[click0]

1.1.69 — TODO §7 MED, all six items in one release

Backup / replicate chain

crate backup and crate replicate shared the backup-* snapshot family and each --auto-incremental took the newest local one, so a file backup between two replications became the base of a zfs send -i the remote could not apply ("incremental source does not match"), and a replicate snapshot could become the base of a file backup no .zstream holds.

• Replicate snapshots are now repl-<utc-time>; with --auto-incremental the base is the newest snapshot present on both sides (zfs list -d 1 locally and over the same ssh hop, ZfsDatasetPure::pickLatestCommonSuffix); nothing in common → full stream with a warning.
• Base lookup is -d 1 and pinned to <dataset>@<prefix> (it was -r with an unpinned regex: a child dataset's snapshot could be chosen).
• A failed replicate destroys its snapshot (as 1.1.48 did for backup); template warm and the run-time base_container / cow / --warm-base clones destroy their snapshot when zfs clone fails.

backup-prune and chained incrementals

Only a kept FULL counted as a base, so with --auto-incremental (F ← inc1 ← inc2) inc2 was always an "orphan" and --delete-orphans deleted it. Now an incremental is kept when its base is kept, full or incremental (oldest-first transitive walk).

crate update --pkg-only

crate create deleted pkg(8), emptied /var/db/pkg and trimmed /usr/sbin/pkg, so the 0.8.41 upgrade path ended in ENOENT on every crate that installs packages. New pkg/keep-pkg (default true) keeps pkg, its database and the stub; false gives the old smaller image. crate update checks the jail first and names what is missing and why.

crate restart of a --warm-base run

The spec registry gained <name> warm-base=<dataset> entries (written by crate run --warm-base … --name); restart replays --warm-base/--name. SpecRegistry::lookup returns "" for them, so crated's control-socket start of a warm-base container stays unsupported (noted in TODO).

crate clean vs a run in progress

New lib/dir_lock: crate run flock(2)s its jail directory (close-on-exec, held through teardown) and crate create its chroot-create-* directory; clean skips a held directory ("in use … — skipped") and now also sweeps dead chroot-create-* directories.

crate top

The DISK column was RCTL writebps, a rate, shown as usage. Now WRITE/s rendered as 12.5 MB/s; CPU% from RCTL pcpu when present, the cputime fallback holds its baseline for 5 s instead of quantising to 0/100/200 %.

Docs / tests

crate.5 (clean, restart, top, backup, backup-prune, replicate, update, pkg/keep-pkg), README en/uk, runbook §4l, CHANGELOG, TODO, VERSION / PORTVERSION 1.1.69.
make test-unit on Linux: 1590/1590 passed (+7 cases: prune chain, pinned/common suffixes, repl prefix, warm-base registry entries, top pcpu/rates). make check-version, make check-kyuafile, mandoc lint (pre-existing style warnings only). zfs(8), flock on jail directories, pkg inside a jail and the ssh remote are runbook §4l.

---

Generated by Claude Code