Description

I’m using Railway’s managed Qdrant plugin alongside my custom wren-ai-service container. The service startup script waits for Qdrant at its in-cluster hostname (qdrant.railway.internal:6333) but the lookup always fails.

To troubleshoot, I removed every hard-coded QDRANT_* variable, then added only variable references via the Railway picker: QDRANT_HOST→qdrant, QDRANT_PORT→PORT (which is 6333), QDRANT_URL→http://${QDRANT_HOST}:${QDRANT_PORT}. I did not type any .up.railway.app domains anywhere.

I enabled Private Networking on both the Qdrant plugin and wren-ai-service, confirmed they share US East (Virginia) Metal, and added dependsOn so wren-ai waits for qdrant. I also stood up the plugin’s TCP proxy on port 6333, which should register the alias qdrant.railway.internal:6333 in Railway DNS.

Despite all of the above, wren-ai-service logs still show “forward host lookup failed: No address associated with name,” and the service times out waiting for Qdrant. I’m looking for any missing setting—feature flag, DNS propagation issue, firewall rule, or plugin config—that would let in-cluster DNS work as intended.