a month ago
Hello Railway Support,
We are preparing a maintenance window for a production application using Railway-managed PostgreSQL and a Railway Storage Bucket backed by Tigris.
We need to establish a consistent PostgreSQL + S3 recovery checkpoint before applying a non-backward-compatible database migration.
During testing with a separate disposable Railway/Tigris bucket, we confirmed that rotating bucket credentials rejects subsequent requests using old credentials (403), but PutObject and UploadPart requests initiated before rotation can still complete successfully afterward (200).
Could you clarify the following?
-
Is there a supported method to temporarily block all writes to a Railway-managed bucket, including writes from existing credentials and independently authorized clients?
-
Is there a provider-supported way to confirm that all in-flight PutObject, UploadPart and CompleteMultipartUpload requests have completed or been cancelled?
-
Can Railway provide a complete inventory of active bucket credentials and principals capable of issuing new credentials?
-
Is there a supported way to obtain a consistent, immutable snapshot or object inventory of a Railway-managed bucket, including protection against late commits from pre-existing requests?
-
If these capabilities are unavailable, what maintenance and backup procedure does Railway recommend for applications that require a consistent PostgreSQL + S3 recovery checkpoint?
We can provide sanitized reproduction details if needed.
Thank you.
1 Reply
a month ago
We don't have a supported way to put a bucket into a read-only or write-blocked state. Resetting credentials, from the bucket's Credentials tab or with railway bucket credentials --reset, stops new requests from authenticating. As your test showed, it doesn't cancel requests that were already accepted.
There's also no Railway-side signal that confirms in-flight PutObject, UploadPart, or CompleteMultipartUpload requests have finished or been abandoned. How those requests complete is handled by the S3 endpoint, which is described in the Tigris documentation.
Each bucket instance has a single credential set, separate in each environment. That set is the one shown in its Credentials tab, and resetting it replaces it. Any project member with access to the bucket can view or reset those credentials. We don't provide a separate inventory of keys or principals beyond that.
Buckets don't currently offer snapshots or backups, and object versioning and object locks aren't supported yet (Storage Buckets). That means there's no immutable point-in-time copy that would exclude late commits.
For a consistent PostgreSQL and bucket checkpoint, the write controls you have are in your own services. Stop or scale down every service that writes to the bucket, and reset the credentials so nothing else can authenticate. Then take your database backup, and copy or list the bucket's objects with your own S3 client into storage you control.
Status changed to Awaiting User Response Railway • 26 days ago
19 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 • 19 days ago