Donate to CanuteTheGreat/horcrux

• github.com/sponsors/CanuteTheGreat

---

See the documentation for more details on donation links.

horcrux-be: /boot snapshot copies content-addressed, retention tied to ZFS snapshots and free space; never fails on a full /boot #19

Merged

CanuteTheGreat merged 3 commits from fix/be-boot-space into main 2026-11-05 18:55:53 -08:00

Conversation 0 Commits 3 Files changed 8 +667 -27

CanuteTheGreat commented 2026-11-05 16:11:56 -08:00

Owner

• Copy link
• Copy Markdown

Incident 2026-11-05: hogshead /boot 97% (8 byte-identical 75 MB kernel copies in /boot/be/.snap, create-* never pruned, no space check).

• hard-linked content store /boot/be/.objects (identical kernels free; atomic replace)
• destroy-snapshot/prune remove the /boot copy with the ZFS snapshot; orphans cleaned; KEEP covers pre-upgrade-* and create-*
• /boot kept <= BOOT_MAX_USED_PCT (80) by destroying the OLDEST automatic snapshots first (never user-named, clone origins, BEs)
• a full /boot never fails a snapshot/upgrade: copy skipped with a message
• tests: 12 unit tests (fake zfs) + real 512 MB ext4 loop test (30 snapshots, worst 61.6%) wired into ci-run.sh
• test-machine allowlist: net-misc/ctg-ssh-ca (-r3)

Incident 2026-11-05: hogshead /boot 97% (8 byte-identical 75 MB kernel copies in /boot/be/.snap, create-* never pruned, no space check). - hard-linked content store /boot/be/.objects (identical kernels free; atomic replace) - destroy-snapshot/prune remove the /boot copy with the ZFS snapshot; orphans cleaned; KEEP covers pre-upgrade-* and create-* - /boot kept <= BOOT_MAX_USED_PCT (80) by destroying the OLDEST automatic snapshots first (never user-named, clone origins, BEs) - a full /boot never fails a snapshot/upgrade: copy skipped with a message - tests: 12 unit tests (fake zfs) + real 512 MB ext4 loop test (30 snapshots, worst 61.6%) wired into ci-run.sh - test-machine allowlist: net-misc/ctg-ssh-ca (-r3)

CanuteTheGreat added 2 commits 2026-11-05 16:11:57 -08:00

horcrux-be: content-addressed /boot snapshot copies, snapshot+copy retention, never fill or fail on a full /boot ... b0320d5278

Every BE snapshot copied the whole kernel set (~75 MB) into /boot/be/.snap/<be>@<snap>, identical kernels included, with no
space check; create-* snapshots were never pruned. A lab laptop's 2 GB /boot hit 97% (Rustinion critical).

- /boot/be/.objects/<sha256> content store; every kernel/initramfs under be/ is a hard link to it (copy when links are
  unavailable): identical kernels cost nothing, an object nobody links to is deleted (st_nlink==1), replacement is atomic
  so a new BE kernel never writes through to a snapshot's link
- destroy-snapshot / prune remove the /boot copy with the ZFS snapshot; copies of snapshots destroyed elsewhere are removed
  (unless ZFS cannot be queried); KEEP now covers pre-upgrade-* and create-* (never user-named, never a clone's origin)
- before saving a kernel, /boot is kept under BOOT_MAX_USED_PCT (80) / BOOT_MIN_FREE_MB (32): oldest automatic snapshots are
  destroyed first; if even that cannot make room the copy is skipped with a message
- a full /boot never fails a snapshot, an auto-snapshot or a kernel sync: the ZFS snapshot stands
- horcrux-be dedupe converts old plain copies; tests: unit (fake zfs) + real 512 MB ext4 loop test, 30 snapshots <= 80%

test machines: allow net-misc/ctg-ssh-ca (host SSH CA trust, same family as -fleet) in world.allowlist (-r3); hx-invariants kept failing on it 5709f4f9e0

CanuteTheGreat added 1 commit 2026-11-05 16:12:50 -08:00

horcrux-be: a repeated sync/snapshot of an unchanged kernel left *.new temp links (rename of two links to one inode is a no-op); test 063c104d7f

CanuteTheGreat merged commit d1b1613f28 into main 2026-11-05 18:55:53 -08:00

CanuteTheGreat referenced this pull request from a commit 2026-11-05 18:55:54 -08:00

Merge pull request 'horcrux-be: /boot snapshot copies content-addressed, retention tied to ZFS snapshots and free space; never fails on a full /boot' (#19) from fix/be-boot-space into main

CanuteTheGreat referenced this pull request 2026-11-05 19:41:46 -08:00

horcrux-test-machine-r4: move config-protected package.use into place in pkg_postinst #20

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

CanuteTheGreat/horcrux!19

No description provided.