Using a view + trigger table swap about pgslice HOT 3 CLOSED

JFYI there is a proof of concept here alice-financial@94bf29c.

• Needs more (manual) testing and refactoring.
• Also the view trigger has to make some assumptions about how to handle inserts that may not generalize well to other table structures (auto-incrementing primary key - this could maybe be handled at runtime, but would obviously make things more complex)
• Only tested on pg 11, it looks like pg 10 supports INSTEAD OF triggers on views, so may be fine there as well.

Because of the second two issues mentioned above, I probably will not bother putting together a PR against this repo, but if you are interested in the functionality I'm happy to rework it a bit (or happy to just have someone else do that).

from pgslice.

Hi @lukeasrodgers, thanks for sharing! It's a neat pattern. I don't think it'll save space (since Postgres doesn't reclaim space on deletes), but does have the advantage of working with updates. I don't think I want to support another pattern right now, but may revisit in the future.

from pgslice.

Thanks @ankane - you seem to be right about this not actually saving space. I had erroneously thought that even though postgres wouldn't "return" the disk space back to the OS, it would internally reuse it, but that's not the case. This doesn't even seem to happen if I issue manual VACUUM commands during the fill process.

from pgslice.