I wonder, is there further work on the lazy GC file walker planned? I see that right now code does collect pieces to be moved to thrash as part of the process with lowered I/O priority, but the part that actually moves files to trash and removes files is not—it’s still part of the main process. Though that would require writing to the piece expiration database from the low-I/O priority thread.
I imagine (though this is just a hypothesis, not something I measured) that in low memory nodes moving files to trash immediatelly might be using less I/O (and hence being faster) than first collecting, only then batch-moving might be faster, as the files’ metadata would not have to be re-read for the move.