Microsandbox /.msb Runtime Disk Quota
When a sandbox runtime path such as /.msb is host-backed and unbounded, guest writes can consume host disk even when other guest paths have quotas. The fix needs evidence from both sides: guest du/df, host bytes, deleted-open files, and a clear do-not-delete boundary for runtime state.
Cleanup request
Get the exact cleanup step.
Leave your email now. The scan summary can follow after the first reply; we send the $29 Deep Cleanup step only if review-first storage remains.
Runbook: Prove The Host-Backed Bytes
- Identify the
/.msbmount type and backing path before deleting anything. A runtime directory can contain overlays, rootfs state, sockets, metadata, or bind-mounted host storage. - Measure guest bytes with both
duanddf. A large gap usually means filesystem metadata, hidden mount contents, deleted-open files, reserved blocks, overlay lower/upper layers, or host-side sparse allocation. - Measure the host path at the same time. The important question is whether a guest write increases host disk usage one-for-one or through sparse/metadata amplification.
- Check deleted-open files from inside the guest if tools are available. If a process still holds an unlinked file,
ducan drop whiledfstays high. - Add an explicit quota boundary for
/.msb, not only/workspace,/tmp, or the OCI overlay upperdir. - Keep cleanup plan separate from quota policy. Quota prevents future host exhaustion; cleanup decides which stale runtime artifacts can be removed safely.
Copy-ready issue reply
Use this for /.msb quota and du/df mismatch reports.
This keeps the discussion focused on read-only evidence, host pressure, and runtime-safe cleanup boundaries.
I would treat `/.msb` as a separate host-backed runtime filesystem, not as part of the normal guest quota surface.
Before adding cleanup, I would capture one read-only evidence block:
mount | grep ' /.msb '
findmnt -T /.msb -o TARGET,SOURCE,FSTYPE,OPTIONS
df -h /.msb
df -i /.msb
du -xhd1 /.msb 2>/dev/null | sort -h
lsof +L1 2>/dev/null | grep '/.msb' || true
Then compare that with the host-side backing directory at the same timestamps. If host usage increases with guest writes, `/.msb` needs its own quota/admission setting, for example `msb_size_mib`, because `/workspace`, `/tmp`, and overlay upper limits do not protect the host from this path.
For cleanup safety, I would not delete `/.msb` broadly. First separate runtime metadata, rootfs/overlay layers, sockets, active sandbox state, and stale scratch files. The acceptance test should prove:
- writes beyond the configured `/.msb` quota fail with ENOSPC
- `du`/`df` deltas are explained or surfaced in diagnostics
- deleted-open files are reported separately
- cleanup removes only stale owned artifacts, never active rootfs/runtime state
Do Not Delete First
- The whole
/.msbtree, rootfs layers, active overlay upperdirs, sockets, or runtime metadata. - Host backing directories until sandbox processes are stopped and open file handles have been checked.
- Any file used as proof of the
du/dfdiscrepancy before the bug has a regression test. - Workspace data or guest state that belongs to an active sandbox session.
Deep Cleanup
Need a safe cleanup order?
Send the issue link, log excerpt, or storage summary first. We reply with the next safe move and offer the $29 Deep Cleanup only if the incident still needs review.