@dsm This ones already asking for 40 sum' gigs. Would really like to repurpose scrap from the recycler so I can actually host this for a couple of years at my own place.
@dsm Pruning activities that serve no purpose (e.g. likes that are getting recorded in object in addition to being sotred separately) does help, though.
@mint vps db space is wildly limited for cheap single user builds, like 20gb after os. not that it should be default, but ot is at least a solution for those instances specifically.
@dsm Ah, right. @iamtakingiteasy was playing around with splitting DB into separate chanks that could be stored on different storage mediums, maybe one could try storing old posts on NFS or something, add that to a list of bad ideas.
The use of vpn backhaul, using the vps for host only, was a good (perhaps better) solution. I think that's what @p is doing here now. Suddenly a few tb is affordable.
Since it needs to migrate object ids to time-ordered snowflakes and rearrange some of the denormalized fields to properly partition the tables anyway, it also changes a bunch of queries to take advantage of new layout, as many joins became easier. Migration includes both up and down, so it is possible to rollback, but migration back from partitioned layout wasn't very well tested, especially after new objects were inserted. Eientei runs on this for a couple of months now, so it is more or less reasonable in terms of completeness. But to be something more than just 15% of overhead, it requires understanding of what you're doing. Then sure, moving all older rows to a partition on cheaper or even a networked storage is possible.
@dsm@mint@iamtakingiteasy Yeah; it was initially my "what if the datacenter *does* kick us out" plan, but modified from "I guess ssh tunnel or reverse-proxy to a Tor address and run the shit out of my house" and luckily wireguard provides something a little more sane, although some of the outbound traffic still does go through Tor.
@mint@dsm have ranted and raved about this, crazy that @iamtakingiteasy went through with an implementation of it, going by month or even week would also help with the constant churn and vacoom that currently forces a repack for bloated instances