there's a timeout on send, which is... weird. writes shouldn't ever be blocking for long.

collection stats

now feat updates and deletes

idk why these are the same commit, except that i'm finally annoyed enough at stubbing the memstore to remove it. it wasn't ever that real since it was just a hacked copy of the fjall stuff.

blehhhhhh

it's done. maybe we won't do it again. that would be nice.

batching doesn't buy anything since we have to commit intermediate batches anyway in case they are too big.

seems like ~same performance, but this is a little less code and doesn't lose progress if interrupted.

i knew i missed this somewhere, idk how it took this long to remember where :s

UFOs: change counting and fix NSID encoding

also actually test that we can prefix against a null-terminated string

this is probably still going to be annoying for building full db-concat'ed keys, but hopefully not too bad

might be possible to go a step further but this is ok for now.

still doesn't show up in the schema output but at least now it's not accidentally required.

get stats for all collections over arbitrary time periods!!!!!!!

still only actually using the single AllTime iter, but seems to work with that.

and cross notes for aarch64 raspi4

musl works -- gnu had some issue with the linker at runtime or something.

will add paging eventually

UFOs: jetstream backfill

UFOs HLL DID-hashing: secret prefix

and use sketches with P=14 (0.8% expected error)

UFOs rank indices

could probably get away with either just ranking records (upper bound for DIDs) or DIDs (lower bound for records) but ordering by the other would probably be messy / approximate. these indices should be pretty cheap to store, so doing it is must a matter of overcoming my laziness

to that effect i'm finally adding a macro for the static_str thing, and made the keys generic-ish.

takes it from ~600ms per trim to ~3ms on my laptop

keeps a list of dirty collections in memory from rollups and trims those

apparently allows browser scripts to set a custom ua

in prod nginx is currently overriding all cors stuff anyway, but