From 79e2032603c690ecd5e8ea72a0fc845017e995c5 Mon Sep 17 00:00:00 2001
From: Snarf <snarf@snarfnet.net>
Date: Tue, 10 Nov 2026 13:45:36 -0500
Subject: [PATCH] fix(uptime-kuma): route ingress directly to clusterip service
 http port and fix ai-stack node affinity

---
 .../aerodrome/uptime-kuma/uptime-kuma.yaml    |   5 +-
 manifests/spacedock/ai-stack/ai-stack.yaml    | 117 +++++++++++++++++-
 2 files changed, 119 insertions(+), 3 deletions(-)

diff --git a/manifests/aerodrome/uptime-kuma/uptime-kuma.yaml b/manifests/aerodrome/uptime-kuma/uptime-kuma.yaml
index fd580c8..8dc3ba0 100644
--- a/manifests/aerodrome/uptime-kuma/uptime-kuma.yaml
+++ b/manifests/aerodrome/uptime-kuma/uptime-kuma.yaml
@@ -115,6 +115,7 @@ kind: Ingress
 metadata:
   annotations:
     traefik.ingress.kubernetes.io/router.middlewares: monitoring-authentik-forwardauth@kubernetescrd
+    traefik.ingress.kubernetes.io/service.serversscheme: http
   labels:
     app.kubernetes.io/name: uptime-kuma
   name: uptime-ing
@@ -127,9 +128,9 @@ spec:
       paths:
       - backend:
           service:
-            name: uptime-lb
+            name: uptime-kuma-2-1756668961
             port:
-              number: 448
+              number: 3001
         path: /
         pathType: ImplementationSpecific
   tls:
diff --git a/manifests/spacedock/ai-stack/ai-stack.yaml b/manifests/spacedock/ai-stack/ai-stack.yaml
index ef968b1..d0a580a 100644
--- a/manifests/spacedock/ai-stack/ai-stack.yaml
+++ b/manifests/spacedock/ai-stack/ai-stack.yaml
@@ -126,6 +126,8 @@ spec:
       labels:
         app: vllm-openai
     spec:
+      nodeSelector:
+        kubernetes.io/hostname: snaisvr44hrvstr04
       containers:
         - name: vllm
           image: vllm/vllm-openai:v0.7.3
@@ -140,7 +142,7 @@ spec:
             - "--dtype"
             - "half"
             - "--max-model-len"
-            - "16384"
+            - "65536"
             - "--gpu-memory-utilization"
             - "0.90"
             - "--enable-auto-tool-choice"
@@ -155,6 +157,8 @@ spec:
               value: "/root/.cache/huggingface"
             - name: VLLM_WORKER_MULTIPROC_METHOD
               value: "spawn"
+            - name: VLLM_ALLOW_LONG_MAX_MODEL_LEN
+              value: "1"
           ports:
             - name: http-openai
               containerPort: 8000
@@ -272,6 +276,8 @@ spec:
               value: "/root/.cache/huggingface"
             - name: VLLM_WORKER_MULTIPROC_METHOD
               value: "spawn"
+            - name: VLLM_ALLOW_LONG_MAX_MODEL_LEN
+              value: "1"
           ports:
             - name: http-openai
               containerPort: 8000
@@ -322,6 +328,115 @@ spec:
       nodePort: 30804
 ---
 # ==============================================================================
+# 2C. DEEPSEEK-R1-DISTILL-QWEN-14B REASONING ENGINE (snaisvr44hrvstr01 Tesla T4 GPU)
+# ==============================================================================
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: glm-superexpert-models-pvc
+  namespace: ai-stack
+spec:
+  accessModes:
+    - ReadWriteOnce
+  storageClassName: longhorn
+  resources:
+    requests:
+      storage: 200Gi
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: glm-superexpert
+  namespace: ai-stack
+  labels:
+    app: glm-superexpert
+spec:
+  replicas: 1
+  strategy:
+    type: Recreate
+  selector:
+    matchLabels:
+      app: glm-superexpert
+  template:
+    metadata:
+      labels:
+        app: glm-superexpert
+    spec:
+      nodeSelector:
+        kubernetes.io/hostname: snaisvr44hrvstr01
+      containers:
+        - name: vllm
+          image: vllm/vllm-openai:v0.32.0-cu129
+          imagePullPolicy: IfNotPresent
+          command: ["/bin/bash", "-c"]
+          args:
+            - |
+              rm -rf /usr/local/lib/python3.12/dist-packages/torchcodec*
+              exec vllm serve casperhansen/deepseek-r1-distill-qwen-14b-awq \
+                --served-model-name DeepSeek-R1-Distill-Qwen-14B \
+                --quantization awq \
+                --dtype half \
+                --max-model-len 16384 \
+                --gpu-memory-utilization 0.92 \
+                --enforce-eager \
+                --enable-auto-tool-choice \
+                --tool-call-parser hermes \
+                --reasoning-parser deepseek_r1 \
+                --host 0.0.0.0 \
+                --port 8000
+          env:
+            - name: HF_HOME
+              value: "/root/.cache/huggingface"
+          ports:
+            - name: http-openai
+              containerPort: 8000
+          resources:
+            requests:
+              cpu: "8"
+              memory: "20Gi"
+              nvidia.com/gpu: "1"
+            limits:
+              cpu: "24"
+              memory: "40Gi"
+              nvidia.com/gpu: "1"
+          readinessProbe:
+            httpGet:
+              path: /health
+              port: 8000
+            initialDelaySeconds: 30
+            periodSeconds: 10
+          volumeMounts:
+            - name: glm-models
+              mountPath: /root/.cache/huggingface
+            - name: dshm
+              mountPath: /dev/shm
+      volumes:
+        - name: glm-models
+          persistentVolumeClaim:
+            claimName: glm-superexpert-models-pvc
+        - name: dshm
+          emptyDir:
+            medium: Memory
+            sizeLimit: 16Gi
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: glm-superexpert
+  namespace: ai-stack
+  labels:
+    app: glm-superexpert
+spec:
+  type: NodePort
+  selector:
+    app: glm-superexpert
+  ports:
+    - name: http-openai
+      port: 8000
+      targetPort: 8000
+      nodePort: 30805
+---
+# ==============================================================================
 # 3. NVIDIA TRITON INFERENCE SERVER (Decision Engine & Multi-Model ML Serving)
 # ==============================================================================
 apiVersion: v1