vh
5de5583762
- playbooks/ana-ml2-pool-health.yaml: rerunnable elway play (scrub-if-idle, nvme-cli, dangling-image + builder prune; never prune -a) - servers/ana-ml2/README.md: 8-drive PM1725b inventory with SMART counters, nvme7 absence + alerting-gap note - persistent-memory: deferred entry closed, outcome + follow-ups (pool-health alerting, nvme7/slot 0-5 watch, boot import race)
45 lines
1.6 KiB
YAML
45 lines
1.6 KiB
YAML
# ana-ml2 pool-health actions (operator-approved 2026-09-09, "all 3 on clean context").
|
|
#
|
|
# 1. kick a scrub of `tank` (2 CKSUM on nvme7n1 after the 09-05 boot resilver,
|
|
# no scrub since 04-12) — polled separately; `zpool clear` only after a
|
|
# clean pass, by hand, never in this playbook.
|
|
# 2. install nvme-cli so nvme7's SMART counters can be read.
|
|
# 3. reclaim zroot (91%): dangling images + build cache ONLY. Never `prune -a`
|
|
# here — tagged idle images include rollback seats.
|
|
#
|
|
# Rerunnable: a second run shows the scrub step skipped (in progress / done
|
|
# within the hour is not detected — the guard is "not already scrubbing"),
|
|
# nvme-cli ok, prunes ok with 0 B reclaimed.
|
|
|
|
steps:
|
|
- name: Start scrub of tank (skip if one is already running)
|
|
shell: zpool scrub tank
|
|
sudo: true
|
|
when: "! zpool status tank | grep -q 'scrub in progress'"
|
|
|
|
- name: Install nvme-cli
|
|
shell: DEBIAN_FRONTEND=noninteractive apt-get install -y -q nvme-cli
|
|
sudo: true
|
|
creates: /usr/sbin/nvme
|
|
|
|
- name: Prune dangling docker images (NOT -a)
|
|
shell: docker image prune -f | tail -1
|
|
sudo: true
|
|
|
|
- name: Prune docker build cache
|
|
shell: docker builder prune -f | tail -1
|
|
sudo: true
|
|
|
|
verify:
|
|
- name: Scrub is in progress or already completed today
|
|
shell: "zpool status tank | grep -Eq 'scrub in progress|scrub repaired .* on .*$(date +%b\\ *%e)'"
|
|
changed_when: "false"
|
|
|
|
- name: nvme binary present
|
|
shell: test -x /usr/sbin/nvme
|
|
changed_when: "false"
|
|
|
|
- name: zroot usage after prune
|
|
shell: df -h / | tail -1
|
|
changed_when: "false"
|