feat(k8s): make single-instance workloads survive node drains

Draining k8s-3 for the Talos 1.13.10 upgrade stopped on four pods
(wp-pg-1, ha-pg-1, gf-pg-1, gitea-0) whose PDBs allow zero disruptions,
and each needed a manual --disable-eviction.

CNPG: ha-pg, gf-pg, wp-pg and tr-pg go to instances: 2. With a replica,
CNPG switches the primary over ahead of a drain, the same way ak-pg,
im-pg and pprl-pg already do. With one instance it creates a primary PDB
and 'will prevent draining the node where the instance is housed'
(CNPG docs, kubernetes_upgrade). smartShutdownTimeout 10 / stopDelay 60
as on ak-pg, which bound switchover time and the barman-plugin demote
hang (cloudnative-pg#9770). Network policies already allow intra-cluster
replication, operator :8000 and WAL-restore egress in all four
namespaces; nothing references a specific instance name.

Single-replica RWO workloads (gitea, woodpecker-server,
protonmail-bridge, syncthing, influxdb): drop the minAvailable:1 PDBs.
They cannot fail over, and eviction already stops the pod gracefully
with SIGTERM and its grace period; the PDB only forced a manual
--disable-eviction on every drain. influxdb's comes from the chart
(pdb.create, influxdb2 2.1.2 defaults).

The 'single-instance cluster, the archive is the only off-node copy'
comments no longer hold and are removed (im-pg's was already stale).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

kubernetes/apps/gitea/kustomization.yaml

@@ -6,7 +6,6 @@ kind: Kustomization
 # and verified. See MIGRATION-gitea-woodpecker.md.
 resources:
   - statefulset.yaml
-  - pdb.yaml
   - service.yaml
   - ingressroute-ssh.yaml
   - ingress.yaml

kubernetes/apps/gitea/pdb.yaml

@@ -1,23 +0,0 @@
----
-# Eviction protection for the single-replica Gitea StatefulSet.
-#
-# Gitea runs on SQLite (single writer) and its repos + DB live on a single
-# ReadWriteOnce PVC, so it is deliberately replicas:1 and cannot fail over.
-# With no PDB, a node drain / maintenance evicts the pod abruptly, mid-write,
-# with zero coordination. minAvailable:1 on a 1-replica workload makes the
-# eviction API refuse *voluntary* evictions: `kubectl drain` blocks and
-# surfaces this pod, forcing a conscious decision (drain with
-# --disable-eviction, or scale down first) instead of yanking the live DB.
-# It does NOT block involuntary disruptions (node crash, OOM).
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: gitea
-  namespace: gitea
-  labels:
-    app.kubernetes.io/name: gitea
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: gitea

kubernetes/apps/grafana-db/postgres.yaml

@@ -4,7 +4,9 @@ metadata:
   name: gf-pg
   namespace: monitoring
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
   imageName: ghcr.io/cloudnative-pg/postgresql:16
   storage:
     storageClass: ceph-block

kubernetes/apps/home-assistant/ha-pg-cluster.yaml

@@ -30,7 +30,9 @@ metadata:
   name: ha-pg
   namespace: home-assistant
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
   imageName: ghcr.io/cloudnative-pg/postgresql:16
   storage:
     storageClass: ceph-block
@@ -124,8 +126,7 @@ spec:
       # padding rather than data.
       #
       # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
       archive_timeout: "30min"
       # Fewer checkpoints => fewer full-page writes, since a page is only
       # fully-logged on its first modification after a checkpoint.

kubernetes/apps/immich/cluster.yaml

@@ -59,8 +59,7 @@ spec:
       # padding rather than data.
       #
       # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
       archive_timeout: "30min"
       # Write-reduction tuning. These databases are largely idle, and the
       # Postgres defaults favour throughput over storage that wears out.

kubernetes/apps/influxdb/helmrelease.yaml

@@ -61,6 +61,9 @@ spec:
       enabled: true
       storageClass: ceph-block
       size: 50Gi
+    # A minAvailable:1 PDB on a single replica blocks every node drain.
+    pdb:
+      create: false
     # Bootstrap admin/org/bucket on the fresh PVC. Admin password + token come
     # from the influxdb-bootstrap Secret (OpenBao via ESO), deterministic across
     # rebuilds so the bootstrap token is the live admin token. `influx setup`

kubernetes/apps/paperless/kustomization.yaml

@@ -11,7 +11,6 @@ resources:
   - protonmail-bridge.yaml
   - ftp-scanner.yaml
   - syncthing.yaml
-  - pdb.yaml
   - network-policy.yaml
   # pprl-pg CNPG Postgres + data/media/consume/redis PVCs (moved here from the
   # retired seed-paperless KS; each carries prune: disabled so ownership

kubernetes/apps/paperless/pdb.yaml

@@ -1,37 +0,0 @@
----
-# Eviction protection for the single-replica stateful paperless sidecars.
-#
-# protonmail-bridge and syncthing each hold state on a single RWO PVC and run
-# replicas:1. minAvailable:1 makes the eviction API refuse voluntary evictions,
-# so a node drain blocks and surfaces these pods instead of evicting them
-# abruptly. Involuntary disruptions (node crash) are not affected.
-#
-# NOTE: the other single-replica paperless pods (paperless, redis, gotenberg,
-# tika, ftp-scanner) are intentionally omitted — they are caches / stateless
-# converters where a blocking PDB would only wedge node drains with no
-# recoverability benefit.
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: protonmail-bridge
-  namespace: paperless
-  labels:
-    app.kubernetes.io/name: protonmail-bridge
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: protonmail-bridge
----
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: syncthing
-  namespace: paperless
-  labels:
-    app.kubernetes.io/name: syncthing
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: syncthing

kubernetes/apps/trmnl/cluster.yaml

@@ -4,7 +4,7 @@ metadata:
   name: tr-pg
   namespace: trmnl
 spec:
-  instances: 1
+  instances: 2
   smartShutdownTimeout: 10
   stopDelay: 60
   # PG 18, NOT the repo-standard 16. Terminus ships a committed
@@ -47,8 +47,7 @@ spec:
       # padding rather than data.
       #
       # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
       archive_timeout: "30min"
       # Write-reduction tuning. These databases are largely idle, and the
       # Postgres defaults favour throughput over storage that wears out.

kubernetes/apps/woodpecker/cluster.yaml

@@ -22,7 +22,9 @@ metadata:
     # migration deliberately withheld this while a rollback still existed.
     kustomize.toolkit.fluxcd.io/prune: disabled
 spec:
-  instances: 1
+  instances: 2
+  smartShutdownTimeout: 10
+  stopDelay: 60
   imageName: ghcr.io/cloudnative-pg/postgresql:16
   storage:
     storageClass: ceph-block
@@ -91,8 +93,7 @@ spec:
       # padding rather than data.
       #
       # The cost is RPO: up to 30min of WAL is unrecoverable if this cluster and
-      # its volume are lost outright. This is a single-instance cluster, so the
-      # archive is the only off-node copy.
+      # its volumes are lost outright.
       archive_timeout: "30min"
   resources:
     requests:

kubernetes/apps/woodpecker/kustomization.yaml

@@ -3,7 +3,6 @@ kind: Kustomization
 resources:
   - rbac.yaml
   - server.yaml
-  - pdb.yaml
   - agent.yaml
   - service.yaml
   - ingress.yaml
@@ -22,7 +21,7 @@ resources:
   # OpenBao access, no NetworkPolicy needed. server.yaml/agent.yaml keep their
   # Reloader annotations; externalsecret.yaml keeps the 10m refreshInterval.
   # --- namespace migration, stages 1-2 (issue #433) ---
-  # Stage 2 has moved the workloads above (server, agent, service, ingress, pdb,
+  # Stage 2 has moved the workloads above (server, agent, service, ingress,
   # rbac, secrets, policies) into the `woodpecker` namespace and pointed the
   # server's DSN at wp-pg-rw.woodpecker.svc. Stage 3 retired the `apps`
   # originals; these four now describe the surviving cluster. History and the

kubernetes/apps/woodpecker/pdb.yaml

@@ -1,27 +0,0 @@
----
-# Eviction protection for the single-replica Woodpecker server StatefulSet.
-#
-# The server is SQLite-backed (single writer) on a single RWO PVC, so it is
-# deliberately replicas:1. minAvailable:1 makes the eviction API refuse
-# voluntary evictions so a node drain blocks and surfaces this pod rather
-# than yanking the live CI DB mid-write. Involuntary disruptions are not
-# affected.
-#
-# The selector MUST include app.kubernetes.io/component: server: the agent
-# Deployment in this same namespace shares app.kubernetes.io/name: woodpecker
-# (component: agent), and a name-only selector would wrongly also cover the
-# stateless, disposable agents.
-apiVersion: policy/v1
-kind: PodDisruptionBudget
-metadata:
-  name: woodpecker-server
-  namespace: woodpecker
-  labels:
-    app.kubernetes.io/name: woodpecker
-    app.kubernetes.io/component: server
-spec:
-  minAvailable: 1
-  selector:
-    matchLabels:
-      app.kubernetes.io/name: woodpecker
-      app.kubernetes.io/component: server