GitHub
04/21/2026, 8:36 PM>= when doing an incremental sync with a bookmark. Its for good reason but if someone knows they arent at risk of missing records using only > then we should allow it.
I couldnt find where its documented but I know its to avoid ties that might have been missed on the last sync. For example if the bookmark is a date, and an incremental sync happens early on in the day, then more records arrive within that day, using > would skip those that arrive later in the day and the incremental sync wouldnt have worked as expected. So >= uses "at least once" semantics and requires deduping on the target end.
The use case is for reverse ETL where the target cant handle dupes. Specifically Squared's target-apprise implementation posting to singer-ecosystem-activity in slack. There will always be < 5 records so ties are almost impossible and I think I'd honestly prefer the edge case to be not sending a record than sending it twice. In this case I'd like the option to override the default >= filter to use >.
Another case is if someone uses an auto incrementing PK integer in a postgres database then theres no reason to be inclusive (although it has pretty minimal harm also), the user should have the preference if they know what theyre doing.
I suggest we add something like incremental_filter_inclusive or inclusive_filter (names are tough 😄 ) that defaults to the status quo but allows us to override.
Am I misunderstanding anything?
meltano/sdkGitHub
04/21/2026, 8:36 PM