Backups start failing after a week or so

I think the issue was with some weird permissions on some of the files, though doesn’t explain how it would sometimes work.

You’re right that doesn’t explain it. It wouldn’t do that so its not it. Duplicati has some code issues at the very least that can work for some time then fail. Other computers won’t have a problem. Some will.

In programming, what happens can be explained as a possible race condition or possible errors that things may throw and at code where Duplicati doesn’t deal with those errors so things can happen or not happen.

What happened is likely a Duplicati problem that you just happened to temporarily or hopefully better workaround.

I did extensive debugging of an issue with Duplicati and it has some problems. I’d expect this possibility for years depending on if and when and if they can even find all the issues here. The Duplicati code is really complex so its not going to be easy for them.

Can you describe the permission situation some? I doubt Duplicati goes out of its way to change these, but because we’re rather low on evidence, might as well look at all that’s known before rejecting it as irrelevant.

I guess I might as well point, although what I really want is for anybody who thinks they might be seeing this somewhat often to maintain a log-file=<path> at log-file-log-level=Information or maybe slightly larger Retry.

Error during compact forgot a dindex file deletion, getting Missing file error next run. #4129 . Alternatively but not as good, use About → Show log → Live → Warning after failure to at least see what files are “missing”.

would also benefit from knowing the type of files, e.g. dblock or dindex, and where they were in the uploads. After the backup itself, version deletion can cause compact to run. This does downloads and uploads while working. It can at least lose track of deletions, per the issue above. Perhaps it can lose track of uploads too.

Keeping a separate log file not only shows whether compact ran, but shows odd endings and all file names.

Posting a link to a database bug report of the database with the problem can possibly provide evidence too, but it’s subject to the database forgetting some things on its way out (transaction rollback) so a log is better.

Finding reported issues is a lot easier when details are available, thus I ask to see if this is the known issue. Permission problem (whether or not it’s a cause) will definitely will be hard to find without better description.