All three of your files seen as missing were actually deleted, according to log comparison below:
The Delete are the end of one compact cycle. The Get start the next one, so is a good time to fail,
except for the forgetting that Delete had happened (forgetting is due to unfinished DB transaction).
Let’s also look at the non-dindex messages to see if there seems to be a similar cleanup problem:
This doesn’t have a match in posted log, but it might be before the posted part. Search log if you wish.
I’ve been looking at the lengthy old issue to see if a good way to fix this is known. Not clear one exists.
Just running Repair pushed up some dindex files, but they were then extra, with no associated dblock.
This bothered DB recreate. I was wondering if going hands-on in DB was needed, and it looks like you
already are in there (in later edits). The basic idea would be to get rid of the Remotevolume table rows thinking they know those dindex files, but first make sure that ID isn’t in IndexBlockLink IndexVolumeID.
Make a copy of the database before trying such experiments. I’ll look at your SQL, but mine’s not great.
You seem to have a similar idea, but details matter, so rather than setting State I call for manual delete.
Having a tested procedure that works on the issue test case would be nice. Maybe yours would be it…
Even better, if you do C# and GitHub, would be to test the proposed patch and submit the pull request.
There is such a shortage of developer volunteers that it might sit, but at least it could be readied to go.
Eventually the original author might show up if nobody else does it first, and would be a good reviewer.