Jason Dusek <jason.dusek@gmail.com> writes:
> The idea would be, to store information about the last XID in
> the last sync and search for XIDs committed since then upon
> reconnecting for sync. Perhaps `txid_current_snapshot()'
> preserves enough information. Is this a plausible technique?
Perfectly plausible, and often done in one guise or another. You can't
expect row XIDs to survive forever --- they'll be replaced by FrozenXID
after awhile to avoid problems due to transaction counter wraparound.
But for delays of a few minutes, in a database with an unremarkable
transaction rate, that's not an issue.
regards, tom lane