[Question]: Behavior after server rebuild with older database

[ogak-github]

Hi, I have a question regarding PowerSync behavior.

Scenario:

Clients already synced and have newer data locally.

The server was rebuilt and restored using an older database snapshot.

PowerSync instance was recreated.

Some record UUIDs may also be different after the rebuild.

Question:

What happens when the client syncs again?

Will the older server data overwrite the newer local data?

Or will the local client push the newer changes back to the server?

If neither, what's the recommended approach to avoid conflicts or duplicated records (e.g., client reset, version check, conflict resolution)?

Looking for guidance on the expected default behavior and best practices for handling this situation.

Thanks!

[rkistner]

For the most part, we can split uploading client-side changes from downloading server-side state:

1. For uploading changes, the client keeps a local queue of all changes (PUT/PATCH/DELETE operations). While this is not directly affected by the server rebuild, there may be conflicts from attempting to apply the client-side changes on an older server-side state. You are in full control of the upload process - check your logic in your PowerSyncBackendConnector, and make sure it can handle conflicts. The best strategy here depends on your use case - for example you can just discard conflicting changes (which reverts them), or save the conflicting changes to a separate table for manual review. For more details on the options, see our guide here.
2. For downloading state, the client waits until the upload queue is empty, then re-applies the state from the server. Since you're re-creating the instance, it would effectively re-download all data, and replace all the old data on the client.

So in short, uploading changes from the client is the main place where issues could come in. Once the upload queue is processed, the client should match the server-side state exactly after completing a sync.

One thing that stands out to me though:

Some record UUIDs may also be different after the rebuild.

Why would UUIDs change due to a snapshot restore? Changing UUIDs would make it much more likely to run into conflicts/issues when the client uploads changes.

[ogak-github]

My client data originally came from Excel. I converted it to CSV and then imported it into the database, which caused the UUIDs to be regenerated automatically. Since this is just a demo app, it made me wonder whether snapshot restore would fully replace all existing rows in the database or simply append new ones.

But after your explanation, everything makes sense now. It really helped clear things up and now I can plan the right strategy moving forward.

Thanks a lot!