Files
helm-charts/sure/templates/sure.sfs.yaml
T
jklein ee62607e2e
Validate Helm Charts / Detect changed charts (pull_request) Successful in 3s
Validate Helm Charts / Resource quantities (pull_request) Successful in 8s
Validate Helm Charts / Build ${{ fromJSON(needs.detect-changed-charts.outputs.charts) }} (pull_request) Successful in 4s
Validate Helm Charts / Trivy config ${{ fromJSON(needs.detect-changed-charts.outputs.scan-charts) }} (pull_request) Successful in 10s
Validate Helm Charts / Install-test ${{ fromJSON(needs.detect-changed-charts.outputs.install-charts) }} (pull_request) Successful in 2m37s
sure: fix the readiness probe, the ephemeral-storage budget and HOME
Second install-test run, three findings, all measured:

- Readiness on "/" failed with "http: server gave HTTP response to HTTPS
  client" on a pod that was answering correctly. The kubelet's HTTP prober
  follows redirects to the same host, and with RAILS_ASSUME_SSL=true the
  app builds its redirect as https://<pod-ip>:3000/registration/new, which
  Puma does not speak. The probe now runs curl without -L and accepts
  2xx-3xx itself. The same log also proves what the probe is for: against
  an empty database the redirect target was /registration/new, the branch
  User.count decides, so the path really does reach Postgres.
- The pod was evicted within two minutes of every start: "ephemeral local
  storage usage exceeds the total limit of containers 60Mi". emptyDirs and
  the stdout log both count there, and the image rebuilds its Bootsnap
  cache under /rails/tmp on every start because the emptyDir hides the one
  baked into the image. The request is now 256Mi (limit 768Mi) per app
  container - headroom, not a measurement.
- /home/rails is read-only, so Bundler created a throwaway home under
  TMPDIR on every invocation, worker probe included. HOME now points at a
  directory the init container creates.

The CI postgres fixture gets -d sure on its readiness probe, so the server
stops logging a missing database that nothing asked for.
2026-09-22 20:43:53 +00:00

232 lines
12 KiB
YAML

apiVersion: apps/v1
kind: StatefulSet
metadata:
name: {{ .Release.Name }}-{{ .Chart.Name }}-sfs
labels:
app: {{ .Release.Name }}-{{ .Chart.Name }}-sfs
spec:
serviceName: {{ .Release.Name }}-{{ .Chart.Name }}-service
replicas: {{ include "sure.replicas" (dict "scaleDown" .Values.scaleDown "replicas" .Values.sure.replicas) }}
selector:
matchLabels:
app: {{ .Release.Name }}-{{ .Chart.Name }}-sfs
template:
metadata:
name: {{ .Release.Name }}-{{ .Chart.Name }}-sfs
labels:
app: {{ .Release.Name }}-{{ .Chart.Name }}-sfs
annotations:
# The chart has no ConfigMap, so there is no checksum/config here. What
# these annotations cover is the SOURCE of the secrets, not their
# content: the Kubernetes Secrets are materialized by the 1Password
# operator, so Helm never sees what is in them. Repointing one of the
# secrets.*SecretRef values at a different item therefore changes the
# pod template and rolls the pod, instead of silently leaving the old
# credentials in the running container. A change of an item's CONTENT
# inside the vault is not covered - see the comment on secrets in
# values.yaml for why that gap cannot be closed here.
# The name differs from checksum/config on purpose - that one promises
# "the config changed", this one only "the config source changed".
checksum/secret-ref: {{ include (print $.Template.BasePath "/sure.secret.yaml") . | sha256sum }}
checksum/secret-ref-db: {{ include (print $.Template.BasePath "/sure-db.secret.yaml") . | sha256sum }}
checksum/secret-ref-redis: {{ include (print $.Template.BasePath "/sure-redis.secret.yaml") . | sha256sum }}
spec:
automountServiceAccountToken: false
securityContext:
{{- include "sure.podSecurityContext" .Values.sure.securityContext | nindent 8 }}
initContainers:
# Ruby refuses a world-writable directory that has no sticky bit as its
# temporary directory (ruby 3.4, tmpdir.rb: a candidate is skipped when
# "world_writable? && !sticky?"), and a Kubernetes emptyDir is created
# mode 0777 WITHOUT the sticky bit. With /tmp mounted as an emptyDir and
# the image's root filesystem read-only, Ruby then finds no candidate at
# all and both containers die during boot - measured in this chart's
# first install-test, which logged
#
# system temporary path is world-writable: /tmp
# /tmp is world-writable: /tmp
# . is not writable: /rails
# ArgumentError: could not find a temporary directory
#
# raised from MiniMagick, which Active Storage loads while Rails boots.
# Note where it does NOT come from: nothing in the app asked for a temp
# file yet.
#
# The fix is a directory that is writable without being world-writable.
# This init container creates two of them per app container inside the
# same emptyDirs (mode 0700, owned by the pod's uid): "ruby" for TMPDIR,
# and "home" for HOME - the image's /home/rails is read-only, and
# Bundler answers that by creating a throwaway home under TMPDIR on
# every single invocation. It runs as the same non-root uid as the app
# and needs no capability at all: the emptyDir root is 0777, so creating an
# entry in it is allowed, and chmod on a directory one has just created
# needs no privilege. That is deliberately weaker than the chown-as-root
# init container paperless-ngx needs.
#
# The image is the app's own, so there is no second image to pin, no
# Renovate block to maintain for it and nothing extra to pull.
- name: {{ .Release.Name }}-{{ .Chart.Name }}-init-tmp
image: "ghcr.io/we-promise/sure:{{ .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
# command, not args: the image entrypoint would start the application.
command:
- /bin/bash
- -ec
- |
for dir in /web-tmp/ruby /web-tmp/home /worker-tmp/ruby /worker-tmp/home; do
mkdir -p "$dir"
chmod 700 "$dir"
done
securityContext:
allowPrivilegeEscalation: false
readOnlyRootFilesystem: true
runAsNonRoot: true
capabilities:
drop:
- ALL
volumeMounts:
- name: web-tmp
mountPath: /web-tmp
- name: worker-tmp
mountPath: /worker-tmp
# Fixed, deliberately small: this container runs two mkdirs and two
# chmods. Not values-configurable, for the same reason as the init
# container in paperless-ngx - there is nothing here to tune.
resources:
requests:
cpu: 10m
memory: 32Mi
ephemeral-storage: 1Mi
limits:
cpu: 100m
memory: 64Mi
ephemeral-storage: 10Mi
containers:
# The web process. Its command is the image's default
# (CMD ["./bin/rails", "server"]), which is exactly what makes the
# entrypoint run "./bin/rails db:prepare" first - the migration only
# happens for that command, so it happens here and nowhere else.
- name: {{ .Release.Name }}-{{ .Chart.Name }}-web
image: "ghcr.io/we-promise/sure:{{ .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
env:
{{- include "sure.commonEnv" . | nindent 12 }}
{{- include "sure.env" (dict "root" $ "env" .Values.sure.web.env) | nindent 12 }}
ports:
- containerPort: 3000
protocol: TCP
name: http
# Probes, resources and both security contexts go through the
# null-safe helpers (#99): an erased block means "chart defaults
# apply" instead of a vanished probe or an unhardened container.
# Never wrap these in an "if" - a guard hides the erased case
# completely instead of rendering a visible null.
livenessProbe:
{{- include "sure.webLivenessProbe" .Values.sure.web.livenessProbe | nindent 12 }}
readinessProbe:
{{- include "sure.webReadinessProbe" .Values.sure.web.readinessProbe | nindent 12 }}
startupProbe:
{{- include "sure.webStartupProbe" .Values.sure.web.startupProbe | nindent 12 }}
resources:
{{- include "sure.webResources" .Values.sure.web.resources | nindent 12 }}
securityContext:
{{- include "sure.containerSecurityContext" (dict "block" .Values.sure.web.securityContext "path" "sure.web.securityContext") | nindent 12 }}
volumeMounts:
# Active Storage's local disk service, i.e. every uploaded file and
# import, lives here (config/storage.yml: Rails.root.join("storage")).
- name: storage
mountPath: /rails/storage
# Writable paths, so readOnlyRootFilesystem: true can stay on.
# Each container gets its own emptyDirs: nothing has to be shared
# between the two processes, and a shared scratch directory would
# only create coupling nobody asked for.
#
# /tmp: Rack writes multipart uploads (CSV imports, attachments)
# here before Active Storage ever sees them. The app does not use
# this path directly, though - TMPDIR points at the private
# subdirectory the init container creates inside it, because Ruby
# rejects a world-writable emptyDir as a temporary directory.
- name: web-tmp
mountPath: /tmp
# /rails/tmp: Rails' own scratch directory - the Bootsnap cache,
# Puma's pidfile when PIDFILE is set, and the combined CA bundle the
# 00_ssl initializer writes when SSL_CA_FILE is given. Note the
# trade-off, read off the Dockerfile: the image ships a Bootsnap
# cache precompiled at build time under this path, and mounting an
# emptyDir over it hides that cache, so every pod start recompiles
# it. Boot time cost, not correctness; the alternative is a
# writable image layer, which the hardening standard rules out.
- name: web-rails-tmp
mountPath: /rails/tmp
# /rails/log: production logs go to STDOUT (config/environments/
# production.rb), so under normal operation nothing is written here
# at all. It stays writable because Rails' default logger path is
# log/<env>.log and any rake or console path that falls back to it
# would otherwise hit a read-only filesystem. Not measured.
- name: web-rails-log
mountPath: /rails/log
# The Sidekiq process. Same image, different command - and because that
# command is NOT "./bin/rails server", the image entrypoint skips
# db:prepare here, so the two containers never migrate against each
# other.
#
# args, NOT command, and that distinction is load-bearing: Kubernetes'
# "command" replaces the image's ENTRYPOINT, which here is
# /rails/bin/docker-entrypoint - the script that preloads jemalloc.
# Upstream counts jemalloc among the three things that make the
# documented memory figures hold, so replacing the entrypoint would
# silently give this container a different memory profile than the one
# its requests are sized for. "args" replaces only the CMD, which is
# what upstream's compose file does too.
- name: {{ .Release.Name }}-{{ .Chart.Name }}-worker
image: "ghcr.io/we-promise/sure:{{ .Chart.AppVersion }}"
imagePullPolicy: IfNotPresent
args:
- bundle
- exec
- sidekiq
env:
{{- include "sure.commonEnv" . | nindent 12 }}
{{- include "sure.env" (dict "root" $ "env" .Values.sure.worker.env) | nindent 12 }}
livenessProbe:
{{- include "sure.workerLivenessProbe" .Values.sure.worker.livenessProbe | nindent 12 }}
readinessProbe:
{{- include "sure.workerReadinessProbe" .Values.sure.worker.readinessProbe | nindent 12 }}
startupProbe:
{{- include "sure.workerStartupProbe" .Values.sure.worker.startupProbe | nindent 12 }}
resources:
{{- include "sure.workerResources" .Values.sure.worker.resources | nindent 12 }}
securityContext:
{{- include "sure.containerSecurityContext" (dict "block" .Values.sure.worker.securityContext "path" "sure.worker.securityContext") | nindent 12 }}
volumeMounts:
# The worker reads and writes the same Active Storage files as the
# web process (imports are processed as background jobs), which is
# why upstream's compose file mounts one volume into both and why
# both containers live in one pod here: a single ReadWriteOnce PVC
# cannot be attached to two pods.
- name: storage
mountPath: /rails/storage
- name: worker-tmp
mountPath: /tmp
- name: worker-rails-tmp
mountPath: /rails/tmp
- name: worker-rails-log
mountPath: /rails/log
restartPolicy: Always
volumes:
- name: storage
persistentVolumeClaim:
claimName: {{ .Release.Name }}-{{ .Chart.Name }}-pvc
- name: web-tmp
emptyDir: {}
- name: web-rails-tmp
emptyDir: {}
- name: web-rails-log
emptyDir: {}
- name: worker-tmp
emptyDir: {}
- name: worker-rails-tmp
emptyDir: {}
- name: worker-rails-log
emptyDir: {}