a month ago
I have a crashed Railway service with an attached persistent volume that is READY. I need a read-only export before any restart or redeployment because the volume contains append-only timestamped research journals and checkpoints. On Railway CLI 5.59.0, railway volume files download fails with a timeout during SFTP initialization even though the volume remains attached. Dashboard shows No Backups, and Hobby plan does not allow manual backups. Is there a supported way to repair the volume file-transfer channel or obtain a crash-consistent, read-only export/snapshot without restarting the service, restoring/replacing the volume, or attaching another writer? Please advise a non-mutating procedure. I do not authorize service or volume changes.
7 Replies
a month ago
The railway volume files commands reach a volume through its attached service, and they need that service to have an active, running deployment. A volume showing READY doesn't mean file access is available. While the service is crashed, the file transfer can't start, and nothing can be repaired on the volume side to change that.
We don't have a read-only export or snapshot that bypasses this. You can't download a volume directly, attach it read-only somewhere else, or pull it as an archive, and Hobby workspaces can't take manual backups. Without a change to the service, there's no supported way to get the files out.
For reference, a Railway volume backup is a point-in-time snapshot of the whole volume taken at the storage level. That makes it crash-consistent, not application-consistent.
Status changed to Awaiting User Response Railway • about 1 month ago
Railway
The `railway volume files` commands reach a volume through its attached service, and they need that service to have an active, running deployment. A volume showing READY doesn't mean file access is available. While the service is crashed, the file transfer can't start, and nothing can be repaired on the volume side to change that. We don't have a read-only export or snapshot that bypasses this. You can't download a volume directly, attach it read-only somewhere else, or pull it as an archive, and Hobby workspaces can't take manual backups. Without a change to the service, there's no supported way to get the files out. For reference, a Railway volume backup is a point-in-time snapshot of the whole volume taken at the storage level. That makes it crash-consistent, not application-consistent.
a month ago
Thank you for the explanation. I still need assistance because the volume contains irreplaceable append-only research evidence. Is there any way for a Railway engineer to help preserve or export the existing volume without restarting the crashed service? If no such option exists, can you confirm whether upgrading to Pro would allow me to create a storage-level snapshot of the existing volume while its service is crashed, before attempting recovery? I want to avoid any action that could alter the existing files. Please advise the safest supported procedure.
Status changed to Awaiting Railway Response Railway • about 1 month ago
a month ago
We don't offer a staff-assisted export, and we can't preserve or copy the volume for you. The only way to keep a copy without the service running is a volume backup.
On Pro, you can take a manual backup of a READY volume even while its service is crashed. A backup is a storage-level snapshot. Creating one doesn't restart or redeploy the service, and it doesn't write to the existing volume. Keep in mind that a backup is not a downloadable file. It lives with the volume and can only be restored into the same project and environment. Manual backups are also limited to 50% of the volume's total size. See Backups for details.
If you restore a backup later, it creates a new volume and stages a swap for you to review. The original volume is kept, unmounted from the service, and nothing changes until you deploy that staged change.
Status changed to Awaiting User Response Railway • about 1 month ago
Railway
We don't offer a staff-assisted export, and we can't preserve or copy the volume for you. The only way to keep a copy without the service running is a volume backup. On Pro, you can take a manual backup of a READY volume even while its service is crashed. A backup is a storage-level snapshot. Creating one doesn't restart or redeploy the service, and it doesn't write to the existing volume. Keep in mind that a backup is not a downloadable file. It lives with the volume and can only be restored into the same project and environment. Manual backups are also limited to 50% of the volume's total size. See [Backups](https://docs.railway.com/volumes/backups) for details. If you restore a backup later, it creates a new volume and stages a swap for you to review. The original volume is kept, unmounted from the service, and nothing changes until you deploy that staged change.
a month ago
Follow-up after upgrading:
I upgraded the workspace to Railway Pro, and a manual storage-level backup completed successfully while the original service remained crashed.
However, Railway CLI 5.68.0 still fails during volume file access with:
Failed to initialize SFTP session — Timeout
The original service remains crashed, and the volume remains READY. No restart, redeployment, restoration, or volume replacement has occurred.
Could a Railway engineer repair the SFTP access path, provide a read-only export of the original volume or the completed backup, or recommend a documented way to inspect the backup without changing the original deployment?
I need to verify the stored files before attempting any recovery action, so I do not want to restart or redeploy the service or restore/swap the volume merely to obtain access.
Status changed to Awaiting Railway Response Railway • about 1 month ago
a month ago
Volume file access through the CLI only works when the volume's attached service has an active deployment, so while the service is crashed the SFTP session cannot start, and staff can't repair that path or export the volume or backup for you.
A backup can't be downloaded or browsed directly either. The only way to see its contents is a restore, which creates a new volume and stages a swap onto the original service, and deploying that swap redeploys the service. If the swap is discarded, the restored copy stays unattached, and the CLI file commands can't reach an unattached volume. That means we don't have a supported way to inspect the backup's files while the original service stays crashed and unchanged.
Your completed backup is a storage-level snapshot of the volume as it was when it ran, and it stays with the volume in the same project and environment.
Status changed to Awaiting User Response Railway • about 1 month ago
a month ago
We understand that the backup cannot be browsed or exported directly and that the normal restore process stages a volume swap. Could you confirm whether there is a supported way to restore our completed backup into an isolated inspection service while keeping the original crashed service and its original volume completely unchanged? Specifically: (1) Can a restored backup volume be attached to a separate service rather than the original service? (2) Can that separate service run an inert inspection-only command without ever launching the original observer? (3) Would the original volume ID, attachment, and original crashed deployment remain unchanged throughout? (4) If supported, what exact sequence avoids an intermediate deployment of the original observer? We are not authorizing any restoration or deployment at this stage; we only request a documented, non-destructive procedure. If unsupported, please confirm explicitly.
Status changed to Awaiting Railway Response Railway • about 1 month ago
a month ago
No, we don't have a supported procedure for this. Restoring a backup creates a new volume and stages a swap onto the original service. Using that restored copy with a separate inspection service isn't a supported recovery workflow, so we can't give you a sequence for (1), (2) or (4).
On (3): the original volume and the service's current attachment stay as they are unless that staged swap is deployed. If you discard the staged change, the service keeps its original volume and the restored copy stays unattached. The restore itself can't be cancelled once it starts. An unattached restored copy also can't be reached with the CLI file commands.
Your completed backup and the original volume both remain in place as they are now.
Status changed to Awaiting User Response Railway • about 1 month ago
24 days ago
This thread has been marked as solved automatically due to a lack of recent activity. Please re-open this thread or create a new one if you require further assistance. Thank you!
Status changed to Solved Railway • 24 days ago