hermes: reclaim healthy worker capacity

infrastructure/core/node-prefer-noschedule-cronjob.yaml

@@ -82,9 +82,8 @@ spec:
                   fi
 
                   if k get node titan-21 >/dev/null 2>&1; then
-                    # Keep replicas off the Jetson, but allow Longhorn to attach
-                    # stateful fallback workloads there.
-                    k label node titan-21 longhorn-host=true --overwrite=true || true
+                    # The Tegra iSCSI initiator cannot host Longhorn engines.
+                    k label node titan-21 longhorn-host- || true
                   else
                     echo "skipping missing node titan-21"
                   fi

infrastructure/vault-injector/helmrelease.yaml

@@ -42,6 +42,8 @@ spec:
       agentImage:
         repository: hashicorp/vault
         tag: "1.17.6"
+      agentDefaults:
+        cpuRequest: 25m
       webhook:
         failurePolicy: Ignore
       nodeSelector:

services/hermes/agent-deployment.yaml

@@ -24,7 +24,7 @@ spec:
         ai.bstein.dev/router-wire-contract: ollama-numeric-keepalive
         ai.bstein.dev/execution: Hermes Kanban with durable direct Codex and Claude Code CLI workers
         ai.bstein.dev/model-policy: Jetson-assisted AUTO routing, low through xhigh, cross-provider fallback
-        ai.bstein.dev/placement: rpi5 preferred; rpi4 next; titan-21 is the stateful fallback
+        ai.bstein.dev/placement: rpi5 preferred; Jetson deferred until state storage is available
         ai.bstein.dev/config-rev: "20260818-ai-usage-exporter-split-v3"
         prometheus.io/scrape: "true"
         prometheus.io/path: /metrics
@@ -107,18 +107,6 @@ spec:
                   - key: kubernetes.io/hostname
                     operator: NotIn
                     values: [titan-04, titan-08, titan-13, titan-14, titan-17, titan-18, titan-19]
-              - matchExpressions:
-                  - key: kubernetes.io/arch
-                    operator: In
-                    values: [arm64]
-                  - key: node-role.kubernetes.io/accelerator
-                    operator: Exists
-                  - key: kubernetes.io/hostname
-                    operator: In
-                    values: [titan-21]
-                  - key: longhorn-host
-                    operator: In
-                    values: ["true"]
           preferredDuringSchedulingIgnoredDuringExecution:
             - weight: 100
               preference:
@@ -126,18 +114,6 @@ spec:
                   - key: hardware
                     operator: In
                     values: [rpi5]
-            - weight: 50
-              preference:
-                matchExpressions:
-                  - key: hardware
-                    operator: In
-                    values: [rpi4]
-            - weight: 25
-              preference:
-                matchExpressions:
-                  - key: kubernetes.io/hostname
-                    operator: In
-                    values: [titan-21]
       initContainers:
         - name: init-config
           image: busybox:1.37
@@ -711,7 +687,7 @@ spec:
             seccompProfile:
               type: RuntimeDefault
           resources:
-            requests: {cpu: 200m, memory: 512Mi}
+            requests: {cpu: 150m, memory: 384Mi}
             limits: {cpu: "3", memory: 6Gi}
         - name: oauth2-proxy
           image: quay.io/oauth2-proxy/oauth2-proxy:v7.15.3@sha256:10a1165743a192e1940b4708fb9647027185ce11a681a1c5519b442ff7f1f561
@@ -881,7 +857,7 @@ spec:
             seccompProfile:
               type: RuntimeDefault
           resources:
-            requests: {cpu: 25m, memory: 64Mi}
+            requests: {cpu: 25m, memory: 48Mi}
             limits: {cpu: 500m, memory: 512Mi}
         - name: cli-lane-runner
           image: registry.bstein.dev/bstein/hermes-agent@sha256:81970563e542f0720773e72297810b3a844b83e381e278f25c0916c78d930107
@@ -939,7 +915,7 @@ spec:
             timeoutSeconds: 5
             failureThreshold: 3
           resources:
-            requests: {cpu: 50m, memory: 192Mi}
+            requests: {cpu: 25m, memory: 128Mi}
             limits: {cpu: "2", memory: 6Gi}
         - name: model-steward
           image: registry.bstein.dev/bstein/hermes-agent@sha256:81970563e542f0720773e72297810b3a844b83e381e278f25c0916c78d930107
@@ -968,7 +944,7 @@ spec:
             - {name: auth-patch, mountPath: /opt/hermes/hermes_cli/auth.py, subPath: auth.py}
             - {name: routing-catalog, mountPath: /routing-catalog}
           resources:
-            requests: {cpu: 25m, memory: 64Mi}
+            requests: {cpu: 25m, memory: 48Mi}
             limits: {cpu: 250m, memory: 512Mi}
         - name: kanban-supervisor
           image: registry.bstein.dev/bstein/hermes-agent@sha256:81970563e542f0720773e72297810b3a844b83e381e278f25c0916c78d930107
@@ -999,7 +975,7 @@ spec:
             - {name: coordinator, mountPath: /opt/coordinator, readOnly: true}
             - {name: auth-patch, mountPath: /opt/hermes/hermes_cli/auth.py, subPath: auth.py}
           resources:
-            requests: {cpu: 10m, memory: 64Mi}
+            requests: {cpu: 10m, memory: 48Mi}
             limits: {cpu: 250m, memory: 256Mi}
         - name: credential-sync
           image: registry.bstein.dev/bstein/hermes-agent@sha256:81970563e542f0720773e72297810b3a844b83e381e278f25c0916c78d930107

services/hermes/execution-coordinator-patch.yaml

@@ -59,7 +59,7 @@ spec:
             initialDelaySeconds: 30
             periodSeconds: 30
           resources:
-            requests: {cpu: 50m, memory: 128Mi}
+            requests: {cpu: 50m, memory: 96Mi}
             limits: {cpu: 500m, memory: 512Mi}
       volumes:
         - name: execution-pool-code

testing/tests/test_hermes_agent_layout.py

@@ -134,13 +134,13 @@ def test_cli_lane_reserves_cpu_headroom_for_ui_and_auth():
 
     assert environment["HERMES_CLI_LANE_CONCURRENCY"] == "2"
     assert lane["resources"] == {
-        "requests": {"cpu": "50m", "memory": "192Mi"},
+        "requests": {"cpu": "25m", "memory": "128Mi"},
         "limits": {"cpu": "2", "memory": "6Gi"},
     }
 
 
-def test_agent_avoids_unhealthy_nodes_and_has_a_stateful_fallback():
-    """The owner agent keeps a bounded fallback when Pi workers are full."""
+def test_agent_avoids_unhealthy_nodes_and_stays_on_storage_workers():
+    """The owner agent stays off unhealthy and attach-incompatible nodes."""
     pod = _agent_deployment()["spec"]["template"]["spec"]
     terms = pod["affinity"]["nodeAffinity"][
         "requiredDuringSchedulingIgnoredDuringExecution"
@@ -154,24 +154,24 @@ def test_agent_avoids_unhealthy_nodes_and_has_a_stateful_fallback():
     assert hostnames["operator"] == "NotIn"
     assert set(hostnames["values"]) >= {"titan-04", "titan-19"}
 
-    fallback = {item["key"]: item for item in terms[1]["matchExpressions"]}
-    assert fallback["node-role.kubernetes.io/accelerator"]["operator"] == "Exists"
-    assert fallback["kubernetes.io/hostname"]["values"] == ["titan-21"]
-    assert fallback["longhorn-host"]["values"] == ["true"]
+    assert len(terms) == 1
 
     preferences = pod["affinity"]["nodeAffinity"][
         "preferredDuringSchedulingIgnoredDuringExecution"
     ]
-    assert [item["weight"] for item in preferences] == [100, 50, 25]
+    assert [item["weight"] for item in preferences] == [100]
 
     hermes = next(item for item in pod["containers"] if item["name"] == "hermes")
-    assert hermes["resources"]["requests"]["cpu"] == "200m"
+    assert hermes["resources"]["requests"] == {
+        "cpu": "150m",
+        "memory": "384Mi",
+    }
 
     node_labels = (
         Path(__file__).parents[2]
         / "infrastructure/core/node-prefer-noschedule-cronjob.yaml"
     ).read_text()
-    assert "k label node titan-21 longhorn-host=true" in node_labels
+    assert "k label node titan-21 longhorn-host-" in node_labels
 
 
 def test_agent_root_is_stock_dashboard_and_terminal_is_a_separate_path():