logs provide historical visibility into node counts, workload activity, and capacity utilization trends.
Learn how to download logs and analyze by exploring the documentation, Download logs on the virtual network data gateway.
Get started
To use autoscaling, create or update a virtual network data gateway and configure the gateway node range in advanced settings. Set the minimum and maximum number of nodes based on expected workload demand, subnet capacity, and cost requirements.
1. Open the Fabric admin experience and go to manage gateways and connections. Select the VNET Data Gateway that you want to configure auto scale.
2. Manage connections and gateways for Virtual network data gateways.
3. Select settings and go to advanced options. Set minimum and maximum nodes for the gateway instance.
4. Configure the inactivity timeout, which determines how long the gateway remains online without workload activity before automatically stopping.
5. Save the changes—then you’re done!
Example scenario
A gateway starts with seven nodes (max instances). Nodes will scale down based on actual usage. A large copy job increases CPU usage and queue length for several sampling intervals. Autoscaling adds nodes, up to the configured maximum, so queued work can progress. After the job finishes and demand drops, autoscaling reduces the node count. If the gateway remains idle for the configured TTL period, the gateway stops and billing pauses.
Ready to try it?
Configure autoscaling on an existing Virtual Network Data Gateway and review diagnostics logs to understand how capacity adjusts to your workload patterns.
Share your feedback, use cases, and questions in the Microsoft Fabric Community. Your input directly shapes the roadmap.