Disk snapshots on FDB

Hello folks,

I am trying to understand how disk snapshots work (technical details).

When the snapshot command is issued and it runs the whitelisted binary, I get that file-system snapshot / copy is outside of the FDB logic and done by underlying filesystem / tool. However, what happens to transactions that are in progress. How does the restored instance know to discard uncommitted data. I don’t see any read-version numbers passed to the snapshot binary. Senthil’s talk did mention that version-number is passed around, but can I assume this is internally handled by the sequence he is describing?

In addition, is there a requirement that the snapshot binary finish within the typical 5-second txn time limit? Is the DB locked or write-transactions blocked during this process? His talk also started a workload in parallel, but he didn’t mention if the write transactions waited/hung during the snapshot process.

I appreciate any pointers.