Skip to content
Notifications
Clear all

Real experience with NordLayer for a Fortune 500 remote workforce

3 Posts
3 Users
0 Reactions
86 Views
(@ci_cd_mechanic_7)
Honorable Member
Joined: 7 months ago
Posts: 410
Topic starter   [#14575]

We rolled out NordLayer to our engineering teams last quarter, ~2000 developers globally. Goal was to secure access to internal build systems and artifact repositories without killing pipeline velocity.

**The good:**
* Setup was straightforward. SCIM integration with Azure AD worked as advertised.
* Split tunneling configuration is critical. We route only internal traffic (10.0.0.0/8) through the tunnel. This prevents public CI/CD services (like GitHub Actions) from being blocked.
```json
// Example NordLayer split tunnel config (simplified)
{
"routes": [
"10.0.0.0/8",
"192.168.100.0/24"
],
"exclude_routes": ["0.0.0.0/0"]
}
```

**The bad:**
* Unpredictable latency spikes. Jenkins agents in EU would occasionally timeout pulling from US-based Artifactory, adding 10+ minutes to build times.
* Agent auto-scaling in AWS became a problem. New instances sometimes failed to establish the tunnel, requiring a retry loop in our user-data script.
* Logging is insufficient for debugging pipeline connectivity issues. Had to rely on our own instance monitoring.

It works for standard VPN use cases, but for high-throughput, low-latency CI/CD traffic, we've had to implement aggressive caching and fallback direct connections. Would not recommend for large-scale build infrastructure without significant tuning.



   
Quote
(@alexh3)
Reputable Member
Joined: 4 months ago
Posts: 254
 

Interesting that you mention latency spikes with Artifactory. We had a similar issue, though with Nexus repositories, and it wasn't consistent across providers. Our team in Singapore connecting to Frankfurt saw it, but the London to Frankfurt path was stable. It points to potential routing table inefficiencies within NordLayer's network backbone for certain geographies.

Your point about the auto-scaling failure is crucial. That's a deal-breaker for immutable infrastructure patterns. We had to implement a similar health-check and retry in our Terraform provisioner, which adds complexity. It feels like their agent isn't designed for headless, automated initialization under load.

I'm curious if you explored their gateway placement options. We found designating a specific gateway for CI/CD traffic, as opposed to using "auto-connect," reduced some variability, but then you lose fault tolerance.


Data is the source of truth.


   
ReplyQuote
(@git_ops_guy)
Reputable Member
Joined: 7 months ago
Posts: 399
 

Yeah, the split tunneling config is key for CI. We had to do something similar but also had to explicitly allow our internal DNS IPs, otherwise container pulls from ECR would hang mysteriously. The agent just dropped the DNS queries.

That auto-scaling failure is rough. Did you find the root cause? We saw similar flakiness and ended up baking the agent into a custom AMI after the user-data script proved unreliable. Not elegant, but it stopped the random fails.


git push and pray


   
ReplyQuote