chore(monitoring): disable metric and log retention pruning

The cluster is being stopped and restarted later to export camp data, so
nothing may be pruned in the meantime.

Prometheus was not honouring the 90d setting: the size cap was the binding
constraint, with 6 size-based deletions in the last 30 days (0 time-based)
and the oldest sample only 65 days old. Camp data was being deleted from
the oldest end. Disable both time and size retention.

Loki already kept data forever (compactor deletes are off by default and no
retention_period was set). Pin both explicitly so a chart-default change
cannot silently start pruning.

Growth is ~0.8GB/day against ~16GB free (~21 days), so with no automatic
pruning the Prometheus PVC needs watching rather than self-limiting.

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

Author: hakan-persson

k8s/infra-manifest/monitoring/helm-values.yaml

@@ -86,11 +86,11 @@ prometheus:
                   operator: In
                   values:
                     - agentpool
-    retention: 90d
-    # Size cap must stay below the PVC's usable disk (~67GB of the 64Gi request)
-    # with headroom for WAL/head/compaction. Raised from 50GB (which was pruning
-    # oldest data before 90d) so camp+evaluation data is retained by time, not size.
-    retentionSize: 60GB
+    # Retention fully disabled (0 = never prune, by time or size): camp data must
+    # survive until it has been exported. Growth is ~0.8GB/day against ~16GB free,
+    # so the PVC must be watched rather than relying on automatic pruning.
+    retention: 0d
+    retentionSize: "0"
     storageSpec:
       volumeClaimTemplate:
         spec:

k8s/infra-manifest/monitoring/loki-values.yaml

@@ -24,6 +24,11 @@ loki:
   auth_enabled: false
   limits_config:
     allow_structured_metadata: false
+    # 0s = keep forever. Explicit so a chart-default change can't start pruning
+    # camp logs before they are exported.
+    retention_period: 0s
+  compactor:
+    retention_enabled: false
 
 singleBinary:
   replicas: 1