Author SHA1 Message Date
DongHyeonka 050a4de235 fix(validate): render the drift check from a clean working-tree copy
In a long-lived checkout the observability renderer finds an ignored
.helm/charts/<chart>-<version> cache left by an earlier render and
refuses ("exact chart cache already exists"), so live-drift-check.sh
fails with exit 2 in /home/donghyeon/workspace/platform, the path the
runbooks use.

When run inside a git working tree, copy the tracked and untracked
(not ignored, not deleted) files to the drift work directory and
render from there. Local edits are still checked; ignored caches are
not carried over. Outside a git tree (the daily timer renders a
git archive of origin/main) nothing changes.

Also accept the one drift this exposed: the kube-prometheus-stack
PrometheusRule still has the two Kubelet certificate alerts that PR #6
disabled in values; it goes away when the observability core is
re-applied in the restart window (until 2026-11-30).

Risk: read-only, as before. The copy is removed with the work
directory.

Verification (desktop, fresh clone of main 8b019d2 with an ignored
services/observability/tempo/.helm/charts/tempo-2.2.3 created):
- old script: "REJECT: exact chart cache already exists", exit 2.
- new script as an uncommitted edit in that clone: SUMMARY drift=1
  accepted=12 (the PrometheusRule above), exit 1.
- new script from a git archive export, as the timer runs it: same
  summary.
2026-11-12 02:00:20 +09:00
DongHyeonka cc5decc538 feat(traefik): re-apply the trust overlay when trust is already live
apply-trust.sh is the observe -> trust transition. When the live
Traefik is already in the trust phase and only another value changed
(PR #3 resources), it stops at "observation phase must not trust
forwarded headers" before applying, so the change had to be applied
with a bare kubectl apply outside the guarded path.

reapply-trust.sh checks that the runtime is in the trust phase with
the reviewed CIDR and that forwarded-header trust works (a request
through Host Nginx is logged with the forwarded client, not the proxy
address), saves the live HelmChartConfig, shows the change with
kubectl diff and does nothing when there is none, asks for
"APPLY <context> RETRUST <cidr>", applies, and repeats the checks plus
the NodePort, Gitea health and site-manifest checks. Any failure after
the apply re-applies the saved HelmChartConfig, so the trusted CIDR is
never dropped.

Risk: the apply path was not run against the cluster (the live values
already match). The restore path reuses the same wait and health
helpers as apply-trust.sh.

Verification (desktop):
- Live (already trust, values match): "Forwarded-header trust in
  effect (ClientHost 127.0.0.1, proxy 10.42.0.1)", then "already
  matches the trust overlay. Nothing to do.", exit 0.
- Copy with memory 64Mi changed to 65Mi, answering "no": the diff
  shows only that line, "ERROR: cancelled", exit 1; live still 64Mi.
- The ClientHost check of apply-trust.sh would have failed here
  (127.0.0.1 is expected in the trust phase), which is why the check
  is inverted.
2026-11-12 02:00:20 +09:00
donghyeon.kang 8b019d26a0 Merge pull request 'docs(runbooks): add restore, rebuild, upgrade and reboot runbooks' (#8) from donghyeon.kang/w6-runbooks into main 2026-11-11 16:48:53 +00:00
donghyeon.kang 28595c47e2 Merge pull request 'feat(keycloak): pin images and attach PriorityClasses for the restart window' (#7) from donghyeon.kang/w6-restart-window-pins into main 2026-11-11 16:48:42 +00:00
donghyeon.kang 00a6680b76 Merge pull request 'feat(backup): back up k3s and Gitea to AIStor with alerts and drills' (#6) from donghyeon.kang/w6-host-backups into main 2026-11-11 16:48:31 +00:00
donghyeon.kang 4ab37ff18c Merge pull request 'feat(host): add the k3s restart bundle and its apply script' (#5) from donghyeon.kang/w6-k3s-restart-bundle into main 2026-11-11 16:48:20 +00:00
DongHyeonka cb78ccdb0b feat(backup): store the Gitea version next to each dump
A Gitea dump must be restored with the Gitea version that wrote it, and
a restore checks the repository count against the dump. Neither was
recorded outside the archive, so the restore runbook had nothing to
compare with.

Upload a small .meta file with each dump: creation time, the
Deployment image, the gitea --version output and the repository count.
docs/runbooks/gitea-restore.md reads it.

Risk: none for the backup itself; the file holds no secret.

Verification: bash -n; kubectl -n gitea exec deploy/gitea -c gitea --
gitea --version prints "gitea version 1.27.0 built with ..." on the
desktop, so the third field is the version.
2026-11-12 01:38:24 +09:00
DongHyeonka de600a0bb1 feat(priority): attach PriorityClasses to Keycloak, Gitea, Loki, Tempo
The PriorityClasses exist in the repository but no pod uses them, so
every platform pod still evicts at priority 0.

Attach platform-critical to Keycloak (CR spec.scheduling) and Gitea
(chart priorityClassName), and platform-low to Loki (singleBinary) and
Tempo. The Keycloak CR also gets spec.image at the running 26.7.0
digest with startOptimized: false: with spec.image the operator
otherwise assumes a pre-built optimized image, and this is the stock
one. Grafana and pgAdmin have no top-level priorityClassName in their
charts and are left for later; platform-postgres is W1's file and
tech-log belongs to its deploy/ directory.

Risk: applying restarts Keycloak, Gitea, Loki and Tempo once, so it
belongs in the k3s restart window, after the PriorityClasses exist (a
pod that names a missing class is rejected). The accepted-drift list
holds these objects until 2026-11-30.

Verification (desktop): render-phase1/2 and render-observability-core
pass with the four priorityClassName values; kubectl diff (server
dry-run) against live shows only these fields.
2026-11-12 01:25:24 +09:00
DongHyeonka 9751db0e97 feat(keycloak): pin the operator source and both images by digest
The Keycloak Operator remote resource is pinned to the tag 26.7.0, and
that tag's content changed once (live build-timestamp 2026-07-09,
later renders 2026-07-23). Both images use the 26.7.0 tag with
imagePullPolicy Always, so every restart depends on quay.io and could
pull different bits.

Pin the remote resource to the tag's commit, the operator image and
RELATED_IMAGE_KEYCLOAK to the digests that were running on 2026-11-12
(pod imageID) and set the operator to IfNotPresent. The env value is
set with a strategic merge patch keyed by name; a JSON patch by list
index hit KUBERNETES_NAMESPACE instead. render-phase2.sh and
apply-keycloak.sh check the commit and the digest.

Risk: applying restarts the Keycloak operator once (restart window).
Keycloak itself keeps running.

Verification (desktop): render-phase2.sh passes; the rendered
Deployment has the pinned image, IfNotPresent and the digest in
RELATED_IMAGE_KEYCLOAK with the other env entries unchanged.
2026-11-12 01:25:24 +09:00
DongHyeonka 9e9ea94461 feat(backup): keep k3s datastore archives outside AIStor on the HDD
AIStor runs inside the k3s cluster that the datastore archive restores.
After losing the whole cluster the AIStor copy cannot be read until k3s
is running again, which is what the archive is for.

Also keep the seven newest archives in
/srv/k3s/aistor/host-backups/k3s-datastore (root 0700) on the same HDD
filesystem (sdb3). The directory is next to AIStor's drive directory
data/, so AIStor does not list it as a bucket, and it is readable by
mounting sdb3 alone. A metric reports how many local copies exist.

Rejected: the NTFS archive partition (sdb2). It has no Unix
permissions and Windows on the same machine can read it; the archive
holds the server token.

Risk: the files live inside the AIStor PersistentVolume root, which is
mounted into the MinIO pod (uid 1000); mode 0700 root keeps the pod
from reading them.

Verification: bash -n; /srv/k3s/aistor contains only data/ (2026-11-12).
2026-11-12 01:14:45 +09:00
DongHyeonka ba3dddc139 feat(validate): add restore drills for the k3s and Gitea backups
A backup nobody has restored is not known to work. The 2026-08 k3s
recovery bundle never got its isolated restore drill.

k3s-datastore-restore-drill.sh downloads the latest archive, checks its
sha256 and integrity, starts the same k3s version in a network-less
throwaway container on the copy, and checks the API comes up, the
Secret count is close to live and every Secret decrypts with the
token. gitea-backup-restore-drill.sh extracts the latest dump and
checks app.ini (INTERNAL_TOKEN, SECRET_KEY), git fsck on every
repository and the repository count against live Gitea. Both accept a
local copy and write a drill timestamp metric.

Risk: read-only for the cluster. The k3s drill runs the restored
control plane with --network none, so its controllers reach nothing.

Verification (desktop):
- test-k3s-datastore-restore-drill.sh: a fresh k3s with three test
  Secrets (stored encrypted, k8s:enc:) restores with all Secrets
  decrypted in 7s; before folding the WAL the test saw 2 of 3 rows,
  which is why the real backup uses sqlite3 .backup.
- gitea-backup-restore-drill.sh on a live repository dump: 29
  repositories, 0 fsck failures, 29 live; the dump was deleted after.
2026-11-12 01:13:02 +09:00
DongHyeonka 012452cf78 feat(observability): alert on host backups, drift and k3s serving certs
Backups, the daily drift check and the k3s serving certificates have no
alert. k3s renews certificates only when it starts within 120 days of
expiry, and the two kube-prometheus-stack kubelet certificate alerts
can never fire because the k3s kubelet exports no
kubelet_certificate_manager_* metric.

Add a PrometheusRule (platform-host-checks) for backup age, failed
runs, missing metrics, drift left for 6h, a stale drift check, and
serving certificate expiry (warning under 100 days, critical under 14).
Add the two remaining host timers (certificate expiry via openssl on
6443 and 10250; the drift check against origin/main) and an installer
that copies the scripts to a root-owned path, creates the textfile and
work directories and enables all four timers. Disable the two dead
kubelet certificate rules in kube-prometheus-stack.

Risk: the metrics need the node-exporter textfile collector, a one-line
change in W1's node-exporter values that was asked of W1. Until then
PlatformHostBackupMetricsAbsent fires after a day, which is the point.

Verification (desktop):
- test-platform-host-checks-alerts.sh: promtool check and unit tests
  pass; with the k3s threshold changed from 2 to 5 days they fail.
- render-observability-core.sh passes; the render has no Kubelet*
  certificate alerts and still has both KubeClientCertificateExpiration.
- k3s-cert-expiry.sh as the desktop user reads both certificates.
- install-host-checks.sh --check: systemd-analyze verify is clean.
2026-11-12 01:13:02 +09:00
DongHyeonka 77daa39a41 feat(backup): back up the k3s datastore and Gitea to AIStor
Nothing backs up the k3s datastore or Gitea's repositories, packages and
app.ini. W1 covers platform-postgres only. One SSD failure loses the
source, CI artefacts, published libraries and every cluster Secret.

Add two root systemd timers that write to a new AIStor bucket on the
HDD (platform-host-backups, same AIStor as W1's postgres backups):
- k3s datastore, daily: sqlite3 .backup of the live state.db plus the
  server token (k3s backup/restore docs: the token is needed to read
  it back), integrity_check, tar.gz, sha256, upload. Off-host copy is
  prepared behind two variables (gpg recipient file, rclone target) and
  stays off until the owner chooses a place.
- Gitea, weekly: gitea dump --skip-db streamed from the live container,
  listing checked for app.ini and repositories, sha256, upload. The DB
  is in the CNPG backup.
Each run writes textfile metrics for the alerts in the next commit.
create-host-backup-credentials.sh makes the bucket (50GiB quota,
lifecycle 56 days for gitea/, 30 days for k3s-datastore/), a
bucket-scoped policy and user, and a root-only env file; no key is
printed and mc gets keys only through a 0600 env file.

Rejected: in-cluster CronJobs. The datastore is a root-owned host file
and the dump needs the Gitea DB connection; a hostPath root pod or a
second Gitea-labelled pod is a larger privilege than a root timer.

Risk: not installed here (owner sudo steps). The k3s archive holds the
token; together with the datastore it decrypts every Secret. It lives
in a bucket only the backup user can read.

Verification (desktop): a dump streamed with --file - is a valid
tar.gz with app.ini; repositories appear as repos/<owner>/<name>.git
(29, matching the live count). The scripts parse with bash -n.
2026-11-12 01:13:02 +09:00
DongHyeonka 457d88e293 feat(bootstrap): apply the k3s restart bundle with one restart
The restart bundle needs root, touches k3s, kubelet and systemd files,
and must not leave the cluster half-configured. The W1 time-sync
drop-ins need a restart window too.

apply-k3s-restart-bundle.sh --check parses the files and prints the
install plan without root. --execute checks the cluster is Ready,
backs up every file it replaces plus a sqlite .backup of the datastore
and the token, installs the W6 files and W1's two time-sync drop-ins
as they are, enables systemd-time-wait-sync, restarts k3s once with
--no-block (the unit has TimeoutStartSec=0), then verifies the kubelet
configz values, the audit log, that the default namespace rejects a
privileged pod, three health URLs and the time-sync unit. Any failure
restores the backup and restarts again. Last, it creates the
local-storage .skip file and clears the local-path default annotation.

Risk: --execute was not run (owner sudo step). The API is unavailable
for one to two minutes; containers keep running (KillMode=process). The
script refuses to run until W1's host files are on main.

Verification (desktop, non-root):
- --check without the W1 files: MISSING ... (W1), exit 1.
- --check with the W1 files overlaid: parse and invariants pass, plan
  printed, exit 0.
- --execute as non-root: "--execute needs root (sudo)".
2026-11-12 00:49:58 +09:00
DongHyeonka e88725aab8 feat(host): add the k3s kubelet, audit and pod security drop-ins
The kubelet has no memory eviction threshold and no reservations, so
under memory pressure the kernel OOM killer picks a victim and
Postgres (oom_score_adj 985) is near the front. The API server writes
no audit log. Namespaces without labels have no Pod Security level, and
graceful node shutdown is off.

Add the files for one k3s restart: a kubelet drop-in (system 1 CPU/2Gi,
kube 0.5 CPU/1Gi, hard eviction memory 500Mi plus the two k3s disk
defaults, soft eviction 1Gi for 1m, shutdown 60s/20s), an API server
drop-in that appends the audit and admission-config flags, a Metadata
audit policy and a cluster Pod Security default (baseline, warn and
audit restricted, kube-system exempt).

Rejected: protect-kernel-defaults, recommended in review 7.3. CIS
dropped it in v1.7, and the sysctls it requires include
kernel.panic_on_oops=1, which reboots this dual-boot desktop on any
kernel oops; a reboot started the 2026-11-11 clock incident. The README
records the reason.

Risk: nothing is installed by this commit. The kubelet merges the
drop-in after k3s's 00-k3s-defaults.conf (k3s passes only --config-dir);
evictionHard lists all three keys because setting the map zeroes the
others.

Verification: the apply script's --check parses every file and asserts
the invariants (next commit).
2026-11-12 00:49:58 +09:00
44 changed files with 1739 additions and 13 deletions
@@ -14,6 +14,11 @@ Kubernetes 리소스를 직접 참조하며 업스트림 매니페스트를 복
- 저장소: <https://github.com/keycloak/keycloak-k8s-resources>
- 리소스 경로: `kubernetes`
- Keycloak 및 Operator 버전: `26.7.0`
- 원격 ref: 태그가 아니라 그 태그의 커밋 `38fd7a0c84e15d4b66b37d1c2ea728ca2f5416a4`(2026-11-12). 같은 태그의 내용이
한 번 바뀌었다(build-timestamp 07-09 → 07-23, 성숙도 점검 A5).
- 이미지: Operator·Keycloak 모두 2026-11-12에 돌던 digest로 고정, Operator `imagePullPolicy: IfNotPresent`.
Keycloak CR의 `spec.image`를 정하면 Operator가 최적화 빌드된 이미지로 보므로 `startOptimized: false`를 같이 둔다.
올릴 때는 태그·커밋·두 digest와 `scripts/validate/render-phase2.sh`·`scripts/bootstrap/apply-keycloak.sh`의 검사 문자열을 함께 바꾼다.
- 설치 모드: `keycloak` namespace 범위
업스트림 namespace 범위 Deployment는 각 컨트롤러의 감시 namespace를
@@ -3,8 +3,38 @@ kind: Kustomization
namespace: keycloak
# keycloak-k8s-resources tag 26.7.0 pointed at this commit on 2026-11-12. The tag
# content changed once (live build-timestamp 2026-07-09, later renders 2026-07-23),
# so the commit is pinned instead of the tag.
resources:
- github.com/keycloak/keycloak-k8s-resources/kubernetes?ref=26.7.0
- github.com/keycloak/keycloak-k8s-resources/kubernetes?ref=38fd7a0c84e15d4b66b37d1c2ea728ca2f5416a4
# The 26.7.0 images that were running on 2026-11-12 (pod imageID), by digest.
images:
- name: quay.io/keycloak/keycloak-operator
digest: sha256:257ee324da2c701b7fc1ed1b2ce0e03979d1bf149952383bdcf6be904ae5b4f3
patches:
# Merged by container and env name (strategic merge), not by list index.
- target:
group: apps
version: v1
kind: Deployment
name: keycloak-operator
patch: |-
apiVersion: apps/v1
kind: Deployment
metadata:
name: keycloak-operator
spec:
template:
spec:
containers:
- name: keycloak-operator
imagePullPolicy: IfNotPresent
env:
- name: RELATED_IMAGE_KEYCLOAK
value: quay.io/keycloak/keycloak@sha256:0f198be292568439d700cdbfb893e69a6009bb43a94a06a945b1d3d506c76b13
labels:
- pairs:
@@ -0,0 +1,12 @@
[Unit]
Description=Back up Gitea (gitea dump --skip-db) to AIStor (platform-host-backups)
Documentation=file:///usr/local/lib/platform-backup/README.md
After=k3s.service docker.service
Requires=docker.service
[Service]
Type=oneshot
ExecStart=/bin/bash /usr/local/lib/platform-backup/backup/gitea-backup.sh
Nice=10
IOSchedulingClass=idle
TimeoutStartSec=1h
@@ -0,0 +1,10 @@
[Unit]
Description=Weekly Gitea backup
[Timer]
OnCalendar=Sun *-*-* 04:00:00
RandomizedDelaySec=10min
Persistent=true
[Install]
WantedBy=timers.target
@@ -0,0 +1,8 @@
[Unit]
Description=Record the k3s API server and kubelet serving certificate expiry
After=k3s.service
[Service]
Type=oneshot
User=donghyeon
ExecStart=/bin/bash /usr/local/lib/platform-backup/validate/k3s-cert-expiry.sh
@@ -0,0 +1,9 @@
[Unit]
Description=Daily k3s certificate expiry check
[Timer]
OnCalendar=*-*-* 05:00:00
Persistent=true
[Install]
WantedBy=timers.target
@@ -0,0 +1,11 @@
[Unit]
Description=Back up the k3s datastore and token to AIStor (platform-host-backups)
Documentation=file:///usr/local/lib/platform-backup/README.md
After=k3s.service docker.service
Requires=docker.service
[Service]
Type=oneshot
ExecStart=/bin/bash /usr/local/lib/platform-backup/backup/k3s-datastore-backup.sh
Nice=10
IOSchedulingClass=idle
@@ -0,0 +1,10 @@
[Unit]
Description=Daily k3s datastore backup
[Timer]
OnCalendar=*-*-* 03:30:00
RandomizedDelaySec=10min
Persistent=true
[Install]
WantedBy=timers.target
@@ -0,0 +1,16 @@
[Unit]
Description=Compare the cluster with platform-core origin/main (kubectl diff, server dry-run)
After=k3s.service network-online.target
Wants=network-online.target
[Service]
Type=oneshot
User=donghyeon
Environment=PLATFORM_HELM_BIN=/home/donghyeon/.local/bin/helm
Environment=PLATFORM_CHECKOUT=/home/donghyeon/workspace/platform
ExecStart=/bin/bash /usr/local/lib/platform-backup/validate/live-drift-check-main.sh
# Exit 1 means "drift found"; the metric and its alert report that. Only 2 (the
# check itself failed) marks the unit failed.
SuccessExitStatus=1
Nice=10
TimeoutStartSec=30min
@@ -0,0 +1,10 @@
[Unit]
Description=Daily live drift check
[Timer]
OnCalendar=*-*-* 05:15:00
RandomizedDelaySec=10min
Persistent=true
[Install]
WantedBy=timers.target
+92
View File
@@ -0,0 +1,92 @@
# k3s 재시작 묶음
k3s를 **한 번만** 재시작해서 다음을 같이 넣는다. 근거는 `docs/reviews/2026-11-11-k3s-maturity-review.md`
A3·A9·A10·7.2·7.3·7.5절, 결정은 같은 문서 8절 결정 5(2026-11-12 총괄 확정)다.
| 파일 | 설치 위치 | 하는 일 |
| --- | --- | --- |
| `kubelet.conf.d/50-platform-resources.conf` | `/var/lib/rancher/k3s/agent/etc/kubelet.conf.d/` | 예약(system 1 CPU·2Gi, kube 0.5 CPU·1Gi), 메모리 축출(hard 500Mi, soft 1Gi/1분), graceful node shutdown(60초, critical 20초) |
| `config.yaml.d/60-platform-apiserver.yaml` | `/etc/rancher/k3s/config.yaml.d/` | API 서버 감사 로그와 PodSecurity admission 설정 파일 지정 |
| `server/audit.yaml` | `/var/lib/rancher/k3s/server/audit.yaml` | 감사 정책: Metadata 수준(본문을 남기지 않아 Secret 값이 로그에 없다), 잦고 값 없는 요청은 뺌 |
| `server/psa.yaml` | `/var/lib/rancher/k3s/server/psa.yaml` | 라벨 없는 namespace의 기본값: enforce `baseline`, warn·audit `restricted`, `kube-system` 예외 |
| (W1) `../systemd/k3s.service.d/10-after-time-sync.conf` | `/etc/systemd/system/k3s.service.d/` | k3s를 시간 동기화 뒤에 시작 |
| (W1) `../systemd/systemd-time-wait-sync.service.d/10-timeout.conf` | `/etc/systemd/system/systemd-time-wait-sync.service.d/` | 동기화를 10분까지만 기다림 |
그리고 `/var/lib/rancher/k3s/server/manifests/local-storage.yaml.skip`을 만들어 k3s가 local-path
addon을 다시 덮어쓰지 않게 한 뒤, `local-path` StorageClass의 기본 표시를 끈다. 쓰는 PVC가 없고
(2026-11-11), 기본으로 남아 있으면 `storageClassName`을 빠뜨린 PVC가 루트 SSD에 Delete 정책으로 생긴다.
k3s 문서의 `.skip` 방식이다(packaged components). 그 뒤 local-path-provisioner는 k3s 업그레이드로
갱신되지 않는다.
## 값을 이렇게 정한 이유
- 예약: 노드 32GiB, 12 CPU. 파드 requests 합이 메모리 11Gi(34%)라 예약 3Gi를 빼도 여유가 크다.
예약이 없으면 Allocatable = Capacity라 파드가 노드 전부를 잡을 수 있고, 같은 호스트의 Docker
빌드·CI·데스크톱 앱과 겹치면 커널 OOM killer가 고른다.
- `evictionHard`: 메모리·디스크 셋을 모두 적는다. 이 맵을 정하면 정하지 않은 항목의 기본값이 0이 된다
(`mergeDefaultEvictionSettings: false`). 디스크 5%는 k3s 기본값 그대로다.
- graceful shutdown: 두 값이 0이면 기능이 꺼진다. 재부팅 때 Postgres 같은 파드가 정상 종료할 시간을
준다. 호스트 종료가 최대 60초 늦어진다. kubelet이 logind `InhibitDelayMaxSec`(지금 30초)가 모자라면
스스로 늘린다. 적용 뒤 `systemd-analyze cat-config systemd/logind.conf`로 확인한다.
- PSA 기본값은 `restricted`가 아니라 `baseline`이다. 라벨 없는 namespace는 `default`·`kube-public`·
`kube-node-lease`뿐이고 파드가 없다. 플랫폼 namespace는 모두 자기 라벨이 있어 영향이 없다.
## 넣지 않은 것: `protect-kernel-defaults`
7.3절에서 권했지만 이 묶음에서 뺐다.
- CIS Kubernetes Benchmark는 v1.7부터 이 항목(옛 4.2.6)을 뺐다.
- 켜려면 kubelet이 기대하는 sysctl을 먼저 맞춰야 한다(k3s hardening 가이드: `vm.overcommit_memory=1`,
`kernel.panic=10`, `kernel.panic_on_oops=1`, `vm.panic_on_oom=0`). `kernel.panic_on_oops=1`은 커널
oops가 한 번만 나도 10초 뒤 호스트를 재부팅한다. 이 기계는 GPU 드라이버가 있는 데스크톱이고
이중 부팅이며, 재부팅 자체가 11-11 시계 사고의 출발점이었다. `vm.overcommit_memory=1`은 호스트 전체의
메모리 할당 방식을 바꾼다.
- 얻는 것(kubelet이 커널 값을 못 바꾸게 함)보다 잃는 것이 커서 뺀다. 다시 넣으려면 sysctl 변경을 따로
소유자가 정한다.
## 적용 (소유자, 한 번)
전제:
- desktop, 이 디렉터리와 W1의 `infrastructure/host/systemd/*`가 **둘 다** 있는 platform-core `main` checkout
(W1 PR 병합 뒤). 없으면 스크립트가 `MISSING …(W1)`을 찍고 멈춘다.
- CI job과 에이전트 게이트가 없는 때. k3s 재시작 중 1~2분 API가 멈춘다. k3s는 `KillMode=process`라
컨테이너는 계속 돈다(파드가 재시작되지는 않는다).
```bash
cd /home/donghyeon/workspace/platform && git pull --ff-only
bash scripts/bootstrap/apply-k3s-restart-bundle.sh --check
sudo bash scripts/bootstrap/apply-k3s-restart-bundle.sh --execute
```
`--execute`가 하는 일:
1. 클러스터 Ready와 health 셋(Keycloak OIDC discovery, tech-log health, Gitea healthz)을 본다.
2. 바꿀 파일을 `/var/backups/platform-k3s-restart-bundle/<시각>/`에 복사하고, datastore를
`sqlite3 .backup`으로 복사한다(`token`도 같이, 모드 0600, `PRAGMA integrity_check` 출력).
3. 파일을 설치하고 `systemd-time-wait-sync`를 켠 뒤 k3s를 한 번 재시작한다.
4. 확인: kubelet configz(예약·eviction·shutdown 값), 감사 로그가 쓰이는지, `default` namespace에서
privileged 파드가 서버 dry-run으로 거절되는지, health 셋이 200인지, time-wait-sync가 enabled인지.
5. 3·4 중 하나라도 실패하면 백업한 파일을 되돌리고 새 파일을 지운 뒤 k3s를 다시 재시작한다.
6. 다 통과하면 `local-path`의 기본 표시를 끈다.
마지막 줄이 `K3S RESTART BUNDLE APPLIED`여야 한다.
## 같은 재시작 창에 이어서 할 것
PriorityClass를 파드에 붙이는 변경과 Keycloak 이미지 고정은 파드 재시작을 부르므로 이 묶음 바로 뒤에
한다(`docs/runbooks/` 재시작 창 절차). Postgres의 `priorityClassName`·Guaranteed는 W1 범위다.
## 되돌리기 (수동)
```bash
B=/var/backups/platform-k3s-restart-bundle/<시각>
sudo rm -f /var/lib/rancher/k3s/agent/etc/kubelet.conf.d/50-platform-resources.conf \
/etc/rancher/k3s/config.yaml.d/60-platform-apiserver.yaml \
/var/lib/rancher/k3s/server/manifests/local-storage.yaml.skip
sudo systemctl daemon-reload && sudo systemctl restart k3s
kubectl annotate storageclass local-path storageclass.kubernetes.io/is-default-class=true --overwrite
```
`audit.yaml`·`psa.yaml`는 참조가 없어지면 쓰이지 않으므로 남겨도 된다. W1 시간 동기화 되돌리기는
`infrastructure/host/README.md` 2단계를 따른다.
@@ -0,0 +1,12 @@
# Installed as /etc/rancher/k3s/config.yaml.d/60-platform-apiserver.yaml.
# "+" appends to kube-apiserver-arg instead of replacing it (docs.k3s.io,
# installation/configuration, multiple config files).
# Audit log: k3s hardening guide "API Server audit configuration".
# PodSecurity default: k3s hardening guide "Pod Security" (baseline here, see README).
kube-apiserver-arg+:
- "audit-log-path=/var/lib/rancher/k3s/server/logs/audit.log"
- "audit-policy-file=/var/lib/rancher/k3s/server/audit.yaml"
- "audit-log-maxage=30"
- "audit-log-maxbackup=10"
- "audit-log-maxsize=100"
- "admission-control-config-file=/var/lib/rancher/k3s/server/psa.yaml"
@@ -0,0 +1,29 @@
# Installed as /var/lib/rancher/k3s/agent/etc/kubelet.conf.d/50-platform-resources.conf.
# k3s starts the kubelet with --config-dir on that directory and keeps its own
# defaults in 00-k3s-defaults.conf; this file is merged after it.
# Why: docs/reviews/2026-11-11-k3s-maturity-review.md A3 and 7.2.
apiVersion: kubelet.config.k8s.io/v1beta1
kind: KubeletConfiguration
# Keep memory and CPU for the host (desktop, Docker builds, CI runner) and for
# k3s itself, so pods cannot take the whole node.
systemReserved:
cpu: "1"
memory: 2Gi
kubeReserved:
cpu: 500m
memory: 1Gi
# All three keys are listed on purpose: setting evictionHard replaces the
# defaults, and k3s's own default only had the two disk keys.
evictionHard:
memory.available: 500Mi
nodefs.available: "5%"
imagefs.available: "5%"
evictionSoft:
memory.available: 1Gi
evictionSoftGracePeriod:
memory.available: 1m
evictionMaxPodGracePeriod: 60
# Graceful node shutdown: on reboot or power-off the kubelet stops normal pods
# first (40s), then critical pods (20s), instead of leaving it to systemd.
shutdownGracePeriod: 60s
shutdownGracePeriodCriticalPods: 20s
+28
View File
@@ -0,0 +1,28 @@
# Installed as /var/lib/rancher/k3s/server/audit.yaml (mode 0600).
# Metadata level only: who did what to which object, never request or
# response bodies, so Secret values are not written to the log.
apiVersion: audit.k8s.io/v1
kind: Policy
omitStages:
- RequestReceived
rules:
# High-volume, low-value traffic.
- level: None
resources:
- group: coordination.k8s.io
resources: ["leases"]
- level: None
resources:
- group: ""
resources: ["events"]
- level: None
nonResourceURLs:
- /healthz*
- /livez*
- /readyz*
- /version
- /metrics
- level: None
users: ["system:kube-proxy"]
verbs: ["watch"]
- level: Metadata
+22
View File
@@ -0,0 +1,22 @@
# Installed as /var/lib/rancher/k3s/server/psa.yaml (mode 0600).
# Cluster-wide Pod Security default for namespaces without their own labels.
# Namespace labels still win: every platform namespace already sets its level.
apiVersion: apiserver.config.k8s.io/v1
kind: AdmissionConfiguration
plugins:
- name: PodSecurity
configuration:
apiVersion: pod-security.admission.config.k8s.io/v1
kind: PodSecurityConfiguration
defaults:
enforce: "baseline"
enforce-version: "latest"
audit: "restricted"
audit-version: "latest"
warn: "restricted"
warn-version: "latest"
exemptions:
usernames: []
runtimeClasses: []
# k3s helm-install jobs and local-path helper pods run here.
namespaces: [kube-system]
@@ -94,6 +94,7 @@ infrastructure/networking/traefik/
├── apply-observe.sh
├── observe-client-host.sh
├── apply-trust.sh
├── reapply-trust.sh
├── rollback-to-observe.sh
└── validate.sh
```
@@ -238,6 +239,20 @@ bash infrastructure/networking/traefik/scripts/observe-client-host.sh
`apply-trust.sh`를 실행한다. Chart, 이미지, context, API server, NodePort 경계 또는
재관측 값이 기대와 다르면 스크립트가 적용을 중단한다.
이미 trust인 상태에서 trust overlay의 다른 값(예: `resources`)만 바꿨을 때는 `apply-trust.sh`가
「observation phase must not trust forwarded headers」로 멈춘다(observe → trust 전환 전용이다). 그때는
`reapply-trust.sh`를 쓴다.
```sh
bash infrastructure/networking/traefik/scripts/reapply-trust.sh --observed-client-host 10.42.0.1 --execute
# 승인: APPLY <현재-context> RETRUST 10.42.0.1/32
```
- 적용 전에 runtime이 trust인지, 전달 헤더 신뢰가 실제로 작동하는지(Host Nginx 경유 요청의 ClientHost가
프록시 주소가 아님)를 본다. 바꿀 것을 `kubectl diff`로 보여 주고, 차이가 없으면 아무것도 하지 않는다.
- 적용 뒤 같은 검사를 다시 하고, 하나라도 실패하면 직전에 라이브였던 HelmChartConfig를 다시 적용한다.
trusted CIDR이 빠지는 일은 없다.
### 롤백
trust만 제거하고 JSON access log를 남기는 표준 롤백은 observe overlay를 적용한다.
+126
View File
@@ -0,0 +1,126 @@
#!/usr/bin/env bash
# Re-apply the trust overlay when the live Traefik is already in the trust phase,
# for example after a values change such as resources. apply-trust.sh is only for
# the observe -> trust transition and stops when the runtime already trusts the
# CIDR ("observation phase must not trust forwarded headers").
#
# bash infrastructure/networking/traefik/scripts/reapply-trust.sh \
# --observed-client-host 10.42.0.1 --execute
#
# On any failure after the apply it restores the HelmChartConfig that was live
# before, so the trusted CIDR is never dropped.
set -Eeuo pipefail
set +x
umask 077
readonly SCRIPT_DIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")" && pwd -P)"
# shellcheck source=common.sh
source "${SCRIPT_DIR}/common.sh"
[[ "$#" -eq 3 && "$1" == "--observed-client-host" && "$3" == "--execute" ]] || {
sed -n '2,11p' "${BASH_SOURCE[0]}" >&2
exit 2
}
readonly REVIEWED_CLIENT_HOST="$2"
require_commands kubectl jq rg curl nc python3 awk sort date cmp find bash
reviewed_cidr="$(host_to_exact_cidr "$REVIEWED_CLIENT_HOST")"
declared_cidr="$(normalize_exact_host_cidr "$(source_trusted_proxy_cidr)")"
[[ "$declared_cidr" != "$SENTINEL_TRUSTED_PROXY_CIDR" ]] || \
fail "trust overlay still contains the non-routable sentinel CIDR"
[[ "$declared_cidr" == "$reviewed_cidr" ]] || \
fail "declared CIDR ${declared_cidr} does not match ClientHost ${REVIEWED_CLIENT_HOST}"
bash "${SCRIPT_DIR}/validate.sh"
work_dir="$(mktemp -d "${TMPDIR:-/tmp}/platform-traefik-retrust.XXXXXX")"
restore_required=false
selected_context=""
cleanup() {
local exit_code=$?
trap - EXIT
if [[ "$exit_code" -ne 0 && "$restore_required" == "true" ]]; then
printf '\nRESTORE: re-applying the HelmChartConfig that was live before.\n' >&2
set +e
kubectl --context "$selected_context" apply --filename "${work_dir}/previous.yaml"
wait_for_runtime trust "$declared_cidr"
kubectl --context "$selected_context" --namespace kube-system \
rollout status deployment/traefik --timeout=5m
wait_for_nodeport_boundary_and_health
printf 'RESTORE complete. The previous trust values are live again.\n' >&2
set -e
fi
case "$work_dir" in
/tmp/platform-traefik-retrust.*|"${TMPDIR:-/tmp}"/platform-traefik-retrust.*) rm -rf -- "$work_dir" ;;
esac
exit "$exit_code"
}
trap cleanup EXIT
trap 'exit 130' INT
trap 'exit 143' TERM
trust_render="${work_dir}/trust.yaml"
render_overlay "$TRUST_OVERLAY" "$trust_render"
assert_live_baseline
assert_runtime trust "$declared_cidr"
assert_nodeport_boundary_and_health
# In the trust phase Traefik logs the forwarded client, not the proxy. Seeing the
# proxy address (the reviewed ClientHost) would mean the trust is not working.
assert_trust_effective() {
local seen
seen="$(observe_host_nginx_client_host)"
[[ "$seen" != "$REVIEWED_CLIENT_HOST" ]] || \
fail "Traefik logged the proxy ${seen} as ClientHost; forwarded-header trust is not in effect"
printf 'Forwarded-header trust in effect (ClientHost %s, proxy %s)\n' "$seen" "$REVIEWED_CLIENT_HOST"
}
assert_trust_effective
kubectl --namespace kube-system get helmchartconfig traefik --output=json |
python3 -I -c '
import json, sys
o = json.load(sys.stdin)
md = o["metadata"]
keep = {k: md[k] for k in ("name", "namespace", "labels", "annotations") if k in md}
keep.get("annotations", {}).pop("kubectl.kubernetes.io/last-applied-configuration", None)
print(json.dumps({"apiVersion": o["apiVersion"], "kind": o["kind"], "metadata": keep, "spec": o["spec"]}))
' >"${work_dir}/previous.yaml"
printf '\nChange to apply (kubectl diff, server dry-run):\n'
set +e
KUBECTL_EXTERNAL_DIFF="diff -u -N" kubectl diff --filename "$trust_render" | grep -E '^[-+] ' | grep -v -E '^[-+] generation:'
diff_rc=${PIPESTATUS[0]}
set -e
if [[ "$diff_rc" -eq 0 ]]; then
printf 'Live HelmChartConfig already matches the trust overlay. Nothing to do.\n'
exit 0
fi
(( diff_rc == 1 )) || fail "kubectl diff failed"
selected_context="$(current_context)"
selected_api_server="$(current_api_server)"
printf '\nKubernetes context: %s\nAPI server: %s\nTrusted CIDR (unchanged): %s\n' \
"$selected_context" "$selected_api_server" "$declared_cidr"
printf 'Type APPLY %s RETRUST %s to continue: ' "$selected_context" "$declared_cidr"
read -r confirmation
[[ "$confirmation" == "APPLY ${selected_context} RETRUST ${declared_cidr}" ]] || fail "cancelled"
[[ "$(current_context)" == "$selected_context" ]] || fail "kubectl context changed after confirmation"
[[ "$(current_api_server)" == "$selected_api_server" ]] || fail "Kubernetes API server changed after confirmation"
assert_runtime trust "$declared_cidr"
restore_required=true
kubectl apply --filename "$trust_render"
wait_for_runtime trust "$declared_cidr"
kubectl --namespace kube-system rollout status deployment/traefik --timeout=5m
assert_runtime trust "$declared_cidr"
assert_live_hcc_matches_overlay "$TRUST_OVERLAY" "${work_dir}/post-apply-trust.yaml"
assert_live_baseline
wait_for_nodeport_boundary_and_health
assert_manifest_https
assert_trust_effective
restore_required=false
printf '\nTRUST PHASE RE-APPLIED\n'
printf 'Traefik web trusted CIDR: %s\n' "$declared_cidr"
printf 'Gitea health and HTTPS site-manifest checks: PASS\n'
+68
View File
@@ -0,0 +1,68 @@
#!/usr/bin/env bash
# Back up Gitea (repositories, packages, attachments, Actions data, app.ini) to
# AIStor (HDD sdb3). Root only. Run by platform-gitea-backup.timer.
#
# `gitea dump --skip-db` runs inside the live Gitea container and streams a
# tar.gz to stdout (logs go to stderr). The database is backed up separately by
# CloudNativePG (services/platform-postgres-backup). The dump contains app.ini,
# whose secrets are needed to restore, so the archive is sensitive.
set -Eeuo pipefail
umask 077
# shellcheck source=../lib/host-backup.sh
source "$(dirname -- "${BASH_SOURCE[0]}")/../lib/host-backup.sh"
readonly NAME=gitea_backup
readonly KCFG=/etc/rancher/k3s/k3s.yaml
[[ "$(id -u)" -eq 0 ]] || hb_fail "run as root"
hb_load_aistor_env
kubectl() { command /usr/local/bin/k3s kubectl --kubeconfig "$KCFG" "$@"; }
started="$(date +%s)"
previous_success="$(awk '/^platform_backup_last_success_timestamp_seconds/ {print $2}' "$TEXTFILE_DIR/$NAME.prom" 2>/dev/null || echo 0)"
on_error() {
hb_write_metrics "$NAME" <<PROM
platform_backup_last_run_success{backup="gitea"} 0
platform_backup_last_run_timestamp_seconds{backup="gitea"} $(date +%s)
platform_backup_last_success_timestamp_seconds{backup="gitea"} ${previous_success:-0}
PROM
}
trap on_error ERR
work="$(mktemp -d "$HB_WORK_ROOT/gitea.XXXXXX")"
trap 'rm -rf -- "$work"' EXIT
ts="$(date -u +%Y%m%dT%H%M%SZ)"
archive="gitea-dump-$ts.tar.gz"
kubectl -n gitea exec deploy/gitea -c gitea -- \
gitea dump --skip-db --type tar.gz --file - >"$work/$archive" 2>"$work/dump.log"
tar -tzf "$work/$archive" >"$work/list.txt"
grep -qx 'app.ini' "$work/list.txt" || hb_fail "the dump has no app.ini"
repos="$(grep -c -E '^repos/[^/]+/[^/]+\.git/HEAD$' "$work/list.txt" || true)"
(( repos > 0 )) || hb_fail "the dump has no repositories"
sha256sum "$work/$archive" | awk '{print $1}' >"$work/$archive.sha256"
size="$(stat -c %s "$work/$archive")"
# What a restore must match: restore with the same Gitea version (docs/runbooks/gitea-restore.md).
{
echo "created=$ts"
echo "gitea_image=$(kubectl -n gitea get deploy gitea -o jsonpath='{.spec.template.spec.containers[0].image}')"
echo "gitea_version=$(kubectl -n gitea exec deploy/gitea -c gitea -- gitea --version 2>/dev/null | awk '{print $3}')"
echo "repositories=$repos"
} >"$work/$archive.meta"
hb_mc "$work" cp --quiet "/work/$archive" "/work/$archive.sha256" "/work/$archive.meta" "backup/$AISTOR_BUCKET/gitea/" >/dev/null
hb_mc "$work" stat "backup/$AISTOR_BUCKET/gitea/$archive" >/dev/null
trap - ERR
hb_write_metrics "$NAME" <<PROM
# HELP platform_backup_last_success_timestamp_seconds Unix time of the last successful backup.
# TYPE platform_backup_last_success_timestamp_seconds gauge
platform_backup_last_success_timestamp_seconds{backup="gitea"} $(date +%s)
platform_backup_last_run_success{backup="gitea"} 1
platform_backup_last_run_timestamp_seconds{backup="gitea"} $(date +%s)
platform_backup_last_size_bytes{backup="gitea"} $size
platform_backup_last_duration_seconds{backup="gitea"} $(( $(date +%s) - started ))
# HELP platform_backup_gitea_repositories Repositories in the last successful Gitea dump.
# TYPE platform_backup_gitea_repositories gauge
platform_backup_gitea_repositories $repos
PROM
hb_log "gitea backup $archive ($size bytes, $repos repositories) uploaded"
+99
View File
@@ -0,0 +1,99 @@
#!/usr/bin/env bash
# Back up the k3s datastore (sqlite) and the server token to AIStor (HDD sdb3).
# Root only. Run by platform-k3s-datastore-backup.timer (infrastructure/host/backup).
#
# k3s docs (datastore/backup-restore, SQLite): copy server/db and server/token;
# the token is required to read the datastore back. `sqlite3 .backup` takes a
# consistent copy of the live database.
#
# Off-host copy (prepared, off by default): set in /etc/platform-backup/aistor.env
# OFFHOST_GPG_RECIPIENT_FILE=/etc/platform-backup/offhost-recipient.asc
# OFFHOST_RCLONE_TARGET=<rclone remote:path>
# The archive is then encrypted to that public key and copied with rclone.
set -Eeuo pipefail
umask 077
# shellcheck source=../lib/host-backup.sh
source "$(dirname -- "${BASH_SOURCE[0]}")/../lib/host-backup.sh"
readonly SERVER=/var/lib/rancher/k3s/server
readonly NAME=k3s_datastore_backup
[[ "$(id -u)" -eq 0 ]] || hb_fail "run as root"
command -v sqlite3 >/dev/null || hb_fail "sqlite3 is required"
hb_load_aistor_env
started="$(date +%s)"
on_error() {
hb_write_metrics "$NAME" <<PROM
# HELP platform_backup_last_run_success 1 when the last run of this backup succeeded.
# TYPE platform_backup_last_run_success gauge
platform_backup_last_run_success{backup="k3s-datastore"} 0
platform_backup_last_run_timestamp_seconds{backup="k3s-datastore"} $(date +%s)
platform_backup_last_success_timestamp_seconds{backup="k3s-datastore"} ${previous_success:-0}
PROM
}
previous_success="$(awk '/^platform_backup_last_success_timestamp_seconds/ {print $2}' "$TEXTFILE_DIR/$NAME.prom" 2>/dev/null || echo 0)"
trap on_error ERR
work="$(mktemp -d "$HB_WORK_ROOT/k3s.XXXXXX")"
trap 'rm -rf -- "$work"' EXIT
ts="$(date -u +%Y%m%dT%H%M%SZ)"
stage="$work/k3s-datastore-$ts"
mkdir -m 0700 "$stage"
sqlite3 "$SERVER/db/state.db" ".backup '$stage/state.db'"
check="$(sqlite3 "$stage/state.db" 'PRAGMA integrity_check;')"
[[ "$check" == "ok" ]] || hb_fail "integrity_check on the copy returned: $check"
install -m 0600 "$SERVER/token" "$stage/token"
{
echo "created=$ts"
echo "k3s_version=$(/usr/local/bin/k3s --version | head -1)"
echo "host=$(hostname)"
echo "kine_rows=$(sqlite3 "$stage/state.db" 'SELECT COUNT(*) FROM kine;')"
} >"$stage/META"
tar -C "$work" -czf "$work/k3s-datastore-$ts.tar.gz" "k3s-datastore-$ts"
sha256sum "$work/k3s-datastore-$ts.tar.gz" | awk '{print $1}' >"$work/k3s-datastore-$ts.tar.gz.sha256"
size="$(stat -c %s "$work/k3s-datastore-$ts.tar.gz")"
hb_mc "$work" cp --quiet "/work/k3s-datastore-$ts.tar.gz" "/work/k3s-datastore-$ts.tar.gz.sha256" "backup/$AISTOR_BUCKET/k3s-datastore/" >/dev/null
hb_mc "$work" stat "backup/$AISTOR_BUCKET/k3s-datastore/k3s-datastore-$ts.tar.gz" >/dev/null
# Local copy outside AIStor on the same HDD (sdb3). AIStor runs inside the
# cluster this archive restores, so after losing the cluster the AIStor copy
# cannot be read until k3s is back; this copy can. The directory sits next to
# AIStor's drive directory (data/), so AIStor does not see it as a bucket.
readonly LOCAL_DIR=/srv/k3s/aistor/host-backups/k3s-datastore
install -d -m 0700 -o root -g root "$(dirname "$LOCAL_DIR")" "$LOCAL_DIR"
install -m 0600 "$work/k3s-datastore-$ts.tar.gz" "$work/k3s-datastore-$ts.tar.gz.sha256" "$LOCAL_DIR/"
ls -1t "$LOCAL_DIR"/k3s-datastore-*.tar.gz | tail -n +8 | while read -r old; do rm -f -- "$old" "$old.sha256"; done
local_copies="$(ls -1 "$LOCAL_DIR"/k3s-datastore-*.tar.gz | wc -l)"
offhost=0
if [[ -n "${OFFHOST_GPG_RECIPIENT_FILE:-}" && -n "${OFFHOST_RCLONE_TARGET:-}" ]]; then
command -v rclone >/dev/null || hb_fail "OFFHOST_RCLONE_TARGET is set but rclone is not installed"
gpg --batch --yes --trust-model always --recipient-file "$OFFHOST_GPG_RECIPIENT_FILE" \
--output "$work/k3s-datastore-$ts.tar.gz.gpg" --encrypt "$work/k3s-datastore-$ts.tar.gz"
rclone copyto "$work/k3s-datastore-$ts.tar.gz.gpg" "$OFFHOST_RCLONE_TARGET/k3s-datastore-$ts.tar.gz.gpg"
offhost=1
fi
trap - ERR
hb_write_metrics "$NAME" <<PROM
# HELP platform_backup_last_success_timestamp_seconds Unix time of the last successful backup.
# TYPE platform_backup_last_success_timestamp_seconds gauge
platform_backup_last_success_timestamp_seconds{backup="k3s-datastore"} $(date +%s)
# HELP platform_backup_last_run_success 1 when the last run of this backup succeeded.
# TYPE platform_backup_last_run_success gauge
platform_backup_last_run_success{backup="k3s-datastore"} 1
platform_backup_last_run_timestamp_seconds{backup="k3s-datastore"} $(date +%s)
# HELP platform_backup_last_size_bytes Size of the last successful backup archive.
# TYPE platform_backup_last_size_bytes gauge
platform_backup_last_size_bytes{backup="k3s-datastore"} $size
platform_backup_last_duration_seconds{backup="k3s-datastore"} $(( $(date +%s) - started ))
# HELP platform_backup_local_copies Archives kept outside AIStor on the HDD.
# TYPE platform_backup_local_copies gauge
platform_backup_local_copies{backup="k3s-datastore"} $local_copies
# HELP platform_backup_offhost_copy 1 when the last run also wrote the encrypted off-host copy.
# TYPE platform_backup_offhost_copy gauge
platform_backup_offhost_copy{backup="k3s-datastore"} $offhost
PROM
hb_log "k3s datastore backup k3s-datastore-$ts.tar.gz ($size bytes) uploaded; off-host=$offhost"
@@ -0,0 +1,238 @@
#!/usr/bin/env bash
# Install the k3s restart bundle and restart k3s exactly once.
#
# bash scripts/bootstrap/apply-k3s-restart-bundle.sh --check # no root, changes nothing
# sudo bash scripts/bootstrap/apply-k3s-restart-bundle.sh --execute
#
# The bundle (infrastructure/host/k3s/README.md):
# W6: kubelet reservations, memory eviction and graceful shutdown (kubelet drop-in),
# API server audit log and cluster Pod Security default (k3s drop-in + two files),
# local-path StorageClass no longer the default (manifest .skip + annotation).
# W1: k3s starts after time sync (infrastructure/host/systemd/*, installed as is).
#
# --execute backs up every file it replaces plus a datastore copy, installs, restarts
# k3s, verifies, and on any failure restores the previous files and restarts again.
set -Eeuo pipefail
umask 077
readonly ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd -P)"
readonly KCFG=/etc/rancher/k3s/k3s.yaml
readonly SERVER=/var/lib/rancher/k3s/server
readonly SKIP_FILE="$SERVER/manifests/local-storage.yaml.skip"
readonly AUDIT_DIR="$SERVER/logs"
readonly BACKUP_ROOT=/var/backups/platform-k3s-restart-bundle
# source | destination | mode | owner of the source file
readonly -a FILES=(
"infrastructure/host/k3s/kubelet.conf.d/50-platform-resources.conf|/var/lib/rancher/k3s/agent/etc/kubelet.conf.d/50-platform-resources.conf|0600|W6"
"infrastructure/host/k3s/config.yaml.d/60-platform-apiserver.yaml|/etc/rancher/k3s/config.yaml.d/60-platform-apiserver.yaml|0644|W6"
"infrastructure/host/k3s/server/audit.yaml|$SERVER/audit.yaml|0600|W6"
"infrastructure/host/k3s/server/psa.yaml|$SERVER/psa.yaml|0600|W6"
"infrastructure/host/systemd/k3s.service.d/10-after-time-sync.conf|/etc/systemd/system/k3s.service.d/10-after-time-sync.conf|0644|W1"
"infrastructure/host/systemd/systemd-time-wait-sync.service.d/10-timeout.conf|/etc/systemd/system/systemd-time-wait-sync.service.d/10-timeout.conf|0644|W1"
)
readonly -a HEALTH_URLS=(
https://id.learn.hyeonworks.com/realms/hyeonworks/.well-known/openid-configuration
https://hyeonworks.com/api/healthcheck
https://git.learn.hyeonworks.com/api/healthz
)
say() { printf '%s\n' "$*"; }
fail() { printf 'ERROR: %s\n' "$*" >&2; exit 1; }
mode="${1:-}"
[[ "$#" -eq 1 && ( "$mode" == "--check" || "$mode" == "--execute" ) ]] || {
sed -n '2,15p' "${BASH_SOURCE[0]}" >&2
exit 2
}
check_sources() {
local entry src dst perm owner missing=0
for entry in "${FILES[@]}"; do
IFS='|' read -r src dst perm owner <<<"$entry"
if [[ ! -f "$ROOT/$src" ]]; then
say "MISSING $src (${owner})"
missing=1
fi
done
(( missing == 0 )) || fail "run this from a platform-core checkout that has both the W6 and the W1 host files on main"
python3 -I - "$ROOT" <<'PY' || fail "bundle files do not parse"
import sys, pathlib, yaml
root = pathlib.Path(sys.argv[1]) / "infrastructure/host/k3s"
k = yaml.safe_load((root / "kubelet.conf.d/50-platform-resources.conf").read_text())
assert k["kind"] == "KubeletConfiguration", "kubelet drop-in kind"
assert set(k["evictionHard"]) == {"memory.available", "nodefs.available", "imagefs.available"}, "evictionHard keys"
assert k["shutdownGracePeriod"] == "60s" and k["shutdownGracePeriodCriticalPods"] == "20s", "shutdown periods"
a = yaml.safe_load((root / "config.yaml.d/60-platform-apiserver.yaml").read_text())
assert list(a) == ["kube-apiserver-arg+"], "apiserver drop-in must only append kube-apiserver-arg"
args = dict(x.split("=", 1) for x in a["kube-apiserver-arg+"])
assert args["audit-policy-file"] == "/var/lib/rancher/k3s/server/audit.yaml"
assert args["admission-control-config-file"] == "/var/lib/rancher/k3s/server/psa.yaml"
p = yaml.safe_load((root / "server/audit.yaml").read_text())
assert p["kind"] == "Policy" and p["rules"][-1] == {"level": "Metadata"}, "audit policy ends with Metadata"
assert all(r["level"] in ("None", "Metadata") for r in p["rules"]), "audit policy must not log bodies"
s = yaml.safe_load((root / "server/psa.yaml").read_text())
d = s["plugins"][0]["configuration"]["defaults"]
assert d["enforce"] == "baseline", "PSA default enforce"
assert s["plugins"][0]["configuration"]["exemptions"]["namespaces"] == ["kube-system"], "PSA exemptions"
print("bundle files parse and keep their invariants")
PY
}
plan() {
local entry src dst perm owner state
for entry in "${FILES[@]}"; do
IFS='|' read -r src dst perm owner <<<"$entry"
if [[ -d "$(dirname "$dst")" && ! -r "$(dirname "$dst")" ]] ||
[[ ! -e "$(dirname "$dst")" && ! -r "$(dirname "$(dirname "$dst")")" ]]; then
state="unknown (needs root to read)"
elif [[ ! -e "$dst" ]]; then
state="new"
elif cmp -s "$ROOT/$src" "$dst"; then
state="same"
else
state="replace"
fi
say " ${owner} ${state} ${dst}"
done
say " W6 touch $SKIP_FILE"
say " W6 mkdir $AUDIT_DIR (0700)"
say " W1 systemctl enable systemd-time-wait-sync.service"
say " -- systemctl daemon-reload && systemctl restart k3s (once)"
}
say "== check"
check_sources
say "== plan"
plan
[[ "$mode" == "--execute" ]] || { say "CHECK ONLY: nothing changed"; exit 0; }
[[ "$(id -u)" -eq 0 ]] || fail "--execute needs root (sudo)"
command -v sqlite3 >/dev/null || fail "sqlite3 is required for the pre-restart datastore copy"
kubectl() { command k3s kubectl --kubeconfig "$KCFG" "$@"; }
wait_ready() {
local i
for i in $(seq 1 60); do
if kubectl get --raw=/readyz >/dev/null 2>&1 &&
[[ "$(kubectl get nodes -o jsonpath='{.items[0].status.conditions[?(@.type=="Ready")].status}' 2>/dev/null)" == "True" ]]; then
return 0
fi
sleep 5
done
return 1
}
say "== preflight"
wait_ready || fail "the cluster is not Ready before the change; not touching it"
for url in "${HEALTH_URLS[@]}"; do
code="$(curl -s -o /dev/null -m 15 -w '%{http_code}' "$url" || true)"
say " before $code $url"
done
ts="$(date -u +%Y%m%dT%H%M%SZ)"
backup="$BACKUP_ROOT/$ts"
install -d -m 0700 "$backup"
time_wait_was_enabled="$(systemctl is-enabled systemd-time-wait-sync.service 2>/dev/null || true)"
say "== backup to $backup"
for entry in "${FILES[@]}"; do
IFS='|' read -r src dst perm owner <<<"$entry"
if [[ -e "$dst" ]]; then
install -D -m 0600 "$dst" "$backup/files$dst"
fi
done
sqlite3 "$SERVER/db/state.db" ".backup '$backup/state.db'"
install -m 0600 "$SERVER/token" "$backup/token"
say " datastore copy: $(sqlite3 "$backup/state.db" 'PRAGMA integrity_check;') $(du -h "$backup/state.db" | cut -f1)"
rollback() {
local entry src dst perm owner
trap - ERR
say "== ROLLBACK"
set +e
for entry in "${FILES[@]}"; do
IFS='|' read -r src dst perm owner <<<"$entry"
if [[ -e "$backup/files$dst" ]]; then
install -D -m "$perm" "$backup/files$dst" "$dst"
else
rm -f -- "$dst"
fi
done
rm -f -- "$SKIP_FILE"
[[ "$time_wait_was_enabled" == "enabled" ]] || systemctl disable systemd-time-wait-sync.service
systemctl daemon-reload
systemctl restart --no-block k3s
sleep 10
if wait_ready; then say "ROLLBACK done: previous files restored, k3s Ready"; else say "ROLLBACK: k3s still not Ready; see journalctl -u k3s"; fi
say "backup kept at $backup"
exit 1
}
trap rollback ERR
say "== install"
for entry in "${FILES[@]}"; do
IFS='|' read -r src dst perm owner <<<"$entry"
install -D -o root -g root -m "$perm" "$ROOT/$src" "$dst"
done
install -d -o root -g root -m 0700 "$AUDIT_DIR"
install -D -o root -g root -m 0600 /dev/null "$SKIP_FILE"
systemctl daemon-reload
systemctl enable systemd-time-wait-sync.service
say "== restart k3s"
# k3s.service has TimeoutStartSec=0, so a blocking restart could wait forever.
systemctl restart --no-block k3s
sleep 10
wait_ready || { say "k3s did not become Ready in 5 minutes"; false; }
say "== verify"
node="$(kubectl get nodes -o jsonpath='{.items[0].metadata.name}')"
kubectl get --raw "/api/v1/nodes/$node/proxy/configz" | python3 -I -c '
import json, sys
k = json.load(sys.stdin)["kubeletconfig"]
want = {
("systemReserved", "memory"): "2Gi", ("kubeReserved", "memory"): "1Gi",
("evictionHard", "memory.available"): "500Mi", ("evictionHard", "nodefs.available"): "5%",
("evictionHard", "imagefs.available"): "5%", ("evictionSoft", "memory.available"): "1Gi",
}
bad = [f"{a}.{b}={k.get(a, {}).get(b)}" for (a, b), v in want.items() if k.get(a, {}).get(b) != v]
if k.get("shutdownGracePeriod") != "1m0s": bad.append(f"shutdownGracePeriod={k.get(\"shutdownGracePeriod\")}")
if bad: sys.exit("kubelet config not applied: " + ", ".join(bad))
print(" kubelet: reservations, eviction and graceful shutdown applied")
'
for i in $(seq 1 12); do [[ -s "$AUDIT_DIR/audit.log" ]] && break; sleep 5; done
[[ -s "$AUDIT_DIR/audit.log" ]] || { say "audit log is empty"; false; }
say " audit log: $(wc -l <"$AUDIT_DIR/audit.log") lines"
psa_out="$(kubectl apply --dry-run=server -f - 2>&1 <<'YAML' || true
apiVersion: v1
kind: Pod
metadata:
name: psa-default-probe
namespace: default
spec:
containers:
- name: probe
image: registry.k8s.io/pause:3.10
securityContext:
privileged: true
YAML
)"
grep -q 'violates PodSecurity "baseline' <<<"$psa_out" || { say "PSA default did not reject a privileged pod: $psa_out"; false; }
say " pod security: default namespace rejects a privileged pod (baseline)"
for url in "${HEALTH_URLS[@]}"; do
ok=0
for i in $(seq 1 30); do
[[ "$(curl -s -o /dev/null -m 15 -w '%{http_code}' "$url" || true)" == "200" ]] && { ok=1; break; }
sleep 10
done
(( ok == 1 )) || { say "health check failed: $url"; false; }
say " after 200 $url"
done
say " time sync: $(systemctl is-enabled systemd-time-wait-sync.service), $(systemctl show systemd-time-wait-sync.service -p TimeoutStartUSec)"
trap - ERR
say "== local-path is no longer the default StorageClass"
kubectl annotate storageclass local-path storageclass.kubernetes.io/is-default-class=false --overwrite
kubectl get storageclass local-path -o jsonpath=' local-path default={.metadata.annotations.storageclass\.kubernetes\.io/is-default-class}{"\n"}'
say "K3S RESTART BUNDLE APPLIED (backup: $backup)"
+2 -2
View File
@@ -189,8 +189,8 @@ assert_regex_count \
assert_regex_count \
"${REPOSITORY_ROOT}/infrastructure/controllers/keycloak-operator/kustomization.yaml" \
'github\.com/keycloak/keycloak-k8s-resources/kubernetes\?ref=26\.7\.0' 1 \
"Keycloak Operator 26.7.0 source pin"
'github\.com/keycloak/keycloak-k8s-resources/kubernetes\?ref=38fd7a0c84e15d4b66b37d1c2ea728ca2f5416a4' 1 \
"Keycloak Operator 26.7.0 commit pin"
assert_regex_count \
"${render_temp_dir}/keycloak-operator.yaml" \
'^kind:[[:space:]]CustomResourceDefinition$' 4 \
@@ -0,0 +1,81 @@
#!/usr/bin/env bash
# Create the AIStor bucket, lifecycle rules, policy and user that the host backup
# timers use, and the root-only /etc/platform-backup/aistor.env.
#
# sudo bash scripts/bootstrap/create-host-backup-credentials.sh
#
# Run once, by the owner, on the desktop. Refuses to run when the env file exists;
# never rotates or deletes a bucket, policy, user or object. No key is printed.
set -Eeuo pipefail
set +x
umask 077
readonly ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd -P)"
# shellcheck source=../lib/host-backup.sh
source "$ROOT/scripts/lib/host-backup.sh"
readonly BUCKET=platform-host-backups
readonly POLICY=platform-host-backups
readonly QUOTA=50GiB
readonly POLICY_FILE="$ROOT/services/platform-host-checks/bucket-policy.json"
readonly PASS_LINE=PLATFORM_HOST_BACKUP_BOOTSTRAP_PASS
kubectl() { command /usr/local/bin/k3s kubectl --kubeconfig /etc/rancher/k3s/k3s.yaml "$@"; }
[[ "$(id -u)" -eq 0 ]] || hb_fail "run with sudo"
for c in docker jq openssl; do command -v "$c" >/dev/null || hb_fail "missing command: $c"; done
jq -e . "$POLICY_FILE" >/dev/null || hb_fail "bucket policy is not valid JSON"
[[ ! -e "$HOST_BACKUP_ENV" ]] || hb_fail "$HOST_BACKUP_ENV already exists; this script does not rotate it"
ip="$(kubectl -n object-storage get service minio -o jsonpath='{.spec.clusterIP}')"
[[ -n "$ip" ]] || hb_fail "object-storage/minio Service not found"
endpoint="http://${ip}:80"
work="$(mktemp -d)"
trap 'rm -rf -- "$work"' EXIT
kubectl -n object-storage get secret aistor-root-configuration -o jsonpath='{.data.config\.env}' | base64 -d >"$work/root.env"
grep -q '^export MINIO_ROOT_USER=' "$work/root.env" || hb_fail "aistor-root-configuration has an unexpected format"
printf 'pcbackup%s' "$(openssl rand -hex 8)" >"$work/ACCESS_KEY"
openssl rand -hex 32 | tr -d '\n' >"$work/SECRET_KEY"
cp "$POLICY_FILE" "$work/policy.json"
printf '%s\n' \
"Bucket: $BUCKET (quota $QUOTA; expire gitea/ after 56 days, k3s-datastore/ after 30 days) on $endpoint" \
"Policy: $POLICY (${POLICY_FILE#"$ROOT/"})" \
"User: a new generated access key attached to $POLICY" \
"File: $HOST_BACKUP_ENV (root, 0600)" \
'No key value will be printed.'
printf 'Type APPLY to create them: '
IFS= read -r confirmation
[[ "$confirmation" == "APPLY" ]] || hb_fail "confirmation did not match APPLY"
# Every mc call is silenced so no key reaches the terminal.
docker run --rm --network host -v "$work:/run/b:ro" --tmpfs /tmp -e MC_CONFIG_DIR=/tmp/mc \
--entrypoint /bin/sh "$MC_IMAGE" -ec '
. /run/b/root.env
ak="$(cat /run/b/ACCESS_KEY)"; sk="$(cat /run/b/SECRET_KEY)"
mc alias set root "'"$endpoint"'" "$MINIO_ROOT_USER" "$MINIO_ROOT_PASSWORD" --api S3v4 >/dev/null
mc mb --ignore-existing "root/'"$BUCKET"'" >/dev/null
mc quota set "root/'"$BUCKET"'" --size "'"$QUOTA"'" >/dev/null
mc ilm rule add --prefix gitea/ --expire-days 56 "root/'"$BUCKET"'" >/dev/null
mc ilm rule add --prefix k3s-datastore/ --expire-days 30 "root/'"$BUCKET"'" >/dev/null
mc admin policy create root "'"$POLICY"'" /run/b/policy.json >/dev/null
mc admin user add root "$ak" "$sk" >/dev/null
mc admin policy attach root "'"$POLICY"'" --user "$ak" >/dev/null
mc alias set backup "'"$endpoint"'" "$ak" "$sk" --api S3v4 >/dev/null
probe=".bootstrap-probe-$(date +%s)"
printf probe | mc pipe "backup/'"$BUCKET"'/$probe" >/dev/null
[ "$(mc cat "backup/'"$BUCKET"'/$probe")" = probe ]
mc rm "backup/'"$BUCKET"'/$probe" >/dev/null
if mc ls backup/loki >/dev/null 2>&1; then echo "the backup user can list another bucket" >&2; exit 1; fi
echo "'"$PASS_LINE"'"' | grep -qx "$PASS_LINE" || hb_fail "bootstrap did not finish; $HOST_BACKUP_ENV was not written"
install -d -m 0700 -o root -g root "$(dirname "$HOST_BACKUP_ENV")"
{
echo "AISTOR_BUCKET=$BUCKET"
echo "AISTOR_ACCESS_KEY=$(cat "$work/ACCESS_KEY")"
echo "AISTOR_SECRET_KEY=$(cat "$work/SECRET_KEY")"
echo "# Off-host encrypted copy of the k3s datastore archive (off until both are set):"
echo "# OFFHOST_GPG_RECIPIENT_FILE=/etc/platform-backup/offhost-recipient.asc"
echo "# OFFHOST_RCLONE_TARGET=remote:path"
} >"$HOST_BACKUP_ENV"
chmod 0600 "$HOST_BACKUP_ENV"
printf '%s\n' "$PASS_LINE" "Created bucket $BUCKET, policy $POLICY, one user and $HOST_BACKUP_ENV."
+78
View File
@@ -0,0 +1,78 @@
#!/usr/bin/env bash
# Install the host backup and check timers (infrastructure/host/backup).
#
# bash scripts/bootstrap/install-host-checks.sh --check # no root, changes nothing
# sudo bash scripts/bootstrap/install-host-checks.sh --execute # install, enable, first runs
#
# Scripts are copied to /usr/local/lib/platform-backup (root-owned) so the root
# timers never execute files the desktop user can edit.
set -Eeuo pipefail
umask 022
readonly ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd -P)"
readonly LIB=/usr/local/lib/platform-backup
readonly UNIT_DIR=/etc/systemd/system
readonly TEXTFILE_DIR=/var/lib/node_exporter/textfile_collector
readonly WORK_ROOT=/var/lib/platform-backup/work
readonly USER_NAME=donghyeon
readonly -a SCRIPTS=(
scripts/lib/host-backup.sh:lib/host-backup.sh
scripts/backup/k3s-datastore-backup.sh:backup/k3s-datastore-backup.sh
scripts/backup/gitea-backup.sh:backup/gitea-backup.sh
scripts/validate/live-drift-check-main.sh:validate/live-drift-check-main.sh
scripts/validate/k3s-cert-expiry.sh:validate/k3s-cert-expiry.sh
services/platform-host-checks/README.md:README.md
)
readonly -a TIMERS=(platform-k3s-datastore-backup platform-gitea-backup platform-live-drift-check platform-k3s-cert-expiry)
fail() { printf 'ERROR: %s\n' "$*" >&2; exit 1; }
mode="${1:-}"
[[ "$#" -eq 1 && ( "$mode" == "--check" || "$mode" == "--execute" ) ]] || { sed -n '2,8p' "$0" >&2; exit 2; }
echo "== check"
for pair in "${SCRIPTS[@]}"; do
src="${pair%%:*}"
[[ -f "$ROOT/$src" ]] || fail "missing $src"
[[ "$src" != *.sh ]] || bash -n "$ROOT/$src" || fail "syntax error in $src"
done
for t in "${TIMERS[@]}"; do
for u in service timer; do [[ -f "$ROOT/infrastructure/host/backup/$t.$u" ]] || fail "missing $t.$u"; done
done
if command -v systemd-analyze >/dev/null; then
verify_dir="$(mktemp -d)"
trap 'rm -rf -- "$verify_dir"' EXIT
cp "$ROOT"/infrastructure/host/backup/*.service "$ROOT"/infrastructure/host/backup/*.timer "$verify_dir/"
# ExecStart paths do not exist before installation; only syntax and keys are checked.
out="$(cd "$verify_dir" && systemd-analyze verify ./*.timer ./*.service 2>&1 | grep -v -E 'is not executable|No such file|Command .* is not executable' || true)"
[[ -z "$out" ]] || { printf '%s\n' "$out" >&2; fail "systemd-analyze verify reported problems"; }
echo " units: systemd-analyze verify clean"
fi
echo "== plan"
for pair in "${SCRIPTS[@]}"; do echo " install ${pair%%:*} -> $LIB/${pair##*:}"; done
for t in "${TIMERS[@]}"; do echo " install $UNIT_DIR/$t.service, $UNIT_DIR/$t.timer; enable --now $t.timer"; done
echo " mkdir $TEXTFILE_DIR (root:$USER_NAME 0775), $WORK_ROOT (root 0700)"
echo " first runs: k3s-cert-expiry, k3s-datastore-backup, gitea-backup, live-drift-check"
[[ "$mode" == "--execute" ]] || { echo "CHECK ONLY: nothing changed"; exit 0; }
[[ "$(id -u)" -eq 0 ]] || fail "--execute needs root (sudo)"
[[ -f /etc/platform-backup/aistor.env ]] || fail "run scripts/bootstrap/create-host-backup-credentials.sh first"
echo "== install"
for pair in "${SCRIPTS[@]}"; do
install -D -o root -g root -m 0755 "$ROOT/${pair%%:*}" "$LIB/${pair##*:}"
done
install -d -o root -g "$USER_NAME" -m 0775 "$TEXTFILE_DIR"
install -d -o root -g root -m 0700 "$WORK_ROOT"
for t in "${TIMERS[@]}"; do
install -o root -g root -m 0644 "$ROOT/infrastructure/host/backup/$t.service" "$ROOT/infrastructure/host/backup/$t.timer" "$UNIT_DIR/"
done
systemctl daemon-reload
for t in "${TIMERS[@]}"; do systemctl enable --now "$t.timer"; done
echo "== first runs"
for t in platform-k3s-cert-expiry platform-k3s-datastore-backup platform-gitea-backup platform-live-drift-check; do
if systemctl start "$t.service"; then echo " $t: ok"; else echo " $t: FAILED (journalctl -u $t)"; fi
done
echo "== metrics"
grep -h -E '^platform_' "$TEXTFILE_DIR"/*.prom || echo " no metrics yet"
systemctl list-timers 'platform-*' --no-pager
echo "HOST CHECKS INSTALLED"
+55
View File
@@ -0,0 +1,55 @@
#!/usr/bin/env bash
# Shared helpers for the host backup and host check scripts. Source only.
# shellcheck shell=bash
readonly HOST_BACKUP_ENV=/etc/platform-backup/aistor.env
readonly TEXTFILE_DIR=/var/lib/node_exporter/textfile_collector
# Backup staging (root 0700, same SSD). Not /var/tmp: dockerd must see the path.
readonly HB_WORK_ROOT=/var/lib/platform-backup/work
# Same pinned AIStor client as scripts/validate/aistor-s3-smoke.sh.
readonly MC_IMAGE="quay.io/minio/aistor/mc@sha256:c5ec777c080fd6292b7529309a0f311c58c3a92e268c9c57c0901da86881949b"
hb_log() { printf '%s %s\n' "$(date -u +%FT%TZ)" "$*"; }
hb_fail() { hb_log "FAIL: $*" >&2; exit 1; }
# Write Prometheus textfile metrics atomically: hb_write_metrics NAME <<'PROM' ... PROM
hb_write_metrics() {
local name="$1" tmp
[[ -d "$TEXTFILE_DIR" ]] || { hb_log "textfile dir $TEXTFILE_DIR missing; metrics not written"; cat >/dev/null; return 0; }
tmp="$(mktemp "$TEXTFILE_DIR/.${name}.XXXXXX")"
cat >"$tmp"
chmod 0644 "$tmp"
mv -f -- "$tmp" "$TEXTFILE_DIR/${name}.prom"
}
# Load the AIStor endpoint and the backup user's keys (root-only file).
hb_load_aistor_env() {
[[ -f "$HOST_BACKUP_ENV" ]] || hb_fail "$HOST_BACKUP_ENV is missing; run scripts/bootstrap/create-host-backup-credentials.sh"
[[ "$(stat -c '%u %a' "$HOST_BACKUP_ENV")" == "0 600" ]] || hb_fail "$HOST_BACKUP_ENV must be owned by root with mode 0600"
# shellcheck disable=SC1090
. "$HOST_BACKUP_ENV"
: "${AISTOR_BUCKET:?}" "${AISTOR_ACCESS_KEY:?}" "${AISTOR_SECRET_KEY:?}"
# The host cannot resolve cluster DNS, so use the Service ClusterIP (port 80 -> 9000).
if [[ -z "${AISTOR_ENDPOINT:-}" ]]; then
local ip
ip="$(/usr/local/bin/k3s kubectl --kubeconfig /etc/rancher/k3s/k3s.yaml -n object-storage \
get service minio -o jsonpath='{.spec.clusterIP}')"
[[ -n "$ip" ]] || hb_fail "cannot find the object-storage/minio Service"
AISTOR_ENDPOINT="http://${ip}:80"
fi
}
# Run the pinned mc client on the host network with the backup user's keys.
# The keys go through a 0600 env file, never the command line (ps would show it).
# Usage: hb_mc <dir mounted at /work> <mc args...>
hb_mc() {
local work="$1"; shift
local envfile="$work/.mc.env"
if [[ ! -f "$envfile" ]]; then
( umask 077; printf 'MC_HOST_backup=%s\nMC_CONFIG_DIR=/tmp/mc\n' \
"${AISTOR_ENDPOINT/:\/\//://${AISTOR_ACCESS_KEY}:${AISTOR_SECRET_KEY}@}" >"$envfile" )
fi
docker run --rm --network host --user "$(id -u):$(id -g)" \
--env-file "$envfile" -v "$work:/work" --tmpfs /tmp \
"$MC_IMAGE" "$@"
}
+10
View File
@@ -596,3 +596,13 @@ PLATFORM_HELM_BIN=/home/donghyeon/.local/bin/helm \
- `--textfile FILE`: `platform_live_drift_objects`, `platform_live_drift_check_success`,
`platform_live_drift_last_run_timestamp_seconds`를 Prometheus textfile 형식으로 쓴다.
- Secret 값의 차이는 보지 않는다.
## 호스트 백업·점검 시험
- `test-platform-host-checks-alerts.sh`: `services/platform-host-checks/prometheus-rules.yaml`의 promtool
단위 시험(백업 신선도, 실패, 지표 없음, drift, k3s 인증서). docker 필요, 클러스터 접근 없음.
- `k3s-cert-expiry.sh`: 6443·10250 serving 인증서 만료를 textfile 지표로 쓴다(타이머가 매일 돌림).
- `live-drift-check-main.sh`: checkout의 작업 트리가 아니라 `origin/main`으로 `live-drift-check.sh`를 돌린다.
- `k3s-datastore-restore-drill.sh`, `gitea-backup-restore-drill.sh`: 복원 리허설. 인자 없이 root로 돌리면 AIStor의
최신 백업을 쓰고, `--from-dir`·`--from-file`로 로컬 사본을 줄 수 있다.
- `test-k3s-datastore-restore-drill.sh`: 운영 데이터 없이 k3s 복원 리허설 경로를 시험한다(일회용 k3s, 시험 Secret 셋).
@@ -0,0 +1,75 @@
#!/usr/bin/env bash
# Restore drill for the Gitea backup (file level).
#
# sudo bash scripts/validate/gitea-backup-restore-drill.sh # latest dump in AIStor
# bash scripts/validate/gitea-backup-restore-drill.sh --from-file DUMP.tar.gz
#
# Extracts the dump into a throwaway directory and checks what a restore needs:
# app.ini is present, every repository passes `git fsck`, and the repository
# count matches the live Gitea. The database is restored from the CloudNativePG
# backup instead (services/platform-postgres-backup); the full restore order is
# in docs/runbooks/gitea-restore.md. Nothing is written to the cluster.
set -Eeuo pipefail
umask 077
readonly ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd -P)"
# shellcheck source=../lib/host-backup.sh
source "$ROOT/scripts/lib/host-backup.sh"
readonly NAME=gitea_restore_drill
from_file=""
case "${1:-}" in
--from-file) from_file="${2:?--from-file needs a path}" ;;
"") ;;
*) sed -n '2,11p' "$0" >&2; exit 2 ;;
esac
if [[ "$(id -u)" -eq 0 ]]; then
kubectl() { command /usr/local/bin/k3s kubectl --kubeconfig /etc/rancher/k3s/k3s.yaml "$@"; }
fi
base="$HB_WORK_ROOT"; [[ -w "$base" ]] || base="$(mktemp -d)"
work="$(mktemp -d "$base/gitea-drill.XXXXXX")"
trap 'rm -rf -- "$work"' EXIT
if [[ -z "$from_file" ]]; then
[[ "$(id -u)" -eq 0 ]] || hb_fail "run as root to read the backup credentials"
hb_load_aistor_env
latest="$(hb_mc "$work" ls "backup/$AISTOR_BUCKET/gitea/" | awk '{print $NF}' | grep -E '^gitea-dump-.*\.tar\.gz$' | sort | tail -1)"
[[ -n "$latest" ]] || hb_fail "no Gitea dump found"
hb_mc "$work" cp --quiet "backup/$AISTOR_BUCKET/gitea/$latest" "backup/$AISTOR_BUCKET/gitea/$latest.sha256" /work/ >/dev/null
[[ "$(sha256sum "$work/$latest" | awk '{print $1}')" == "$(cat "$work/$latest.sha256")" ]] || hb_fail "checksum mismatch for $latest"
from_file="$work/$latest"
hb_log "drill on $latest"
fi
mkdir "$work/restore"
tar -C "$work/restore" -xzf "$from_file"
[[ -s "$work/restore/app.ini" ]] || hb_fail "app.ini is missing from the dump"
for key in INTERNAL_TOKEN SECRET_KEY; do
grep -q "^${key}[[:space:]]*=" "$work/restore/app.ini" || hb_fail "app.ini has no ${key} (needed to restore)"
done
total=0 broken=0
while IFS= read -r -d '' repo; do
total=$((total + 1))
if ! git -c safe.directory='*' --git-dir="$repo" fsck --no-dangling --no-progress >/dev/null 2>"$work/fsck.err"; then
broken=$((broken + 1))
hb_log "fsck failed: ${repo#"$work/restore/"}: $(head -1 "$work/fsck.err")"
fi
done < <(find "$work/restore/repos" -mindepth 2 -maxdepth 2 -type d -name '*.git' -print0)
hb_log "repositories in dump: $total, fsck failures: $broken"
(( total > 0 )) || hb_fail "the dump has no repositories"
(( broken == 0 )) || hb_fail "$broken repositories failed git fsck"
live="$(kubectl -n gitea exec deploy/gitea -c gitea -- sh -c 'find /data/git/gitea-repositories -mindepth 2 -maxdepth 2 -name "*.git" -type d | wc -l' 2>/dev/null || echo "")"
if [[ -n "$live" ]]; then
hb_log "live Gitea repositories: $live"
(( total >= live - 2 )) || hb_fail "the dump has $total repositories, live Gitea has $live"
fi
hb_write_metrics "$NAME" <<PROM
# HELP platform_backup_restore_drill_last_success_timestamp_seconds Unix time of the last successful restore drill.
# TYPE platform_backup_restore_drill_last_success_timestamp_seconds gauge
platform_backup_restore_drill_last_success_timestamp_seconds{backup="gitea"} $(date +%s)
PROM
echo "GITEA RESTORE DRILL PASS"
+25
View File
@@ -0,0 +1,25 @@
#!/usr/bin/env bash
# Write the notAfter time of the k3s API server (6443) and kubelet (10250)
# serving certificates as Prometheus textfile metrics. No root needed.
# k3s renews these only when it starts and they are within 120 days of expiry
# (docs.k3s.io cli/certificate), and the kubelet exposes no certificate metric,
# so nothing else in the cluster watches them.
set -Eeuo pipefail
# shellcheck source=../lib/host-backup.sh
source "$(dirname -- "${BASH_SOURCE[0]}")/../lib/host-backup.sh"
lines=""
for target in apiserver:6443 kubelet:10250; do
name="${target%%:*}" port="${target##*:}"
end="$(echo | timeout 10 openssl s_client -connect "127.0.0.1:$port" 2>/dev/null | openssl x509 -noout -enddate 2>/dev/null | sed 's/^notAfter=//')" || true
[[ -n "$end" ]] || hb_fail "could not read the $name certificate on 127.0.0.1:$port"
lines+="platform_k3s_certificate_not_after_seconds{certificate=\"$name\"} $(date -d "$end" +%s)"$'\n'
hb_log "$name certificate notAfter $end"
done
hb_write_metrics platform_k3s_certificates <<PROM
# HELP platform_k3s_certificate_not_after_seconds notAfter of a k3s serving certificate (Unix time).
# TYPE platform_k3s_certificate_not_after_seconds gauge
${lines}# HELP platform_k3s_certificate_check_timestamp_seconds Unix time of the last certificate check.
# TYPE platform_k3s_certificate_check_timestamp_seconds gauge
platform_k3s_certificate_check_timestamp_seconds $(date +%s)
PROM
@@ -0,0 +1,99 @@
#!/usr/bin/env bash
# Restore drill for the k3s datastore backup.
#
# sudo bash scripts/validate/k3s-datastore-restore-drill.sh # latest backup in AIStor
# bash scripts/validate/k3s-datastore-restore-drill.sh --from-dir DIR [--expect-secrets N]
# DIR holds state.db and token (an extracted archive, or a test datastore)
#
# Starts a throwaway k3s server (same version as production, --network none,
# no agent) on a copy of the backup, and checks that the API answers, that the
# Secret count matches and that every Secret decrypts (the token unlocks the
# encryption config stored in the datastore). Prints no Secret value. The
# container and the copy are removed at the end.
set -Eeuo pipefail
umask 077
readonly ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd -P)"
# shellcheck source=../lib/host-backup.sh
source "$ROOT/scripts/lib/host-backup.sh"
# Same version as the desktop server (k3s --version: v1.36.2+k3s1).
readonly K3S_IMAGE="docker.io/rancher/k3s:v1.36.2-k3s1@sha256:6a47cea22c4b834d4ba72c89d291696b79ebe406251f90b446e4dff03513dd87"
readonly NAME=k3s_datastore_restore_drill
from_dir="" expect_secrets=""
while (( $# > 0 )); do
case "$1" in
--from-dir) from_dir="${2:?}"; shift 2 ;;
--expect-secrets) expect_secrets="${2:?}"; shift 2 ;;
*) sed -n '2,13p' "$0" >&2; exit 2 ;;
esac
done
base="${HB_WORK_ROOT}"
[[ -w "$base" ]] || base="$(mktemp -d)"
work="$(mktemp -d "$base/k3s-drill.XXXXXX")"
container="k3s-restore-drill-$$"
cleanup() {
docker rm -f "$container" >/dev/null 2>&1 || true
# The drill container writes as root; remove its files with the same image.
docker run --rm -v "$work:/w" --entrypoint /bin/sh "$K3S_IMAGE" -c 'rm -rf /w/*' >/dev/null 2>&1 || true
rm -rf -- "$work"
}
trap cleanup EXIT
if [[ -z "$from_dir" ]]; then
[[ "$(id -u)" -eq 0 ]] || hb_fail "run as root to read the backup credentials"
hb_load_aistor_env
latest="$(hb_mc "$work" ls "backup/$AISTOR_BUCKET/k3s-datastore/" | awk '{print $NF}' | grep -E '^k3s-datastore-.*\.tar\.gz$' | sort | tail -1)"
[[ -n "$latest" ]] || hb_fail "no k3s datastore backup found"
hb_mc "$work" cp --quiet "backup/$AISTOR_BUCKET/k3s-datastore/$latest" "backup/$AISTOR_BUCKET/k3s-datastore/$latest.sha256" /work/ >/dev/null
[[ "$(sha256sum "$work/$latest" | awk '{print $1}')" == "$(cat "$work/$latest.sha256")" ]] || hb_fail "checksum mismatch for $latest"
tar -C "$work" -xzf "$work/$latest"
from_dir="$(find "$work" -mindepth 1 -maxdepth 1 -type d -name 'k3s-datastore-*' | head -1)"
hb_log "drill on $latest"
if [[ -z "$expect_secrets" ]]; then
expect_secrets="$(/usr/local/bin/k3s kubectl --kubeconfig /etc/rancher/k3s/k3s.yaml get secrets -A --no-headers | wc -l)"
hb_log "live cluster has $expect_secrets Secrets (counts may differ by objects created since the backup)"
fi
fi
[[ -f "$from_dir/state.db" && -f "$from_dir/token" ]] || hb_fail "$from_dir must contain state.db and token"
install -d -m 0700 "$work/server/db"
cp "$from_dir/state.db" "$work/server/db/state.db"
cp "$from_dir/token" "$work/server/token"
[[ "$(sqlite3 "$work/server/db/state.db" 'PRAGMA integrity_check;')" == "ok" ]] || hb_fail "integrity_check failed"
started="$(date +%s)"
docker run -d --name "$container" --network none --tmpfs /run --tmpfs /var/run \
-v "$work/server:/var/lib/rancher/k3s/server" \
"$K3S_IMAGE" server --disable-agent --secrets-encryption \
--token-file /var/lib/rancher/k3s/server/token \
--bind-address 127.0.0.1 --advertise-address 127.0.0.1 --node-ip 127.0.0.1 \
--disable traefik,servicelb,metrics-server,local-storage,coredns >/dev/null
ready=0
for _ in $(seq 1 60); do
if docker exec "$container" kubectl get --raw /readyz >/dev/null 2>&1; then ready=1; break; fi
sleep 5
done
if (( ready == 0 )); then
docker logs --tail 20 "$container" 2>&1 | sed 's/^/ k3s: /' >&2
hb_fail "restored k3s did not become ready in 5 minutes"
fi
namespaces="$(docker exec "$container" kubectl get namespaces --no-headers | wc -l)"
secrets="$(docker exec "$container" kubectl get secrets -A --no-headers | wc -l)"
# -o json decrypts every Secret on the server side; any failure makes kubectl exit non-zero.
decrypted="$(docker exec "$container" kubectl get secrets -A -o json | python3 -I -c 'import json,sys; d=json.load(sys.stdin)["items"]; print(sum(1 for s in d if s.get("data")))')"
hb_log "restored: namespaces=$namespaces secrets=$secrets with-data=$decrypted in $(( $(date +%s) - started ))s"
if [[ -n "$expect_secrets" ]]; then
diff=$(( secrets - expect_secrets ))
(( ${diff#-} <= 5 )) || hb_fail "Secret count $secrets is far from the expected $expect_secrets"
fi
(( decrypted > 0 )) || hb_fail "no Secret carried data after restore"
hb_write_metrics "$NAME" <<PROM
# HELP platform_backup_restore_drill_last_success_timestamp_seconds Unix time of the last successful restore drill.
# TYPE platform_backup_restore_drill_last_success_timestamp_seconds gauge
platform_backup_restore_drill_last_success_timestamp_seconds{backup="k3s-datastore"} $(date +%s)
PROM
echo "K3S DATASTORE RESTORE DRILL PASS"
+6 -1
View File
@@ -3,8 +3,13 @@
Cluster platform-data platform-postgres 2026-11-30 pg_hba의 techlog 두 줄이 repo에 없다. w1이 CNPG 백업과 함께 넣는다(1c4ef9a).
DatabaseRole platform-data platform-postgres-gitea 2026-11-30 렌더가 false 기본값 필드를 명시할 뿐 값은 같다. 다음 Phase 1 적용 때 사라진다.
DatabaseRole platform-data platform-postgres-keycloak 2026-11-30 렌더가 false 기본값 필드를 명시할 뿐 값은 같다. 다음 Phase 2 적용 때 사라진다.
Deployment keycloak keycloak-operator 2026-11-30 원격 ref 26.7.0 태그 내용이 바뀌어 build-timestamp 라벨만 다르다. 재시작 창에 커밋 SHA 고정과 함께 적용한다.
Deployment keycloak keycloak-operator 2026-11-30 커밋 SHA·이미지 digest·IfNotPresent 고정(재시작). 재시작 창에 apply-keycloak.sh로 적용한다.
Service keycloak keycloak-operator 2026-11-30 위와 같은 build-timestamp 라벨.
ServiceAccount keycloak keycloak-operator 2026-11-30 위와 같은 build-timestamp 라벨.
Deployment observability observability-core-kube-pr-operator 2026-11-30 이미지 문자열이 tag@digest와 digest로 다르고 digest는 같다. 적용하면 재시작되므로 재시작 창에.
Deployment observability observability-core-kube-state-metrics 2026-11-30 위와 같은 이미지 문자열 차이.
Keycloak keycloak keycloak 2026-11-30 이미지 digest·startOptimized false·platform-critical(Keycloak 재시작). 재시작 창에 PriorityClass를 먼저 만든 뒤 적용한다.
Deployment gitea gitea 2026-11-30 priorityClassName platform-critical(Gitea 재시작). 재시작 창에 apply-gitea-oidc.sh로 적용한다.
StatefulSet observability loki 2026-11-30 priorityClassName platform-low(Loki 재시작). 재시작 창에 관측 core 적용으로.
StatefulSet observability tempo 2026-11-30 priorityClassName platform-low(Tempo 재시작). 재시작 창에 관측 core 적용으로.
PrometheusRule observability observability-core-kube-pr-kubernetes-system-kubelet 2026-11-30 Kubelet 인증서 경보 둘을 values에서 껐다(PR #6). 관측 core를 다시 적용하면 사라진다(재시작 창 2단계).
+16
View File
@@ -0,0 +1,16 @@
#!/usr/bin/env bash
# Daily drift check against origin/main (not the working tree of the checkout),
# so local edits or a stale checkout do not hide or invent drift.
# Run by platform-live-drift-check.timer as the desktop user.
set -Eeuo pipefail
readonly CHECKOUT="${PLATFORM_CHECKOUT:-/home/donghyeon/workspace/platform}"
readonly TEXTFILE=/var/lib/node_exporter/textfile_collector/platform_live_drift.prom
git -C "$CHECKOUT" fetch --quiet origin main
work="$(mktemp -d /tmp/platform-drift-main.XXXXXX)"
trap 'rm -rf -- "$work"' EXIT
git -C "$CHECKOUT" archive origin/main | tar -x -C "$work"
printf 'origin/main %s\n' "$(git -C "$CHECKOUT" rev-parse --short origin/main)"
args=(--accepted "$work/scripts/validate/live-drift-accepted.txt")
[[ -d "$(dirname "$TEXTFILE")" ]] && args+=(--textfile "$TEXTFILE")
bash "$work/scripts/validate/live-drift-check.sh" "${args[@]}"
+24 -5
View File
@@ -64,10 +64,29 @@ die() {
work="$(mktemp -d /tmp/platform-drift.XXXXXX)"; made_dirs+=("$work")
mkdir -p "$work/in" "$work/out"
# Render from a clean copy when run inside a git working tree. The observability
# renderer refuses to run when an ignored chart cache (.helm/) is left over from an
# earlier render, which happens in long-lived checkouts. The copy keeps tracked and
# untracked-but-not-ignored files, so local edits are still checked.
SOURCE_ROOT="$REPOSITORY_ROOT"
if git -C "$REPOSITORY_ROOT" rev-parse --is-inside-work-tree >/dev/null 2>&1; then
SOURCE_ROOT="$work/src"
mkdir -p "$SOURCE_ROOT"
python3 -I - "$REPOSITORY_ROOT" <<'PY' | tar -C "$REPOSITORY_ROOT" --null -T - -cf - | tar -x -C "$SOURCE_ROOT" || die "could not copy the working tree"
import subprocess, sys
root = sys.argv[1]
def ls(*args):
out = subprocess.run(["git", "-C", root, "ls-files", "-z", *args], check=True, capture_output=True).stdout
return [p for p in out.decode().split("\0") if p]
keep = set(ls("--cached", "--others", "--exclude-standard")) - set(ls("--deleted"))
sys.stdout.write("\0".join(sorted(keep)) + "\0")
PY
fi
render_root() {
local script="$1" prefix="$2" dir
dir="$(mktemp -d "/tmp/${prefix}.XXXXXX")"; made_dirs+=("$dir")
bash "$REPOSITORY_ROOT/scripts/validate/${script}" --verified-output-dir "$dir" \
bash "$SOURCE_ROOT/scripts/validate/${script}" --verified-output-dir "$dir" \
>"$work/${script}.log" 2>&1 || { tail -5 "$work/${script}.log" >&2; die "${script} failed"; }
printf '%s\n' "$dir"
}
@@ -86,10 +105,10 @@ declare -a inputs=(
for f in "$phase2"/*.yaml "$admin"/*.yaml; do inputs+=("$f"); done
# Live Traefik is the trust overlay (infrastructure/networking/traefik/README.md).
kubectl kustomize "$REPOSITORY_ROOT/infrastructure/networking/traefik/overlays/trust" >"$work/traefik-trust.yaml" || die "traefik render failed"
kubectl kustomize "$REPOSITORY_ROOT/services/tenants/techlog" >"$work/tenant-techlog.yaml" || die "tenant render failed"
kubectl kustomize "$REPOSITORY_ROOT/services/priority-classes" >"$work/priority-classes.yaml" || die "priority class render failed"
kubectl kustomize "$REPOSITORY_ROOT/infrastructure/networking/keycloak" >"$work/keycloak-network-policies.yaml" || die "keycloak policy render failed"
kubectl kustomize "$SOURCE_ROOT/infrastructure/networking/traefik/overlays/trust" >"$work/traefik-trust.yaml" || die "traefik render failed"
kubectl kustomize "$SOURCE_ROOT/services/tenants/techlog" >"$work/tenant-techlog.yaml" || die "tenant render failed"
kubectl kustomize "$SOURCE_ROOT/services/priority-classes" >"$work/priority-classes.yaml" || die "priority class render failed"
kubectl kustomize "$SOURCE_ROOT/infrastructure/networking/keycloak" >"$work/keycloak-network-policies.yaml" || die "keycloak policy render failed"
inputs+=("$work/traefik-trust.yaml" "$work/tenant-techlog.yaml" "$work/priority-classes.yaml" "$work/keycloak-network-policies.yaml")
n=0
+5 -1
View File
@@ -266,7 +266,8 @@ aistor_objectstore_kustomization="${REPOSITORY_ROOT}/services/minio-aistor/base/
aistor_operator_values="${REPOSITORY_ROOT}/infrastructure/controllers/aistor-operator/values/home.yaml"
aistor_objectstore_values="${REPOSITORY_ROOT}/services/minio-aistor/base/deployment/values/home.yaml"
assert_regex_count "$keycloak_operator_kustomization" '^[[:space:]]*-[[:space:]]github\.com/keycloak/keycloak-k8s-resources/kubernetes\?ref=26\.7\.0[[:space:]]*$' 1 "Keycloak Operator remote ref must be 26.7.0"
assert_regex_count "$keycloak_operator_kustomization" '^[[:space:]]*-[[:space:]]github\.com/keycloak/keycloak-k8s-resources/kubernetes\?ref=38fd7a0c84e15d4b66b37d1c2ea728ca2f5416a4[[:space:]]*$' 1 "Keycloak Operator remote ref must be the 26.7.0 commit"
assert_regex_count "$keycloak_operator_kustomization" '^[[:space:]]*digest:[[:space:]]sha256:257ee324da2c701b7fc1ed1b2ce0e03979d1bf149952383bdcf6be904ae5b4f3[[:space:]]*$' 1 "Keycloak Operator image must be pinned by digest"
assert_regex_count "$keycloak_operator_kustomization" 'github\.com/keycloak/keycloak-k8s-resources/kubernetes' 1 "exactly one Keycloak Operator remote resource is allowed"
assert_regex_count "$keycloak_operator_kustomization" '\?ref=' 1 "exactly one pinned remote ref is allowed in the Keycloak Operator root"
@@ -342,6 +343,9 @@ assert_regex_count "${render_dir}/keycloak.yaml" '^kind: Ingress$' 1 "Keycloak I
assert_regex_count "${render_dir}/minio-aistor.yaml" '^kind: ObjectStore$' 1 "AIStor ObjectStore count"
assert_regex_count "${render_dir}/keycloak.yaml" '^[[:space:]]*instances:[[:space:]]1[[:space:]]*$' 1 "Keycloak instance count field"
assert_regex_count "${render_dir}/keycloak.yaml" '^[[:space:]]*image:[[:space:]]quay\.io/keycloak/keycloak@sha256:[0-9a-f]{64}[[:space:]]*$' 1 "Keycloak server image must be pinned by digest"
assert_regex_count "${render_dir}/keycloak.yaml" '^[[:space:]]*startOptimized:[[:space:]]false[[:space:]]*$' 1 "the stock Keycloak image must not start optimized"
assert_regex_count "${render_dir}/keycloak.yaml" '^[[:space:]]*priorityClassName:[[:space:]]platform-critical[[:space:]]*$' 1 "Keycloak priority class"
assert_regex_count "${render_dir}/keycloak.yaml" '^[[:space:]]*hostname:[[:space:]]https://id\.learn\.hyeonworks\.com[[:space:]]*$' 1 "Keycloak external hostname"
assert_regex_count "${render_dir}/keycloak.yaml" '^[[:space:]]*strict:[[:space:]]true[[:space:]]*$' 1 "Keycloak strict hostname mode"
assert_regex_count "${render_dir}/keycloak.yaml" '^[[:space:]]*httpEnabled:[[:space:]]true[[:space:]]*$' 1 "Keycloak internal HTTP mode"
@@ -0,0 +1,41 @@
#!/usr/bin/env bash
# Test k3s-datastore-restore-drill.sh without production data: start a fresh,
# network-less k3s with secrets encryption, add three test Secrets, stop it,
# and run the drill on its state.db and token. Needs docker; no root.
set -Eeuo pipefail
umask 077
readonly ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd -P)"
readonly K3S_IMAGE="$(sed -n 's/^readonly K3S_IMAGE="\(.*\)"$/\1/p' "$ROOT/scripts/validate/k3s-datastore-restore-drill.sh")"
[[ -n "$K3S_IMAGE" ]] || { echo "cannot read K3S_IMAGE from the drill script" >&2; exit 1; }
work="$(mktemp -d)"
seed="w6-k3s-drill-seed-$$"
cleanup() {
docker rm -f "$seed" >/dev/null 2>&1 || true
docker run --rm -v "$work:/w" --entrypoint /bin/sh "$K3S_IMAGE" -c 'rm -rf /w/*' >/dev/null 2>&1 || true
rm -rf -- "$work"
}
trap cleanup EXIT
mkdir -p "$work/server" "$work/copy"
docker run -d --name "$seed" --network none --tmpfs /run --tmpfs /var/run \
-v "$work/server:/var/lib/rancher/k3s/server" "$K3S_IMAGE" server --disable-agent --secrets-encryption \
--bind-address 127.0.0.1 --advertise-address 127.0.0.1 --node-ip 127.0.0.1 \
--disable traefik,servicelb,metrics-server,local-storage,coredns >/dev/null
for _ in $(seq 1 60); do docker exec "$seed" kubectl get --raw /readyz >/dev/null 2>&1 && break; sleep 5; done
for i in 1 2 3; do docker exec "$seed" kubectl create secret generic "drill-test-$i" --from-literal=value="test-$i" >/dev/null; done
expected="$(docker exec "$seed" kubectl get secrets -A --no-headers | wc -l)"
docker stop -t 30 "$seed" >/dev/null
docker run --rm -v "$work:/w" --entrypoint /bin/sh "$K3S_IMAGE" -c \
"cp /w/server/db/state.db* /w/server/token /w/copy/ && chown $(id -u):$(id -g) /w/copy/*"
# Fold the WAL into state.db, as `sqlite3 .backup` does in the real backup.
sqlite3 "$work/copy/state.db" 'PRAGMA wal_checkpoint(TRUNCATE);' >/dev/null
rm -f "$work/copy/state.db-wal" "$work/copy/state.db-shm"
# The test Secrets must be stored encrypted (k8s:enc:...), so the drill really
# proves that the token unlocks the encryption key.
encrypted="$(sqlite3 "$work/copy/state.db" "SELECT COUNT(*) FROM kine WHERE name LIKE '/registry/secrets/default/drill-test-%' AND CAST(value AS TEXT) LIKE 'k8s:enc:%';")"
echo "seed: $expected Secrets, encrypted test Secret rows=$encrypted"
(( encrypted >= 3 )) || { echo "test Secrets are not encrypted at rest in the seed datastore" >&2; exit 1; }
bash "$ROOT/scripts/validate/k3s-datastore-restore-drill.sh" --from-dir "$work/copy" --expect-secrets "$expected"
echo "K3S DATASTORE RESTORE DRILL TEST PASS"
@@ -0,0 +1,151 @@
#!/usr/bin/env bash
# promtool unit tests for services/platform-host-checks/prometheus-rules.yaml.
# Needs docker and python3 (PyYAML). No cluster access.
set -Eeuo pipefail
readonly ROOT="$(cd -- "$(dirname -- "${BASH_SOURCE[0]}")/../.." && pwd -P)"
# Same pinned image as test-platform-observability-rules.sh.
readonly PROMETHEUS_IMAGE="quay.io/prometheus/prometheus:v3.13.2-distroless@sha256:ce95cfa77eff5aad28bd7a65aff19868cf78d9e17e4c254da7dfe22ade78318b"
work="$(mktemp -d)"
trap 'rm -rf -- "$work"' EXIT
python3 -I - "$ROOT/services/platform-host-checks/prometheus-rules.yaml" "$work/rules.yaml" <<'PY'
import sys, yaml
rule = yaml.safe_load(open(sys.argv[1]))
yaml.safe_dump({"groups": rule["spec"]["groups"]}, open(sys.argv[2], "w"), sort_keys=False)
PY
# promtool evaluates from Unix time 0, so "now" in each test is the eval_time.
cat >"$work/tests.yaml" <<'EOF'
rule_files:
- rules.yaml
evaluation_interval: 1m
tests:
# k3s datastore: last success at t=0; stale after 2 days, alert after 1h more.
- interval: 1m
input_series:
- series: 'platform_backup_last_success_timestamp_seconds{backup="k3s-datastore"}'
values: '0x5500'
- series: 'platform_backup_last_run_timestamp_seconds{backup="k3s-datastore"}'
values: '0x5500'
- series: 'platform_backup_last_run_timestamp_seconds{backup="gitea"}'
values: '0x5500'
- series: 'platform_backup_last_success_timestamp_seconds{backup="gitea"}'
values: '0x5500'
alert_rule_test:
- eval_time: 24h
alertname: PlatformK3sDatastoreBackupTooOld
exp_alerts: []
- eval_time: 50h
alertname: PlatformK3sDatastoreBackupTooOld
exp_alerts:
- exp_labels: {severity: warning}
exp_annotations:
summary: k3s datastore has not been backed up for more than 2 days
description: Check `journalctl -u platform-k3s-datastore-backup` on desktop. Daily timer, 03:30 KST.
- eval_time: 90h
alertname: PlatformGiteaBackupTooOld
exp_alerts: []
- eval_time: 90h
alertname: PlatformHostBackupMetricsAbsent
exp_alerts: []
- interval: 1m
input_series:
- series: 'platform_backup_last_success_timestamp_seconds{backup="gitea"}'
values: '0x11700'
alert_rule_test:
- eval_time: 194h
alertname: PlatformGiteaBackupTooOld
exp_alerts:
- exp_labels: {severity: warning}
exp_annotations:
summary: Gitea has not been backed up for more than 8 days
description: Check `journalctl -u platform-gitea-backup` on desktop. Weekly timer, Sunday 04:00 KST.
# A failed run is reported with its backup label.
- interval: 1m
input_series:
- series: 'platform_backup_last_run_success{backup="gitea",instance="desktop"}'
values: '0x30'
- series: 'platform_backup_last_run_success{backup="k3s-datastore",instance="desktop"}'
values: '1x30'
alert_rule_test:
- eval_time: 15m
alertname: PlatformHostBackupFailed
exp_alerts:
- exp_labels: {severity: warning, backup: gitea}
exp_annotations:
summary: "The last gitea backup run failed"
description: "Check `journalctl -u platform-gitea-backup` on desktop."
# Metrics never written (timers or textfile collector missing).
- interval: 1h
input_series:
- series: 'up{job="prometheus-node-exporter"}'
values: '1x30'
alert_rule_test:
- eval_time: 25h
alertname: PlatformHostBackupMetricsAbsent
exp_alerts:
- exp_labels: {severity: warning, backup: k3s-datastore}
exp_annotations:
summary: Host backup metrics are missing
description: The backup timers or the node-exporter textfile collector (/var/lib/node_exporter/textfile_collector) are not running.
- exp_labels: {severity: warning, backup: gitea}
exp_annotations:
summary: Host backup metrics are missing
description: The backup timers or the node-exporter textfile collector (/var/lib/node_exporter/textfile_collector) are not running.
# Drift for 7h fires; drift 0 does not.
- interval: 1m
input_series:
- series: 'platform_live_drift_objects'
values: '3x500'
- series: 'platform_live_drift_last_run_timestamp_seconds'
values: '0x500'
- series: 'platform_live_drift_check_success'
values: '1x500'
alert_rule_test:
- eval_time: 5h
alertname: PlatformLiveDriftDetected
exp_alerts: []
- eval_time: 7h
alertname: PlatformLiveDriftDetected
exp_alerts:
- exp_labels: {severity: warning}
exp_annotations:
summary: "3 live objects differ from platform-core origin/main"
description: Run `scripts/validate/live-drift-check.sh --accepted scripts/validate/live-drift-accepted.txt` on desktop to list them.
- eval_time: 7h
alertname: PlatformLiveDriftCheckStale
exp_alerts: []
# Certificates: 50 days left -> warning only; 10 days left -> warning and critical.
- interval: 1m
input_series:
- series: 'platform_k3s_certificate_not_after_seconds{certificate="apiserver"}'
values: '4320000x200'
- series: 'platform_k3s_certificate_not_after_seconds{certificate="kubelet"}'
values: '864000x200'
- series: 'platform_k3s_certificate_check_timestamp_seconds'
values: '0x200'
alert_rule_test:
- eval_time: 2h
alertname: PlatformK3sServingCertificateExpiry
exp_alerts:
- exp_labels: {severity: warning, certificate: apiserver}
exp_annotations:
summary: "k3s apiserver serving certificate expires in less than 100 days"
description: k3s renews certificates only when it starts within 120 days of expiry. Plan a k3s restart window (docs/runbooks).
- exp_labels: {severity: warning, certificate: kubelet}
exp_annotations:
summary: "k3s kubelet serving certificate expires in less than 100 days"
description: k3s renews certificates only when it starts within 120 days of expiry. Plan a k3s restart window (docs/runbooks).
- exp_labels: {severity: critical, certificate: kubelet}
exp_annotations:
summary: "k3s kubelet serving certificate expires in less than 14 days"
description: Restart k3s (it renews certificates within 120 days of expiry) or run `k3s certificate rotate`.
- eval_time: 2h
alertname: PlatformK3sCertificateCheckStale
exp_alerts: []
EOF
docker run --rm --user "$(id -u):$(id -g)" -v "$work:/work:ro" -w /work --entrypoint /bin/promtool "$PROMETHEUS_IMAGE" check rules rules.yaml
docker run --rm --user "$(id -u):$(id -g)" -v "$work:/work:ro" -w /work --entrypoint /bin/promtool "$PROMETHEUS_IMAGE" test rules tests.yaml
echo "PLATFORM HOST CHECK ALERTS PASS"
@@ -2,6 +2,9 @@ namespace: gitea
replicaCount: 1
# services/priority-classes: evicted after everything that is not critical.
priorityClassName: platform-critical
image:
registry: docker.gitea.com
repository: gitea
+9
View File
@@ -9,6 +9,15 @@ metadata:
spec:
instances: 1
# The 26.7.0 server image running on 2026-11-12, by digest. With spec.image set
# the operator assumes a pre-built (optimized) image; this is the stock image,
# so startOptimized stays false and Keycloak builds at start as before.
image: quay.io/keycloak/keycloak@sha256:0f198be292568439d700cdbfb893e69a6009bb43a94a06a945b1d3d506c76b13
startOptimized: false
scheduling:
priorityClassName: platform-critical
db:
vendor: postgres
host: platform-postgres-rw.platform-data.svc.cluster.local
@@ -34,6 +34,11 @@ kubeProxy:
defaultRules:
disabled:
Watchdog: true
# k3s kubelet exposes no kubelet_certificate_manager_* metrics, so these two
# can never fire. Serving certificates are watched by platform-host-checks
# (PlatformK3sServingCertificateExpiry) instead. Review 2026-11-11 A11.
KubeletClientCertificateExpiration: true
KubeletServerCertificateExpiration: true
rules:
etcd: false
kubeControllerManager: false
@@ -80,6 +80,8 @@ loki:
max_compaction_parallelism: 1
singleBinary:
# services/priority-classes: logs can be lost first under node pressure.
priorityClassName: platform-low
replicas: 1
sidecar: false
podLabels:
@@ -4,6 +4,9 @@ global:
replicas: 1
# services/priority-classes: traces can be lost first under node pressure.
priorityClassName: platform-low
config: |
stream_over_http_enabled: false
memberlist:
+86
View File
@@ -0,0 +1,86 @@
# 호스트 백업과 점검
desktop 호스트에서 systemd 타이머로 도는 백업과 점검이다. 결과는 node-exporter textfile 수집기
(`/var/lib/node_exporter/textfile_collector/*.prom`)로 Prometheus에 올라가고, 이 폴더의 PrometheusRule이
경보를 낸다. 근거와 결정은 `docs/reviews/2026-11-11-k3s-maturity-review.md` A1·A11·A13·A14, 8절 결정 1·6이다.
## 무엇을 어디에 얼마나
| 대상 | 방법 | 주기 | 보관 | 위치 |
| --- | --- | --- | --- | --- |
| k3s datastore + `server/token` | `sqlite3 .backup` → tar.gz + sha256 | 매일 03:30 KST | AIStor 30일 + 로컬 최근 7개 | AIStor `platform-host-backups/k3s-datastore/`와 `/srv/k3s/aistor/host-backups/k3s-datastore/`(root 0700). 둘 다 HDD sdb3 |
| Gitea (저장소·패키지·첨부·Actions·`app.ini`) | 라이브 파드에서 `gitea dump --skip-db` | 일요일 04:00 KST | 56일(8주) | AIStor `platform-host-backups/gitea/` |
| platform-postgres (Gitea·Keycloak·tech-log DB) | CloudNativePG base backup + WAL | 매일 03:00 KST | 30일 | AIStor `platform-postgres-backups/` (W1, `services/platform-postgres-backup`) |
| AIStor 데이터(Loki·Tempo) | **백업하지 않는다** | — | — | 8절 결정 6: 잃어도 되는 관측 데이터. 디스크를 잃으면 로그·추적 기록만 사라진다 |
| Prometheus·Grafana·Alertmanager PV | 백업하지 않는다 | — | — | 설정은 이 저장소가 정본이고, 지표는 다시 쌓인다 |
- Gitea DB는 dump에 없다(`--skip-db`). Gitea를 복원하려면 CNPG 복원이 먼저다(`docs/runbooks/gitea-restore.md`).
- k3s 백업에는 `token`이 들어 있다. token과 datastore가 같이 있으면 클러스터의 모든 Secret을 풀 수 있다.
Gitea dump의 `app.ini`에도 비밀 값이 있다. 그래서 둘 다 AIStor의 전용 사용자만 읽고 쓴다.
- AIStor는 같은 desktop의 HDD다. SSD가 죽으면 남지만 desktop 자체를 잃으면 같이 잃는다.
- k3s 백업을 AIStor 밖에도 두는 이유: AIStor는 바로 그 k3s 위에서 돈다. 클러스터를 통째로 잃으면 AIStor를
다시 띄우기 전에는 AIStor 안의 사본을 꺼낼 수 없다. 로컬 사본은 sdb3를 mount하기만 하면 읽힌다
(`docs/runbooks/k3s-datastore-restore.md`). 폴더는 AIStor 드라이브 폴더(`data/`) 옆이라 bucket으로 보이지 않는다.
## off-host 사본 (준비만, 기본 꺼짐)
위치는 소유자가 정한다(8절 결정 1). `/etc/platform-backup/aistor.env`에 두 줄을 넣으면 k3s datastore
백업이 그 공개 키로 암호화한 사본을 `rclone`으로 보낸다.
```bash
OFFHOST_GPG_RECIPIENT_FILE=/etc/platform-backup/offhost-recipient.asc # 공개 키. 개인 키는 desktop에 두지 않는다
OFFHOST_RCLONE_TARGET=<rclone remote>:<path> # rclone 설치·설정 필요
```
지표 `platform_backup_offhost_copy{backup="k3s-datastore"}`가 1이 된다. Gitea dump의 off-host 사본은 아직 없다.
## 점검
| 타이머 | 사용자 | 하는 일 | 지표 |
| --- | --- | --- | --- |
| `platform-k3s-cert-expiry` | donghyeon | 6443·10250 serving 인증서 notAfter | `platform_k3s_certificate_not_after_seconds` |
| `platform-live-drift-check` | donghyeon | origin/main 렌더 → `kubectl diff`(서버 dry-run) | `platform_live_drift_objects` |
## 경보 (`prometheus-rules.yaml`)
| 경보 | 조건 |
| --- | --- |
| `PlatformK3sDatastoreBackupTooOld` | 마지막 성공이 2일보다 오래됨 |
| `PlatformGiteaBackupTooOld` | 8일보다 오래됨 |
| `PlatformHostBackupFailed` | 마지막 실행 실패 |
| `PlatformHostBackupMetricsAbsent` | 지표가 하루 동안 없음(타이머나 textfile 수집기가 안 돎) |
| `PlatformLiveDriftDetected` | drift가 6시간 넘게 남음 |
| `PlatformLiveDriftCheckStale` | drift 검사가 2일 넘게 완료되지 않음 |
| `PlatformK3sServingCertificateExpiry` | 100일 미만 warning, 14일 미만 critical |
| `PlatformK3sCertificateCheckStale` | 인증서 점검이 2일 넘게 안 돎 |
`scripts/validate/test-platform-host-checks-alerts.sh`가 promtool 단위 시험으로 각 경보가 울려야 할 때 울리고
아닐 때 안 울리는지 본다. 같은 PR에서 kube-prometheus-stack의 `KubeletClientCertificateExpiration`·
`KubeletServerCertificateExpiration`을 껐다. k3s kubelet은 그 지표를 내지 않아 울릴 수 없었다.
## 적용 (소유자, 순서대로)
전제: node-exporter textfile 수집기(W1, `services/observability/node-exporter/values/home.yaml`의
`--collector.textfile.directory=/host/root/var/lib/node_exporter/textfile_collector`)가 들어간 main checkout.
```bash
cd /home/donghyeon/workspace/platform && git pull --ff-only
sudo bash scripts/bootstrap/create-host-backup-credentials.sh # bucket·정책·사용자·/etc/platform-backup/aistor.env (APPLY 입력)
bash scripts/bootstrap/install-host-checks.sh --check
sudo bash scripts/bootstrap/install-host-checks.sh --execute # 타이머 설치·켜기·첫 실행
kubectl apply -k services/platform-host-checks # 경보 규칙
sudo bash scripts/validate/k3s-datastore-restore-drill.sh # 첫 백업으로 복원 리허설
sudo bash scripts/validate/gitea-backup-restore-drill.sh
```
`install-host-checks.sh --execute` 마지막에 `platform_backup_last_run_success{backup="k3s-datastore"} 1`과
`{backup="gitea"} 1`이 보여야 한다. 리허설 둘은 각각 `K3S DATASTORE RESTORE DRILL PASS`, `GITEA RESTORE DRILL PASS`로 끝나야 한다.
## 복원 리허설
- k3s: 백업을 내려받아 sha256을 맞춘 뒤, 운영과 같은 버전의 k3s를 **네트워크 없는** 일회용 컨테이너로
띄워 API가 뜨는지, Secret 수가 라이브와 비슷한지(±5), 모든 Secret이 복호화되는지 본다.
`scripts/validate/test-k3s-datastore-restore-drill.sh`는 운영 데이터 없이 같은 경로를 시험한다(새 k3s에
시험 Secret 셋 → 암호화돼 저장된 것을 확인 → 복원).
- Gitea: dump를 풀어 `app.ini`(INTERNAL_TOKEN·SECRET_KEY), 모든 저장소의 `git fsck`, 라이브와 저장소 수를 본다.
- 분기마다 한 번 돌린다. 성공하면 `platform_backup_restore_drill_last_success_timestamp_seconds`가 갱신된다.
@@ -0,0 +1,15 @@
{
"Version": "2012-10-17",
"Statement": [
{
"Effect": "Allow",
"Action": ["s3:GetBucketLocation", "s3:ListBucket"],
"Resource": ["arn:aws:s3:::platform-host-backups"]
},
{
"Effect": "Allow",
"Action": ["s3:GetObject", "s3:PutObject", "s3:DeleteObject"],
"Resource": ["arn:aws:s3:::platform-host-backups/*"]
}
]
}
@@ -0,0 +1,5 @@
apiVersion: kustomize.config.k8s.io/v1beta1
kind: Kustomization
resources:
- prometheus-rules.yaml
@@ -0,0 +1,88 @@
apiVersion: monitoring.coreos.com/v1
kind: PrometheusRule
metadata:
name: platform-host-checks
namespace: observability
labels:
observability.hyeonworks.com/instance: home
app.kubernetes.io/part-of: platform
spec:
groups:
- name: platform-host-backups
rules:
- alert: PlatformK3sDatastoreBackupTooOld
expr: time() - max(platform_backup_last_success_timestamp_seconds{backup="k3s-datastore"}) > 2 * 86400
for: 1h
labels:
severity: warning
annotations:
summary: k3s datastore has not been backed up for more than 2 days
description: Check `journalctl -u platform-k3s-datastore-backup` on desktop. Daily timer, 03:30 KST.
- alert: PlatformGiteaBackupTooOld
expr: time() - max(platform_backup_last_success_timestamp_seconds{backup="gitea"}) > 8 * 86400
for: 1h
labels:
severity: warning
annotations:
summary: Gitea has not been backed up for more than 8 days
description: Check `journalctl -u platform-gitea-backup` on desktop. Weekly timer, Sunday 04:00 KST.
- alert: PlatformHostBackupFailed
expr: max by (backup) (platform_backup_last_run_success) == 0
for: 10m
labels:
severity: warning
annotations:
summary: "The last {{ $labels.backup }} backup run failed"
description: "Check `journalctl -u platform-{{ $labels.backup }}-backup` on desktop."
- alert: PlatformHostBackupMetricsAbsent
expr: absent(platform_backup_last_run_timestamp_seconds{backup="k3s-datastore"}) or absent(platform_backup_last_run_timestamp_seconds{backup="gitea"})
for: 1d
labels:
severity: warning
annotations:
summary: Host backup metrics are missing
description: The backup timers or the node-exporter textfile collector (/var/lib/node_exporter/textfile_collector) are not running.
- name: platform-host-drift
rules:
- alert: PlatformLiveDriftDetected
expr: max(platform_live_drift_objects) > 0
for: 6h
labels:
severity: warning
annotations:
summary: "{{ $value }} live objects differ from platform-core origin/main"
description: Run `scripts/validate/live-drift-check.sh --accepted scripts/validate/live-drift-accepted.txt` on desktop to list them.
- alert: PlatformLiveDriftCheckStale
expr: time() - max(platform_live_drift_last_run_timestamp_seconds) > 2 * 86400 or max(platform_live_drift_check_success) == 0
for: 1h
labels:
severity: warning
annotations:
summary: The daily live drift check has not completed
description: Check `journalctl -u platform-live-drift-check` on desktop.
- name: platform-k3s-certificates
rules:
- alert: PlatformK3sServingCertificateExpiry
expr: (min by (certificate) (platform_k3s_certificate_not_after_seconds) - time()) < 100 * 86400
for: 1h
labels:
severity: warning
annotations:
summary: "k3s {{ $labels.certificate }} serving certificate expires in less than 100 days"
description: k3s renews certificates only when it starts within 120 days of expiry. Plan a k3s restart window (docs/runbooks).
- alert: PlatformK3sServingCertificateExpiry
expr: (min by (certificate) (platform_k3s_certificate_not_after_seconds) - time()) < 14 * 86400
for: 10m
labels:
severity: critical
annotations:
summary: "k3s {{ $labels.certificate }} serving certificate expires in less than 14 days"
description: Restart k3s (it renews certificates within 120 days of expiry) or run `k3s certificate rotate`.
- alert: PlatformK3sCertificateCheckStale
expr: time() - max(platform_k3s_certificate_check_timestamp_seconds) > 2 * 86400
for: 1h
labels:
severity: warning
annotations:
summary: The daily k3s certificate expiry check has not run
description: Check `journalctl -u platform-k3s-cert-expiry` on desktop.
+4 -3
View File
@@ -6,13 +6,14 @@
| 이름 | 값 | 선점 | 붙일 곳 |
| --- | --- | --- | --- |
| `platform-critical` | 1000000 | 낮은 것을 선점 | platform-postgres, Gitea, Keycloak, tech-log |
| `platform-low` | -1000 | 선점 안 함 | Grafana, pgAdmin, Loki, Tempo |
| `platform-critical` | 1000000 | 낮은 것을 선점 | Gitea(`services/gitea` values), Keycloak(CR `spec.scheduling`). platform-postgres는 W1, tech-log는 tech-log `deploy/` |
| `platform-low` | -1000 | 선점 안 함 | Loki(`singleBinary`), Tempo. Grafana·pgAdmin은 chart에 최상위 키가 없어 아직 안 붙였다 |
- 아무것도 안 붙인 파드는 기본값 0이다. 시스템 클래스(`system-*`, 20억)는 그대로 가장 높다.
- `globalDefault`는 둘 다 끈다. 새 파드가 모르게 우선순위를 얻지 않게 하려는 것이다.
- 클래스를 만드는 것은 아무 파드도 바꾸지 않는다. **붙이는 것은 파드 재시작을 부르므로**
k3s 재시작 창에 한다(`infrastructure/host/k3s/README.md`).
k3s 재시작 창에 한다(`docs/runbooks/k3s-restart-window.md`).
- **클래스를 먼저 만든다.** 없는 클래스를 가리키는 파드는 생성이 거절돼 Deployment가 새 파드를 못 만든다.
```sh
kubectl apply -k services/priority-classes