Alright, so I've been running Versa for about six months now, primarily for connecting our dev and staging Kubernetes clusters across different clouds. I was really excited about the "cloud-optimized" routing feature they've been talking about – the promise of smarter, more efficient paths to our SaaS dependencies and between cloud providers.
But I'm hitting something weird, and I'm wondering if it's just me. Every so often – and it feels almost random – our inter-service pings and traceroutes between, say, our GCP cluster and an AWS-based service, will go from a normal ~20ms to over 2000ms. We're talking full seconds of latency. It lasts for 2-5 minutes, then snaps back like nothing happened.
My first thought was our own code or the cloud providers, but after a lot of digging:
* The spikes correlate exactly with the path showing as routed through a Versa "cloud-optimized" node in the analytics.
* When it falls back to the standard path (shows as a different PoP in the portal), latency immediately normalizes.
* This happens across different source and destination regions, so it's not isolated to one location.
I'm not seeing packet loss during these events, just insane latency. It's throwing off our automated canary deployments and causing false positives in our monitoring alerts.
Has anyone else deployed this feature and seen similar behavior? I'm trying to figure out if:
* This is a known issue with a workaround.
* My configuration is somehow forcing suboptimal "optimized" paths.
* It's just a scaling/growing pain in their cloud network.
I love the idea, but these intermittent spikes are a killer for anything performance-sensitive. Would appreciate any data points or config tips you all might have.
Ship fast, measure faster.
We saw something similar last quarter. The "cloud-optimized" path kept trying to route our East-to-West traffic through Singapore, of all places.
Is your latency spike consistent for all destinations during an event, or is it specific to certain IP ranges? Ours seemed tied to how their route selection handled particular AWS prefixes.
null