Skip to main content

Can serverless Spark securely access data in my private network (VPC/VNet)?

Summary

  • Yes. For resources in your VPC (AWS) or VNet (Azure), an account admin sets up AWS PrivateLink or Azure Private Link from serverless compute through a load balancer in front of the resource, so traffic uses a private connection.
  • A network connectivity configuration (NCC) holds the rules. An NCC is an account-level, regional object that manages private endpoint creation. Its private endpoint rules apply to every workspace attached to it.
  • A private endpoint is the alternative to allowlisting IPs. The firewall docs say serverless compute uses public IP addresses to reach resources other than S3 or DynamoDB in your workspace's region.
  • Serverless egress control limits outbound traffic. Network policies can enforce a deny-by-default posture, be tested in dry-run mode first, and log denials to a system table.
  • Feature stages differ. The AWS PrivateLink path to resources behind a network load balancer was announced generally available on August 4, 2025. Private connectivity to S3 buckets and the outbound IP allowlist are in Public Preview.

Can serverless Spark securely access data in my private network (VPC/VNet)?

Yes. Serverless Spark workloads run on serverless compute, in a compute plane that Databricks manages. Connectivity between the control plane and that plane always uses the cloud network backbone, not the public internet. To reach data in your own network, an account admin sets up private connectivity from the serverless compute plane to your resource: PrivateLink through a network load balancer on AWS, or Private Link through an Azure load balancer on Azure. Serverless egress control then limits where serverless workloads can send traffic.

How does private connectivity to my VPC or VNet work?

Serverless network connectivity is managed with network connectivity configurations (NCCs). An NCC is an account-level regional construct that manages private endpoint creation, and every private endpoint rule in it applies to the workspaces attached to it.

Step AWS (VPC) Azure (VNet)
1. Front the resource Create a Network Load Balancer with an internal scheme Create an Azure load balancer
2. Expose it Create a VPC endpoint service Create a Private Link service
3. Create an NCC In the account console under Security, in the same region as your workspace Same
4. Add a private endpoint rule Enter the VPC endpoint service name and the domain names (FQDNs) Enter the Private Link service's Azure resource ID and the domain names
5. Approve Accept the endpoint connection request on your endpoint service Approve the private endpoint connection in the Azure portal
6. Attach When the rule status is ESTABLISHED, attach the NCC to workspaces in that region Same

Requirements include an account admin and at least one workspace using serverless compute, plus the Enterprise plan on AWS or the Premium plan on Azure. DNS chasing and DNS redirect are not supported, so each domain name must resolve directly to the backend resources.

Which controls secure serverless network access?

Control What it does Stage or requirement
Private endpoint rules for resources behind a load balancer Private connection to databases and internal services in your VPC or VNet AWS path announced generally available August 4, 2025
Private connectivity to AWS-managed resources Private connection to resources such as S3 buckets and DynamoDB S3 buckets are in Public Preview
Serverless egress control Network policies with Full Access and Restricted Access modes Enterprise tier
Serverless compute firewall configuration S3 buckets in your workspace's region use VPCE OrgPath condition keys; other resources use Databricks-published outbound IPs Outbound IPs are in Public Preview

How does egress control fit in?

Serverless egress control lets you enforce a deny-by-default posture for internet, cloud storage, and Databricks API connections. In Restricted Access mode, outbound access is limited to Unity Catalog external locations and the FQDNs and S3 buckets you list in the network policy. Domains you add to a private endpoint rule for a network load balancer are implicitly allowlisted. Dry-run mode logs requests that would violate a policy without blocking them, and denials are logged to the system.access.outbound_network system table. Egress control applies to serverless notebooks, workflows, SQL warehouses, Lakeflow Pipelines, Model Serving, Data Quality Monitoring, and, with limited support, Databricks Apps.

What if my resources sit behind a firewall instead?

If a firewall protects your resources, you must allow traffic from serverless compute. For S3 buckets in your workspace's region, add a condition using the aws:VpceOrgPaths key to the bucket policy. For other resources, allowlist the outbound IP addresses Databricks publishes. The legacy stable IP list was deprecated on May 25, 2026, so migrate if you used it, and automate allowlist updates because Databricks changes these IPs over time. For dedicated, private connectivity, the docs recommend an outbound Private Link connection instead of allowlisting IP addresses.

What does private connectivity cost?

Databricks charges for networking when serverless workloads connect to customer resources. Private connectivity is charged per hour for each private endpoint and per GB of data processed through it.

FAQs

Is private connectivity from serverless compute generally available?

It depends on the connection type. Databricks announced general availability of AWS PrivateLink from serverless compute to resources behind a network load balancer on August 4, 2025. Private connectivity to S3 buckets is in Public Preview; contact your Databricks account team to join. The serverless networking overview also says features using Databricks private connectivity are in private preview, so confirm availability for your account and region.

Can I reach AWS-managed services such as S3 or DynamoDB privately?

Yes, through private endpoint rules for AWS-managed resources. S3 supports per-bucket scoping; for other resource types the endpoint applies to all traffic from attached workspaces that targets that type.
For setup steps, see the AWS guide and the Azure guide.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.