Hi guys,
My final goal is to create an extension that does direct data-file to data-file transfer (no intermediate tables, no triggers) with no blocking at all in order to simulate a zero-downtime FULL VACUUM and work on continuously maintained clustering (which I'm a big fan of). This of course should involve fixing inexes, etc... There are multiple steps to have a final product but the first thing I would do is have two pointers: one iterates from the beginning till it finds enough space, and the other iterates from the end and adds this row to the space pointed by the first pointer.
I would like to first know whether this is useful (in my previous companies this was a complete game changer), and whether there are any alternative algorithms that you would suggest.
Of course, there are many essential features that would follow later like running automatically when system is under light load, adding a threshold for when to do this "online FULL-VACUUM", and utilizing index CLUSTERing and/or a set of predefined columns.
Any and all comments are welcome,
Ahmed