a month ago
We need a documented, reversible maintenance window for a service using HTTP functions and HTTP/1.1 WebSockets. We will separately drain existing connections before deployment. We have read Edge Rules, Edge Networking, and Public Networking Specs & Limits, but need clarification before making any production change.
- A service currently returns edgeConfig.enabled=true and edgeRules=null. What supported API operation and exact input restore this identical state after temporarily setting a ruleset? Does updateServiceEdgeRules accept edgeRules:null, or is another reset/delete operation required?
- Are HTTP/1.1 WebSocket Upgrade requests evaluated by Edge Rules before forwarding? Which ingress paths bypass these rules (for example private networking or TCP Proxy), and how can we authoritatively inventory or disable those paths during maintenance?
- Once fully propagated, does a service/environment ruleset remain continuously effective across deployments, traffic switches, and deploymentRedeploy(usePreviousImageTag:true)? How can propagation completion be established before switching traffic?
- Do platform deployment health checks traverse the public edge rules? Can an exact /health allow rule followed by a catchall 403 block preserve health checks while denying all new business requests?
If Edge Rules cannot provide these guarantees, please identify a supported service-level maintenance/admission mechanism that remains effective throughout deployment and rollback, together with its reversible API procedure.
Please provide authoritative documentation or API schema references. No account identifiers, project IDs, domains, customer data, credentials, or logs are included in this inquiry.
1 Reply
Status changed to Solved medim • about 1 month ago
Status changed to Open medim • about 1 month ago
a month ago
Hey!
1- Looking at the GraphQL schema (https://railway.com/graphiql), updateServiceEdgeRules takes UpdateServiceEdgeRulesInput { serviceId: String!, environmentId: String!, edgeRules: JSON }.
The mutation's description says it's a full section replace where null clears the rules, so to reset you have to send "edgeRules": null:
mutation updateServiceEdgeRules($input: UpdateServiceEdgeRulesInput!) {
updateServiceEdgeRules(input: $input) { id enabled edgeRules }
}
It returns the config so you can confirm you're back to null.
2- Edge rules match incoming HTTP requests and handle them at Railways edge before they reach the service, and they apply to every domain attached to the service, and on the Specs & Limits page says the edge supports WS over HTTP/1.1, WS upgrade is just an HTTP request, but you should test it just to be sure. TCP Proxy and private networking aren't HTTP at the edge, so I'd expect rules to skip them.
3- Edge Rules docs says "Rules are configured for a service in the selected environment and apply to every domain attached to that service.", that's service-level config, and the updateServiceEdgeRules input has no deployment ID so I'd expect it to persist across deploys. About propagation, the same page says "Saved changes propagate to Railway's edge locations within a few seconds.". To check, add a block rule with a unique response body and use globalping.io to probe it from multiple locations.
4- Healthchecks docs says that Railway polls your healthcheck path until it gets a 200, and only then activates the new deployment. It sends the host healthcheck.railway.app, and it only runs at deploy time, not continuously. Your allow block condition looks valid but I have yet to test it. Block only takes 4xx, so use 403 or 429.
If someone from the community can test any of this on a service, that would confirm parts I could only get from the docs