Restricting Network Access to Snowflake

Consistent network access for a distributed team and the automation behind it.

By Bernie Pruss
Restricting Network Access to Snowflake - Cloud Data Consulting

Snowflake network policies let connections in only from approved public Internet Protocol (IP) addresses. That works for an office with a static IP. It breaks when your IP address changes—from a coffee shop, hotel, or cell hotspot, or when a provider changes your office or home address—or when a GitHub Actions runner connects from a new address every run: the login and role are fine, but Snowflake blocks the connection.

We fixed this with Tailscale. Snowflake traffic from our workstations and automated jobs goes out through a small Amazon Web Services (AWS) server whose fixed address is on the allowlist, while everyday browsing stays on the local connection.

Network overview: a remote laptop reaches Snowflake through a Tailscale app connector on Amazon EC2, while everyday browsing uses the local internet connection.

An Amazon Elastic Compute Cloud (EC2) server in Northern Virginia and a Snowflake account in the same region. Transport Layer Security (TLS) encrypts the connection from the server to Snowflake. The route uses an app connector, explained below.

What Snowflake checks

A network policy lists the public IP addresses, or ranges of addresses, that are allowed to connect. This list is usually called an allowlist, also called a whitelist. Adding an IP address to the list is called whitelisting that address. A public IP address identifies where a connection comes from, as Snowflake sees it. Several computers in one office may share an address because their traffic leaves through the same router. The allowlist limits which addresses can connect; a person on an approved address still signs in, and their Snowflake role still determines what they can read and change.

In Snowflake, a network rule holds a group of addresses, such as the public addresses of our AWS servers. A network policy lists the rules to allow and can also list rules to block. Our setup reaches Snowflake at its public hostnames over the internet. Snowflake also supports private connectivity (AWS PrivateLink), which this post doesn’t cover.

Other tools need access too

The allowlist also has to cover the tools that connect to Snowflake for us. Fivetran loads data, Coalesce runs transformations, and Omni queries Snowflake for analysis. When they connect, Snowflake sees their outbound addresses.

Each vendor documents those addresses. Fivetran lists them by cloud and region, Coalesce publishes its network requirements, and Omni shows its addresses on the connection settings page. We add the addresses each vendor lists for our region.

When we open an Omni dashboard, our browsers talk to Omni, and Omni’s servers query Snowflake. Snowflake sees Omni’s address, so that address belongs on the allowlist even when our workstations use Tailscale.

Office, home office, and travel

An office has a stable public IP when its internet service includes a static address. A dynamic home address may stay the same for months and then change. Each hotel, client office, or mobile hotspot we connect through can present another public address to Snowflake.

We could ask an administrator to update the allowlist each time. Instead, we send Snowflake connections through an AWS server whose public address stays the same when we move.

Where Tailscale fits

Tailscale is a Virtual Private Network (VPN). It creates encrypted connections between our devices and AWS servers. A workstation can be a laptop, a desktop, or a virtual machine in the cloud. We install Tailscale on the devices that need to reach Snowflake and configure them to send that traffic through our AWS servers.

At Cloud Data Consulting, we’ve had good experience using a minimal Amazon EC2 server configuration to run Linux and Tailscale. Our east-coast server is a t3.nano running Amazon Linux, with 512 mebibytes (MiB) of memory. It uses an Elastic IP—a public address we keep even when we replace the server. Snowflake sees that address, so that’s what we add to the allowlist. We leave the server running so workstations and automated jobs can connect whenever they need Snowflake.

We allocate the Elastic IP separately from the server. Our Tailscale Terraform module attaches that existing address to the server. Destroying the module removes the server and detaches the address, but does not release it.

Exit nodes and App Connectors

The same server can provide two ways to route traffic.

Exit nodes

An exit node routes all internet traffic through the selected server. In the Tailscale application, open Exit Nodes and choose a machine. Selecting our server in us-east-1, Northern Virginia, makes Snowflake see that server’s approved public address. A more-specific app-connector route takes precedence for the destinations it covers.

Other websites see the exit node’s address too. With the Northern Virginia exit node selected, BestBuy.com showed us stores near that server instead of stores near us. We also encountered extra sign-in checks, multi-factor authentication (MFA), and grids of fuzzy photos asking us to pick out cars or motorcycles.

App connectors

An app connector routes traffic for configured domain names through the server assigned to them. With the exit node turned off, we can send Snowflake traffic through AWS while everyday browsing uses our local internet connection.

We assign domains to a connector tag, a label on the server that should carry their traffic. Separate regional tags let us choose a server near each Snowflake account. If multiple servers have the same tag, each can serve those domains.

Choose a connector in the region hosting each Snowflake account. The same pattern applies wherever the accounts are hosted; Northern Virginia and Oregon are two examples:

Snowflake account locationApp connector locationTraffic path
Northern Virginia, us-east-1AWS us-east-1Workstation → east connector → east Snowflake account
Oregon, us-west-2AWS us-west-2Workstation → west connector → west Snowflake account

Tailscale does not infer the region from the account name. We choose the domain and tag mapping so the connection avoids a cross-country detour.

Assigning domains to regional connectors

The mapping lives in the Tailscale admin console, under Apps or in the nodeAttrs section of Access controls. In Apps, choose Add an app, name it, select Custom, and enter the domains. Under Connectors, choose the server’s tag and save. Tailscale distributes the configuration to connected devices.

Here is a policy fragment with separate east and west mappings. It uses clouddata-prd for the east account and clouddata-prd-west as an example west account. The tags are examples; replace the domains and tags with your own:

{
  "nodeAttrs": [
    {
      "target": ["*"],
      "app": {
        "tailscale.com/app-connectors": [
          {
            "name": "snowflake-east",
            "connectors": ["tag:snowflake-east"],
            "domains": ["clouddata-prd.snowflakecomputing.com"]
          },
          {
            "name": "snowflake-west",
            "connectors": ["tag:snowflake-west"],
            "domains": ["clouddata-prd-west.snowflakecomputing.com"]
          }
        ]
      }
    }
  ]
}

domains holds the Snowflake hostname, connectors holds the server’s tag, and name labels the entry.

tag:snowflake-east belongs on the server in us-east-1, and tag:snowflake-west on the server in us-west-2. Each Snowflake account’s allowlist must permit the public address of the server handling its connections.

Tailscale’s setup guide covers permissions, route approval, and Domain Name System (DNS) settings. DNS translates a domain name into an IP address. Linux devices must accept the connector’s routes with tailscale set --accept-routes=true.

Snowsight and some Snowflake operations use additional hostnames, so include the names your tools use. Listing account hostnames separately lets us choose a region for each. A broad wildcard such as *.snowflakecomputing.com would send many accounts through the same mapping. A shared hostname such as app.snowflake.com has one mapping; Tailscale cannot choose a region from an account name later in the web address.

App connectors route to the IP addresses DNS returns for the configured domains. If two domains resolve to the same IP address, traffic to both goes through the connector—even if only one domain is listed. The route matches the destination address, so it cannot distinguish the two domain names.

The same problem applies to automation

Our GitHub Actions workflows run dbt jobs that create and update objects in Snowflake as part of our CI/CD deployment process. This is continuous integration and continuous delivery or deployment (CI/CD). A hosted runner, the computer executing the job, may have a different public address each time it runs.

Each job gets its own Snowflake login and role, then joins Tailscale and selects our EC2 instance as its exit node. Snowflake sees the instance’s Elastic IP whichever runner GitHub assigns. Other CI/CD platforms face the same address check.

GitHub Actions connects to Snowflake through an AWS EC2 exit node.

GitHub Actions reaches Snowflake through the EC2 instance and its approved public address.

What does it cost?

One Linux t3.nano running all month in AWS us-east-1, with its public Internet Protocol version 4 (IPv4) address, costs about US$7.45 per month before storage and usage charges. The calculation uses 730 hours and published prices checked on October 1, 2026:

ItemPublished hourly priceCost for 730 hours
Linux t3.nanoUS$0.0052US$3.80
One public IPv4 addressUS$0.005US$3.65
SubtotalUS$7.45

Disk storage, billable data transfer, and any extra processor-usage charges come on top. A second server adds another instance, public address, and disk.

Tailscale is billed separately. Its published Standard plan is US$8 per user per month, and Premium is US$18; the free Personal plan is for non-commercial use. Standard lists 1,000 minutes a month for ephemeral resources such as short-lived CI runners, and Premium lists 10,000. The pricing links below have current rates.

Checking that it works

Test an exit node

Open WhatIsMyIP.com with the exit node turned off. It should show your local public IP address. Select an exit node and refresh the page; the address should change to that server’s Elastic IP. Repeat for each exit node you want to test. This assumes the lookup site itself is not routed through an app connector or another VPN.

Test the Snowflake connection

Run this query in Snowsight, the Snowflake command-line tool (snow), or any other tool that can run SQL against Snowflake:

SELECT CURRENT_IP_ADDRESS();

The result should match the Elastic IP of the server assigned to that account. To test an app connector, leave Tailscale connected and turn off the exit node before running the query. The same query can check the connection from an automated job. This function reports IPv4 addresses.


Need help configuring Snowflake network access for your team and automated jobs? Talk with Cloud Data Consulting.

References

Share: