2 hours ago
Hello,
I had a question about specs - https://docs.railway.com/networking/public-networking/specs-and-limits
The rate limits table lists "Maximum Connections: 10,000 concurrent connections", but it doesn't say what the limit applies to. The HTTP Requests/Sec row says "to a given domain", while the connections row has no qualifier.
Is the 10,000 concurrent connections limit per service, per domain, per replica, or per client IP? In other words, if a service has 10 replicas behind one domain, can it hold 10 × 10,000 connections, or just 10,000 in total?
1 Replies
2 hours ago
Our edge counts concurrency per domain, separately on each edge node it passes through. That count is shared across all client IPs and all the replicas behind that domain, so it isn't a per-replica or per-client-IP allowance.
So 10 replicas behind one domain do not get 10 × 10,000. Adding replicas, or upgrading your plan, does not multiply the edge limit on its own. Because each edge node keeps its own count, traffic from users spread across regions reaches the cap later than traffic that all comes through the same nodes.
What's counted is in-flight requests. A long poll holds a slot while it waits for its response. An established WebSocket connection gives its slot back after the upgrade, but a new WebSocket request can still be refused at the start if the cap is already full.
Status changed to Awaiting User Response Railway • about 2 hours ago
Status changed to Solved amitkulkarni23 • about 2 hours ago