GitLab cross-site replication Phase 1a: Postgres network plumbing

Per the approved plan (jiggly-snacking-iverson.md, 'GitLab True
Cross-Site Replication' section) - replacing the git-mirror CronJob
with real CNPG streaming replication for pg-gitlab and pg-praefect,
reusing the exact pattern already proven for pg-authentik.

This commit is network plumbing only, no DB replication yet:
- Repoint both CNPG clusters' barmanObjectStore backup target from
  in-cluster gitlab-minio-svc to VPS MinIO (also becomes the initial
  seed source for the VPS's replicas in Phase 1c).
- New ha-gitlab-postgres/ha-praefect-postgres NodePort Services
  (bypass CNPG's own -rw Services) on both home and VPS.
- New PeerAuthentication with portLevelMtls PERMISSIVE on the replication
  ports - ambient mesh STRICT default would otherwise reset the
  VPS/witness's connection before the Postgres TLS handshake starts.
- Extended allow-gitlab-access AuthorizationPolicy with a port-scoped
  rule for the same traffic (no source.namespaces rule can match
  external, non-mesh peers).
- CoreDNS rewrites for pg-gitlab.ha.huskypup.net / pg-praefect.ha.huskypup.net
  so home's own pods reach the floating hostname locally instead of
  round-tripping through the WAN forward (no NAT hairpin support).

Still needed before Phase 1b/1c: UniFi WAN port-forwards for external
ports 61442/61443 -> talos-cp-01:32442/32443, source-restricted to the
VPS + witness public IPs (manual, same as the existing 61432 rule).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

apps/gitlab/manifests/cnpg-cluster.yaml

@@ -53,17 +53,25 @@ spec:
       max_parallel_workers_per_gather: "2"
       max_parallel_workers: "4"
 
-  # Backup configuration to MinIO
+  # Backup configuration - repointed to VPS MinIO 2026-08-22 (GitLab
+  # cross-site replication plan, see
+  # /home/scooby/.claude/plans/jiggly-snacking-iverson.md's "GitLab True
+  # Cross-Site Replication" section). Was gitlab-minio-svc (in-cluster,
+  # same Ceph the backup is meant to protect against). Also doubles as
+  # the initial-seed source for the VPS's own pg-gitlab replica
+  # (bootstrap.recovery, Phase 1c) - same barmanObjectStore block serves
+  # both purposes, matching the Authentik precedent
+  # (infrastructure/authentik/manifests/cnpg-cluster.yaml).
   backup:
     barmanObjectStore:
-      destinationPath: s3://gitlab-backups/pg-gitlab
-      endpointURL: http://gitlab-minio-svc.gitlab.svc.cluster.local:9000
+      destinationPath: s3://cnpg-backups/pg-gitlab
+      endpointURL: http://100.108.113.41:9000
       s3Credentials:
         accessKeyId:
-          name: gitlab-minio-secret
+          name: vps-minio-secret
           key: accesskey
         secretAccessKey:
-          name: gitlab-minio-secret
+          name: vps-minio-secret
           key: secretkey
       wal:
         compression: gzip

apps/gitlab/manifests/ha-peerauth.yaml

@@ -0,0 +1,51 @@
+# GitLab cross-site replication (see
+# /home/scooby/.claude/plans/jiggly-snacking-iverson.md) - direct copy of
+# infrastructure/authentik/manifests/ha-postgres-peerauth.yaml's proven
+# fix. The gitlab namespace carries the same ambient-mesh STRICT default
+# as authentik; without this, ztunnel resets the VPS/witness's connection
+# (no mesh identity - they're not in this cluster) before the Postgres
+# TLS handshake can even start, surfacing as "server closed the
+# connection unexpectedly" on the replica side despite the NodePort/
+# firewall path being entirely correct.
+#
+# Scoped to just each CNPG primary pod's ports via portLevelMtls, not the
+# whole namespace - GitLab's own in-mesh east-west traffic (webservice
+# <-> gitaly <-> redis <-> sidekiq <-> praefect) stays STRICT.
+apiVersion: security.istio.io/v1
+kind: PeerAuthentication
+metadata:
+  name: allow-ha-gitlab-postgres-replication
+  namespace: gitlab
+spec:
+  selector:
+    matchLabels:
+      cnpg.io/cluster: pg-gitlab
+  mtls:
+    mode: STRICT
+  portLevelMtls:
+    "5432":
+      mode: PERMISSIVE
+    # Floating-hostname port (see ha-postgres-nodeport.yaml) - needs its
+    # own entry, ambient's port-level mTLS enforcement is keyed on the
+    # port actually dialed (61442), not just the pod's real containerPort
+    # (5432) it eventually reaches (confirmed live for the authentik
+    # precedent, same mechanism here).
+    "61442":
+      mode: PERMISSIVE
+---
+apiVersion: security.istio.io/v1
+kind: PeerAuthentication
+metadata:
+  name: allow-ha-praefect-postgres-replication
+  namespace: gitlab
+spec:
+  selector:
+    matchLabels:
+      cnpg.io/cluster: pg-praefect
+  mtls:
+    mode: STRICT
+  portLevelMtls:
+    "5432":
+      mode: PERMISSIVE
+    "61443":
+      mode: PERMISSIVE

apps/gitlab/manifests/ha-postgres-nodeport.yaml

@@ -0,0 +1,56 @@
+---
+# GitLab cross-site replication (see
+# /home/scooby/.claude/plans/jiggly-snacking-iverson.md's "GitLab True
+# Cross-Site Replication" section) - direct copy of the proven pattern
+# from infrastructure/authentik/manifests/ha-postgres-nodeport.yaml.
+# Exposes pg-gitlab's and pg-praefect's current primaries to the
+# VPS/witness over the Netbird mesh for streaming replication - same
+# selectors CNPG's own -rw ClusterIP Services use, just NodePort instead,
+# since CNPG doesn't manage these objects and won't fight with them.
+# Needs infrastructure/kyverno/policies/ha-failover-nodeport-exception.yaml
+# (name pattern ha-* already matches, no change needed there).
+#
+# Second port on each (61442/61443, ClusterIP only) for the CoreDNS
+# rewrite in infrastructure/coredns/manifests/: home's own pods resolving
+# pg-gitlab.ha.huskypup.net / pg-praefect.ha.huskypup.net rewrite to
+# THESE services specifically so they can reach them on the SAME port
+# number the app tier will eventually be configured with (Phase 3) -
+# rewrite only changes the resolved NAME, not the port a client then
+# connects to (confirmed live for the authentik precedent).
+apiVersion: v1
+kind: Service
+metadata:
+  name: ha-gitlab-postgres
+  namespace: gitlab
+spec:
+  type: NodePort
+  selector:
+    cnpg.io/cluster: pg-gitlab
+    cnpg.io/instanceRole: primary
+  ports:
+    - name: nodeport-5432
+      port: 5432
+      targetPort: 5432
+      nodePort: 32442
+    - name: floating-port-61442
+      port: 61442
+      targetPort: 5432
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: ha-praefect-postgres
+  namespace: gitlab
+spec:
+  type: NodePort
+  selector:
+    cnpg.io/cluster: pg-praefect
+    cnpg.io/instanceRole: primary
+  ports:
+    - name: nodeport-5432
+      port: 5432
+      targetPort: 5432
+      nodePort: 32443
+    - name: floating-port-61443
+      port: 61443
+      targetPort: 5432

apps/gitlab/manifests/praefect-cnpg-cluster.yaml

@@ -26,10 +26,53 @@ spec:
 
   primaryUpdateStrategy: unsupervised
 
+  # Explicit params added 2026-08-22 (GitLab cross-site replication plan)
+  # - this cluster ran on bare Postgres defaults before. A VPS replica
+  # can't replay WAL past a checkpoint recorded with higher settings than
+  # its own ("recovery aborted because of insufficient parameter
+  # settings", confirmed live for pg-authentik) - giving both sides an
+  # identical explicit block up front means nobody can tune one side
+  # later and silently break replication.
+  postgresql:
+    parameters:
+      max_connections: "200"
+      shared_buffers: "512MB"
+      effective_cache_size: "1536MB"
+      maintenance_work_mem: "128MB"
+      checkpoint_completion_target: "0.9"
+      wal_buffers: "16MB"
+      default_statistics_target: "100"
+      random_page_cost: "1.1"
+      effective_io_concurrency: "200"
+      work_mem: "4MB"
+      min_wal_size: "1GB"
+      max_wal_size: "4GB"
+      max_worker_processes: "4"
+
   bootstrap:
     initdb:
       database: praefect_production
       owner: app
 
+  # Backup to VPS MinIO, also the initial-seed source for the VPS's own
+  # pg-praefect replica (bootstrap.recovery, Phase 1c) - this cluster had
+  # no backup block at all before. See the matching comment on
+  # cnpg-cluster.yaml (pg-gitlab) in this same directory.
+  backup:
+    barmanObjectStore:
+      destinationPath: s3://cnpg-backups/pg-praefect
+      endpointURL: http://100.108.113.41:9000
+      s3Credentials:
+        accessKeyId:
+          name: vps-minio-secret
+          key: accesskey
+        secretAccessKey:
+          name: vps-minio-secret
+          key: secretkey
+      wal:
+        compression: gzip
+        maxParallel: 2
+    retentionPolicy: "30d"
+
   monitoring:
     enablePodMonitor: true