possum-db 0.3.1

concurrent disk-backed cache supporting efficient direct file I/O, transactions, and snapshots using file cloning and sparse files
Documentation 
1
2
3
4
5
6
7
8
9
10
11

On my M1 Macbook, on APFS:

 * Reads/views are 10-12µs when clone files can be reused.
 * List operations: 22µs.
 * File cloning is 60µs, tending toward 75µs for larger files.
 * Empty reads (and the sqlite read transaction): 360ns.
 * Empty writes (but a forced immediate write transaction): 3µs.

 Need to offload hole punching. The fcntl and opening files takes a good chunk of time. Perhaps values can be skipped if previous punches include them. Values can also be streamed over a channel and batched up.

 Batch writer commits are also slow.