Hello Railway Support,

We are experiencing intermittent PostgreSQL latency in production.

Project: Bober Travel

Environment: production

Region: europe-west4-drams3a

Services:

• Backend: 790333bb-3999-42bc-8ec7-628b6c537ee6
• PostgreSQL: c0c3ced9-ac84-4f07-9739-6aef32e13816

Affected request:

• Time window: 2026-06-19 12:07:35–12:08:34 UTC
• Method/path: POST /route
• Status: 201
• Railway request ID: vEq6_lXBTZKmYccv63j4Ww
• Backend deployment: 18b78625-e39d-46a5-8518-5d80ed7c7f6b
• Deployment instance: 11b3b600-bf9b-4617-a08b-9e87e6430a61
• Sentry trace ID: 4c0c27d79d7b49d5a5a30429535c3d05

Timing:

• Railway totalDuration: 59,152 ms
• Railway upstreamRqDuration: 59,148 ms
• Application duration: 59,122 ms
• Prisma/DB duration: 59,103 ms across 6 sequential queries

The same endpoint normally completes in 38–189 ms.

The services were not under meaningful CPU or memory load. PostgreSQL had 4 connections out of max_connections=100 when inspected, no recorded

deadlocks, and the volume was approximately 320 MB / 500 MB.

PostgreSQL logs also show unusually long checkpoint/fsync operations. For example:

2026-06-19 10:51:30.008 UTC

checkpoint complete: wrote 6 buffers; write=0.536 s, sync=110.643 s, total=114.366 s

Other checkpoints today took approximately 30–105 seconds despite writing only 5–22 buffers.

This appears consistent with intermittent volume/WAL fsync latency rather than application CPU load or connection-pool exhaustion. No checkpoint

was logged during the exact affected request, so we cannot prove that its specific wait was an fsync; however, 59.103 seconds were spent inside

the database client and the storage stalls are independently visible in PostgreSQL logs.

Could you please:

1. Inspect storage/volume latency for the PostgreSQL service during 2026-06-19 12:07:35–12:08:34 UTC.
2. Check whether the underlying host or volume experienced an I/O incident.
3. Advise whether the database should be migrated/reprovisioned onto a new volume or host.
4. Confirm whether similar latency is visible in Railway’s internal telemetry.

Thank you.