Backblaze reliability

Hi
Question for the folks that are using backlaze b2 as a backend

How often do you have errors that require you to repair the database?

I have had quite a few and I am wondering if this is something to do with my usage or something with the backend and the way it handles interrupted uploads

Cheers

What errors are you getting? “Unexpected difference in fileset…”? I used to get those errors sometimes, but it was unrelated to the backend. I haven’t seen the issue since using more modern Canary releases. Hopefully the next Beta will be released soon which will incorporate fixes made over the past year.

Yes those one. None of them since I move my data to some other provider but at the same time I followed the releases of the new canaries

Yeah that bug was independent of the back end. I use B2 without issue…

If you’re willing to help debug (or at least say more about your usage), tracking down whatever happened previously would be helpful, unless software is older than 2.0.5.1 which made this issue much more rare.

Interrupted uploads are supposed to get retried 5 times (default) for any destination type. Logs can show.
Interruptions through process kills or restarts of the computer are more likely to leave some mess behind.

I moved everything to wasabi (so I can run more regular data verification process). I haven’t noticed any issue since. Happy to assist with debugging if problems start reoccurring. I can be in touch via forum

If you still have an unused Backblaze account, the free limits might be enough for a small test backup.
Or if you prefer it to not be a test backup, you can pick some valuable data and keep a second copy…
This might not show that problem though. Maybe time and specific files matter. I have no way to know.

What kind of output would you need?

Not clear what’s needed. To see the damage, a database bug report is good. Because file names are sanitized (but might be significant), you might have to help lookup specified things with a DB Browser.

These reports can be awkwardly produced later if need be, provided you kept a copy of bad Database.

Environmental clues might be good (and might be good even now, if you recall) such as PC shutdown, messages coming from Duplicati, etc. Some of that can sort of be reconstructed from Windows data.

When specifically hoping to observe an issue, logs help. Interruptions and retries would be visible with
log-file=<path> at log-file-log-level=retry. Higher than that gets large and might need personal info edits.

Highest possible level is profiling with profile-all-database-queries set, but it’s a huge log, so probably would be reserved for developer use after initial clue. Ideally, a way is found to reproduce this.