Environment: Railway (multi-region services)

Runtime: Bun

Database: MongoDB Atlas (multi-region)

Project ID: fed40ff9-1ec4-4036-a208-6055169b0a8d

Service ID: cf0a99b7-d7d0-4908-8afd-bd0838d79431

• Requests frequently hang on Railway until client timeouts (no quick 5xx response) → we’re losing dozens of sales per outage.
• RAM usage grows steadily over time (appears to worsen the longer the instance runs) only on Railway; we don’t observe this elsewhere.
• Deploys are slow and often stall, and attempting to switch to “Metal” never finishes (deploy never completes).

This is affecting production at scale and is revenue-impacting.

1) Requests hang until timeout (most urgent)

• Periodically (several times per day), requests to our Bun service remain “loading” until the client reaches its timeout threshold.

• We see this on normal HTTP endpoints and long-lived SSE endpoints.

• When this occurs, there is no immediate 5xx from the platform—just a stall/hang.

2) Exaggerated RAM growth over time (Railway only)

• Memory consumption monotonically increases the longer the service is up on Railway.

• If we redeploy/restart, RAM drops and then begins rising again.

• We do not see the same growing pattern running the exact same code outside Railway (local/other infra).

• No obvious GC/heap snapshots pointing to a clear leak on our side;

3) Deploys slow and frequently fail to progress

• A significant number of deploys take very long to move from build → start → healthy.

• We’ve observed deploys that never reach healthy, requiring manual intervention/cancel/retry.

4) Unable to move to “Metal” (deploy never completes)

• We attempted to upgrade/migrate this service to Railway Metal, but the deploy never completes (appears to be stuck indefinitely).

• We need help to unblock/force this migration or diagnose why it can’t complete.