feat(multiplayer): spread allocator replicas

deploy/k8s/base/allocator-deployment.yaml

@@ -25,6 +25,22 @@ spec:
       terminationGracePeriodSeconds: 10
       serviceAccountName: allocator
       automountServiceAccountToken: false
+      topologySpreadConstraints:
+        - maxSkew: 1
+          topologyKey: topology.kubernetes.io/zone
+          whenUnsatisfiable: ScheduleAnyway
+          labelSelector:
+            matchLabels:
+              app.kubernetes.io/name: allocator
+      affinity:
+        podAntiAffinity:
+          preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              podAffinityTerm:
+                topologyKey: kubernetes.io/hostname
+                labelSelector:
+                  matchLabels:
+                    app.kubernetes.io/name: allocator
       securityContext:
         runAsNonRoot: true
         runAsUser: 10001

multiplayer-next.md

@@ -1251,7 +1251,7 @@ the local/CI/community transport, not a silent production fallback.
 | 8.49 `[D:8.25,8.26,8.28,8.29,8.30,8.31,8.35,8.36]` | **IN PROGRESS.** `scripts/verify_kind_agones.sh` creates a disposable kind cluster, installs pinned Agones, loads the real `game-server` image, applies the Fleet in an explicitly separate Agones-only supervisor/UDP readiness mode, and verifies readiness plus allocation of a dynamic UDP endpoint; `.github/workflows/agones-integration.yml` runs it for infrastructure changes and on demand | The cloud-free runner is committed and fails clearly when Docker/kind/Helm are unavailable. CI/live evidence for production control-plane registration, roster/no-show, both readiness stages, races, multi-match node, result-pending reconciliation, drain, and rollback remains open |
 | 8.50 `[D:8.25,8.37,8.43,8.49]` | **IN PROGRESS.** `make verify-chaos-recovery` provides a disposable PostgreSQL + real testkit API + real maintenance flow: it restarts the API, injects a stale allocation, and verifies no-penalty requeue plus a durable participant-targeted lifecycle event | The API-restart/stalled-allocation slice is implemented and documented; 100 ms RTT/jitter/loss, matcher/client restart, game-pod death, node drain, Redis failover, control-plane loss, and live chaos evidence remain |
 | 8.51 `[D:8.17,8.18,8.30,8.31,8.45]` | **IN PROGRESS.** The opt-in `make verify-multiplayer-load` gate drives 10,000 real HTTP queue-create requests through the service with 256 in flight and records p95/p99, plus 100 concurrent proposal formations through the real matcher/domain path; the handler and in-process ownership boundary are exercised without weakening normal tests | Local API load passes at p95 <250 ms in normal and race runs, and the matcher forms 100 unique proposals; PostgreSQL saturation, durable matcher fencing under load, forecast launch concurrency x2, and replica scaling remain live infrastructure gates |
-| 8.52 `[D:8.32,8.34,8.45,8.51]` | **IN PROGRESS.** Allocator supports both an opt-in per-replica fixed-window quota (`--allocation-quota` / `--allocation-quota-window`) and an optional PostgreSQL-backed EU/NA quota table consumed inside the serializable allocation transaction before any provider call; idempotent replays do not double-count. The allocator now exposes bounded EU/NA Prometheus counters at an explicit `/metrics` listener (`--metrics-addr`), including quota denials; the checked-in rule warns on regional denial activity, and hardened Kubernetes Deployment/Service/ServiceMonitor/PDB resources provide the provisioning, health, rollout, disruption, and discovery contract | Normal/race/vet tests cover local quota exhaustion, window reset, region isolation, invalid input, atomic concurrent consumption, bounded metrics labels, read-only endpoint behavior, and hardened deployment/network policy/lifecycle/PDB invariants; migration/SQL and manifest coverage define the shared quota/metrics boundaries; real image digest/secrets, measured regional cost model, threshold tuning, and denial-of-wallet rehearsal remain |
+| 8.52 `[D:8.32,8.34,8.45,8.51]` | **IN PROGRESS.** Allocator supports both an opt-in per-replica fixed-window quota (`--allocation-quota` / `--allocation-quota-window`) and an optional PostgreSQL-backed EU/NA quota table consumed inside the serializable allocation transaction before any provider call; idempotent replays do not double-count. The allocator now exposes bounded EU/NA Prometheus counters at an explicit `/metrics` listener (`--metrics-addr`), including quota denials; the checked-in rule warns on regional denial activity, and hardened Kubernetes Deployment/Service/ServiceMonitor/PDB/placement resources provide the provisioning, health, rollout, disruption, discovery, and failure-domain-spreading contract | Normal/race/vet tests cover local quota exhaustion, window reset, region isolation, invalid input, atomic concurrent consumption, bounded metrics labels, read-only endpoint behavior, and hardened deployment/network policy/lifecycle/PDB/placement invariants; migration/SQL and manifest coverage define the shared quota/metrics boundaries; real image digest/secrets, measured regional cost model, threshold tuning, and denial-of-wallet rehearsal remain |
 | 8.53 `[D:7.8,8.13,8.38,8.45,8.46,8.48,8.49,8.50,8.51,8.52]` | **IN PROGRESS.** `scripts/verify_release_gate.py` provides a fail-closed promotion check for the ordered development → internal → casual canary → casual → provisional ranked → ranked stages, requiring an evidence report for SLO, security, cost, rollback, EU+NA playtests, and both legacy gates | Validator and adversarial tests cover skipped stages, unknown stages, missing gates, non-boolean gate values, and blank release IDs; the actual reports, production rollback rehearsal, regional playtests, and live promotion remain open |
 
 Implementation invariants for every task above:

server/security/test_kubernetes_policies.py

@@ -48,6 +48,18 @@ class KubernetesPolicyTest(unittest.TestCase):
         ):
             self.assertIn(required, deployment)
 
+    def test_allocator_replicas_prefer_separate_failure_domains(self):
+        deployment = self.read("allocator-deployment.yaml")
+        for required in (
+            "topologySpreadConstraints:", "maxSkew: 1",
+            "topologyKey: topology.kubernetes.io/zone",
+            "whenUnsatisfiable: ScheduleAnyway",
+            "podAntiAffinity:", "preferredDuringSchedulingIgnoredDuringExecution:",
+            "topologyKey: kubernetes.io/hostname",
+        ):
+            self.assertIn(required, deployment)
+        self.assertGreaterEqual(deployment.count("app.kubernetes.io/name: allocator"), 4)
+
     def test_allocator_network_policy_has_only_metrics_data_agones_and_dns_flows(self):
         policies = self.read("network-policies.yaml")
         allocator = policies.split("name: allocator-allowed-flows", 1)[-1]