fix: in-cluster Host, job-execution warning, assistant image tag

CHANGELOG.md

@@ -31,6 +31,14 @@ and the chart uses SemVer (`version`) plus the upstream Forail CalVer
   than replacing the loopback entries.
 - **Chart render checks in CI** pass a throwaway `secrets.forailAdminPassword`,
   so the now-required password does not fail `helm lint` / `helm template`.
+- **In-cluster callers are no longer rejected by `forail.allowedHosts`.** Clients
+  that reach the API through the `forail-web` Service send the Service DNS name
+  as their `Host`, which the hardened list did not cover, so Django answered
+  `400` — the documented forail-operator install
+  (`--set forail.url=http://forail-web.<ns>.svc.cluster.local:8013`) could not
+  resolve a single object. The chart now appends `forail-web`,
+  `forail-web.<ns>`, `forail-web.<ns>.svc` and `forail-web.<ns>.svc.cluster.local`
+  to whatever `forail.allowedHosts` is set to. Unknown hosts are still rejected.
 
 ### Security
 - **No working secret defaults**: `postgresPassword`, `forailSecretKey` and

templates/NOTES.txt

@@ -0,0 +1,39 @@
+Forail {{ .Chart.AppVersion }} installed into namespace {{ .Release.Namespace }}.
+
+{{- if .Values.ingress.enabled }}
+
+  URL:    http{{ if .Values.ingress.tls.enabled }}s{{ end }}://{{ .Values.ingress.host }}
+          Map {{ .Values.ingress.host }} to any node running the ingress controller.
+{{- end }}
+  Admin:  admin
+
+Check the rollout with:
+
+  kubectl -n {{ .Release.Namespace }} get pods
+
+{{- if not .Values.task.privileged }}
+
+────────────────────────────────────────────────────────────────────────
+ WARNING — job execution is NOT enabled in this install.
+────────────────────────────────────────────────────────────────────────
+
+forail-task runs project updates and automation jobs through podman inside
+the task pod, and podman cannot mount its overlay storage in a
+non-privileged container. With task.privileged=false every job fails a
+moment after launch with
+
+  [graphdriver] prior storage driver overlay failed:
+    mount /var/lib/containers/storage/overlay: permission denied
+
+and the only thing visible in the UI is a project or job stuck in "Pending".
+
+This is the safe default on purpose: a privileged pod with a host cgroup
+mount is a trivial container escape to node-root. Enable it only when you
+have somewhere safe to run it -- ideally dedicated, tainted execution
+nodes:
+
+  helm upgrade {{ .Release.Name }} ... \
+    --set task.privileged=true \
+    --set task.hostCgroup=true
+
+{{- end }}

templates/_helpers.tpl

@@ -76,8 +76,18 @@ Wraps DB, Redis, secrets, OTel, admin into one block to avoid drift.
     secretKeyRef:
       name: forail-secrets
       key: forailAdminPassword
+{{- $ns := include "forail.namespace" . }}
 - name: FORAIL_ALLOWED_HOSTS
-  value: {{ .Values.forail.allowedHosts | quote }}
+  {{- /*
+    In-cluster consumers reach the API through the forail-web Service, so their
+    Host header is the Service DNS name, never the ingress host. Django answers
+    400 ("The request could not be understood by the server.") for any Host not
+    listed, which broke the documented forail-operator install
+    (--set forail.url=http://forail-web.<ns>.svc.cluster.local:8013). These four
+    names are cluster-internal and derived from the release, so appending them
+    keeps needtofix M9 tight while making in-cluster access work by default.
+  */}}
+  value: {{ printf "%s,forail-web,forail-web.%s,forail-web.%s.svc,forail-web.%s.svc.cluster.local" .Values.forail.allowedHosts $ns $ns $ns | quote }}
 - name: FORAIL_CSRF_TRUSTED_ORIGINS
   value: {{ .Values.forail.csrfTrustedOrigins | quote }}
 - name: FORAIL_COOKIE_SECURE

values.yaml

@@ -21,7 +21,12 @@ images:
     pullPolicy: IfNotPresent
   assistant:
     repository: ghcr.io/forail-platform/forail-assistant
-    tag: "2026.07.0"
+    # forail-assistant is NOT part of the 2026.07.0 release -- it ships
+    # disabled (assistant.enabled=false below) and was not built or tested for
+    # it, so this stays on the last published build. Pointing it at a tag that
+    # was never pushed would give ImagePullBackOff the moment anyone enabled
+    # it. Bump this in lockstep the next time the assistant is released.
+    tag: "2026.06.0"
     pullPolicy: IfNotPresent
   postgres:
     repository: postgres
@@ -73,6 +78,12 @@ forail:
   # here; 127.0.0.1 and localhost MUST stay in the list because the in-cluster
   # liveness/readiness probes reach the app over the loopback address (Django
   # rejects any Host not listed with HTTP 400, which would crash-loop the pod).
+  # The chart also appends the forail-web Service DNS names (forail-web,
+  # forail-web.<ns>, forail-web.<ns>.svc, forail-web.<ns>.svc.cluster.local) to
+  # whatever you set here, so in-cluster clients that call the Service directly
+  # — forail-operator with
+  # --set forail.url=http://forail-web.<ns>.svc.cluster.local:8013 — are not
+  # rejected. Do not list them here yourself.
   allowedHosts: "forail.lan,127.0.0.1,localhost"
   # NOTE: avoid the .local TLD — Avahi/mDNS (`mdns_minimal [NOTFOUND=return]`
   # in nsswitch.conf on most desktop Linux distros) intercepts every
@@ -242,3 +253,6 @@ ingress:
 #   1) kubectl -n forail exec deploy/forail-web -- forail-manage create_oauth2_token --user admin
 #   2) helm install forail-operator <chart-path> -n forail \
 #        --set forail.token=<PAT> --set forail.url=http://forail-web.forail.svc.cluster.local:8013
+# That Service DNS name is covered by forail.allowedHosts automatically (see
+# above). If you point the operator at the ingress host instead, or front it
+# with another name, pass --set forail.hostHeader=<host in allowedHosts>.