Indeed it does not preserve the state of traversin directories across node restarts.
That’s right.
Indeed. IIRC, before the implementation of the lazy file walker, files were put into GC immediatelly during the scan, as opposed to first performing the scan, then removing files, so the impact was much smaller. I’ve asked about this in this thread.
Largest single nodes are probably around 20-25 TB. A good setup can scan 1 TB of files in around 10 minutes. Some more complex setups with SSD cache can do even less. This means few hours, but not days.
Unfortunately most setups discussed in this forum aren’t as good, and might may this days indeed.