Prune repeatedly fails with exit code 100 — B2 fossils accumulating

I’m looking for advice on a persistent pruning issue with Duplicacy CLI and Backblaze B2.

Environment

  • Duplicacy CLI 3.3.0 (previously 3.2.3; same issue)
  • Debian Linux, Backblaze B2 storage
  • Snapshot ID: HomeBackup
  • Repository size: ~3.8 TB
  • Backups twice weekly

Problem

My retention policy is:

duplicacy prune -keep 0:210 -keep 30:56

The dry run completes successfully and identifies 49 snapshots for deletion. However, the actual prune repeatedly fails with exit code 100, and storage consumption continues to grow.

One of the final log messages is:

Chunk 0b8cb1be1375c6ab934d96bf321ed57492b0439e4b422f5a262245c1515f855c does not exist in the storage

Interestingly, this same chunk appears as “marked fossil” in 19 prune logs dating back to December 2025. I’m not certain whether it’s causing the failure.

Troubleshooting performed

  1. Upgraded from 3.2.3 to 3.3.0 — no change.
  2. duplicacy check -id HomeBackup -r 1937 — latest snapshot passed; all ~420,000 referenced chunks exist.
  3. duplicacy check -a -fossils — ran for approximately two days before I interrupted it. The captured output showed extensive fossilized chunk references across 39 revisions.
  4. Compared these revisions against the prune dry run. Most are scheduled for deletion, but revisions 1910 and 1915 are retained despite referencing fossils.
  5. Checked those revisions individually:
Revision Fossil references Missing chunks Exit code
1910 20,048 0 0
1915 15,015 0 0

Both checks reported that all referenced chunks exist, although many are fossils.

Questions

  1. What might cause prune to fail repeatedly with exit code 100 while the dry run succeeds?
  2. Should I run check -resurrect on revisions 1910 and 1915 before attempting another prune?
  3. Would testing prune -delete-only and -collect-only help isolate the failure, or is there a better diagnostic approach?
  4. What’s the safest way to recover from these apparently incomplete prune operations without risking retained backups?

I’ve paused pruning for now and can provide the relevant logs.

I’d appreciate any recommendations on how best to proceed. Thanks!

I ran into this thread. Should I follow those suggested actions?

Is there a reason you aborted the check? You should let check -a complete, and not bother with -id -r, as you want to make sure the whole storage isn’t missing chunks otherwise prune will abort and you don’t want that.

Almost certainly the reason the prunes are failing is because a prior run was aborted and the chunks were deleted before the snapshot revision files were deleted, so you may need to manually remove the revisions to which you think would have been pruned.

Personally, I’d run the -fossils -resurrect once and then start figuring out which snapshots are meant to be pruned (-dry-run?), then remove the revisions manually, then run another full check. Also yes, remove cache is a good idea.

Edit: Oh and if check -a errors out at any point, running it again with -persist flag will hopefully get you a full list of revisions which you may need to delete manually.

Thank you for the feedback. I will restart the check -a again and let it run until completion. At the time, 2 days seeemd excessive so I was wondering if there was an issue. There was nothing in the output suggesting that though so it was more impatience.

The full duplicacy check -a -fossils has now completed.

Results:

  • Exit code: 100
  • 861,177 fossil-reference messages
  • Only 4 missing chunks, all referenced by revision 1862
  • All other revisions (1867–1937) report that their referenced chunks exist.

Importantly, one of the four missing chunks is:

0b8cb1be1375c6ab934d96bf321ed57492b0439e4b422f5a262245c1515f855c

This is the same chunk that repeatedly appears in my failed prune logs.

Revision 1862 is from April 7, 2025, and is already scheduled for deletion under my retention policy.

Would the recommended next step be to remove revision 1862’s snapshot metadata so pruning can proceed? If so, what’s the safest supported method?

Also, should I resurrect the fossils referenced by retained snapshots before attempting another prune, or would you recommend a different recovery sequence?

I haven’t made any changes to the repository since completing the check.

Manually delete the revision file from the B2 bucket - i.e. /snapshots/<id>/1862. You can clean up with a prune -exhaustive and then personally I’d run another check -a after just to be sure.

I don’t think resurrecting fossils is necessary in this case.