Plakar Cache Fills Disk During Backup

When a Plakar backup of a large tree fills ~/.cache/plakar or a dedicated cache partition, the next run can fail before the old cache is cleared. The safe fix is to classify cache, repository metadata, and active backup state before deleting anything.

Free scan first

Submit the cache symptom first.

Leave an email after the browser scan step and we reply with the next safe cleanup move before suggesting the $29 Deep Cleanup.

Use the runbook

What To Measure First

Confirm which filesystem is full: root, the Plakar cache mount, temp, or the destination repository.
Measure $HOME/.cache/plakar by child directory: scan, vfs, repository, plugins, and any versioned subfolder.
Record whether the failed run was interrupted mid-backup. A failed scan cache is usually safer than repository metadata, but do not assume.
If the service runs under systemd, verify the effective home and environment with systemctl show and the unit's User=, not only the shell you tested manually.
For a large Docker tree, separate Docker's own storage growth from Plakar's cache amplification. They can fill different filesystems at the same time.

Copy-ready issue reply

Use this when Plakar cache growth blocks backups.

This reply gives maintainers a reproducible storage contract without asking users to share secrets or backup contents.

I would treat this as a cache-budget and failed-run cleanup boundary, not only a bigger-partition problem.

Checks that would make the behavior safer:
- Print the effective cache path at backup start, including the systemd User/HOME and whether XDG_CACHE_HOME was honored.
- Before scan/vfs cache writes, compare available bytes with an estimated cache budget and fail before the root filesystem reaches 100%.
- On ENOSPC, mark the run failed and list the cache subdirectories created by that run.
- Provide a documented cleanup command for failed-run scan/vfs cache only, while leaving repository metadata and previous restore state untouched.
- Add a regression test where a large source tree fills the cache mount and the next backup does not crash on stale failed-run cache.

Do Not Delete First

Repository metadata under Plakar's cache until you know whether it is only a local acceleration layer.
Any cache directory from an active backup process.
Logs that show the first ENOSPC path, because that tells you whether cache, temp, or repository storage filled first.
Docker volumes from the source tree before proving they are not local databases or stateful services.

Free scan first

Want the next cleanup step?

Submit the cache summary. We tell you what looks safe to clean first, or say no-pay if the issue needs a product fix rather than a cleanup order.