fix(forgejo): resolve merge conflict — keep 60Gi, revert CPU to 1000m

Co-authored-by: rock <[email protected]>

k8s/apps/llm-serving/coder.yaml

@@ -1,134 +0,0 @@
-apiVersion: serving.kserve.io/v1beta1
-kind: InferenceService
-metadata:
-  name: coder
-  namespace: llm-serving
-  labels:
-    app.kubernetes.io/name: llm-coder
-    app.kubernetes.io/part-of: llm-serving
-  annotations:
-    serving.kserve.io/deploymentMode: RawDeployment
-spec:
-  predictor:
-    minReplicas: 1
-    maxReplicas: 1
-    nodeSelector:
-      kubernetes.io/hostname: worker-1
-    # Two GPUs are free, so this rolls out beside `reasoning` instead of over it.
-    # Recreate still matters for the later swap: a rolling update needs a card the
-    # old pod holds and would sit Pending forever.
-    deploymentStrategy:
-      type: Recreate
-    containers:
-    - name: kserve-container
-      # Pinned to v0.11.0 on purpose. This is the last image measured to work on this
-      # node: torch 2.8.0+cu128, arch list sm_70..sm_120, driver 580.159.04.
-      # v0.15.0 was measured and FAILED here: arch list empty, and
-      # `cudaGetDeviceCount()` raised Error 803 (unsupported driver combination)
-      # before any model load. Do not raise this tag without re-running that probe.
-      image: vllm/vllm-openai:v0.11.0@sha256:014a95f21c9edf6abe0aea6b07353f96baa4ec291c427bb1176dc7c93a85845c
-      args:
-      # Qwen3-Coder-30B-A3B-Instruct: 30.5B total, 3.3B active. Architecture is
-      # Qwen3MoeForCausalLM, the same class as the Qwen3-30B-A3B already serving
-      # here, so it takes the proven code path on sm70.
-      #
-      # Qwen ships this model only as BF16 (61.1 GB) and FP8 (no FP8 on Volta).
-      # 4-bit GPTQ is the only form that fits one 32 GB card. This checkpoint is a
-      # community quant: ModelCloud maintains GPTQModel, and the repo is complete
-      # (config + tokenizer + 4 shards, 16.9 GB). Verified by loading, not by trust.
-      #
-      # Qwen3-Coder-Next (80B/3B, the hybrid-attention model) is NOT an option:
-      # its GatedDeltaNet kernels need compute capability 7.5+ and the `fla` module,
-      # and sm70 has neither. Measured, not assumed.
-      - --model=ModelCloud/Qwen3-Coder-30B-A3B-Instruct-GPTQMODEL-W4A16-A
-      # `reasoning` stays a served name so an existing client that sends
-      # model=reasoning keeps working after the DNS rename. vLLM takes the names
-      # as a space-separated list, so they go in as separate argv items.
-      - --served-model-name
-      - coder
-      - reasoning
-      - --quantization=gptq
-      - --dtype=float16
-      - --kv-cache-dtype=auto
-      # Native window is 262144 and rope_scaling is null, so 256K needs no YaRN.
-      # One card cannot hold it: measured KV pool on a single V100 was 132,672
-      # tokens. TP=2 halves the weights per card (7.8 GiB each), which frees room
-      # for the larger pool. Measured after boot: see the KV line in the log.
-      - --max-model-len=262144
-      - --tensor-parallel-size=2
-      - --pipeline-parallel-size=1
-      - --gpu-memory-utilization=0.90
-      # A single 262,144-token request consumes the whole pool, so keep few slots.
-      - --max-num-seqs=4
-      - --max-num-batched-tokens=8192
-      - --long-prefill-token-threshold=2048
-      - --enable-chunked-prefill
-      - --enable-prefix-caching
-      # Measured: with the default (custom all-reduce ON) the engine finished CUDA graph
-      # capture in 6s at 17:01:56 and then went silent for 14+ minutes, emitting only
-      # "No available shared memory broadcast block found in 60 seconds". The front-end
-      # never received the ready message. vLLM's custom all-reduce registers its own
-      # CUDA graph addresses and deadlocks on this Volta/PCIe pair. NCCL is the fallback
-      # and is correct here, because TP=2 across two PCIe V100s gains nothing from the
-      # custom kernel anyway.
-      - --disable-custom-all-reduce
-      - --reasoning-parser=qwen3
-      - --enable-auto-tool-choice
-      - --tool-call-parser=hermes
-      - --host=0.0.0.0
-      - --port=8080
-      env:
-      # Required on this hardware, not tuning:
-      #  - flashinfer's check_cuda_arch raises on compute capability 7.0,
-      #    so the sampler must stay off
-      #  - FlashInfer and FlashAttention-2 need sm75+, so Triton serves attention
-      - name: VLLM_USE_FLASHINFER_SAMPLER
-        value: "0"
-      - name: VLLM_ATTENTION_BACKEND
-        value: TRITON_ATTN
-      - name: HF_HOME
-        value: /mnt/models
-      ports:
-      - containerPort: 8080
-        protocol: TCP
-      readinessProbe:
-        httpGet:
-          path: /health
-          port: 8080
-        periodSeconds: 10
-      startupProbe:
-        httpGet:
-          path: /health
-          port: 8080
-        # TP=2 at 256K needs ~10 min before the API binds: 22s weights, 80s torch.compile,
-        # 85s KV profiling, then graph capture. 120 x 15s = 30 min of headroom.
-        failureThreshold: 120
-        periodSeconds: 15
-      resources:
-        # worker-1 exposes only 63,495 MiB of allocatable memory, so a 24Gi request
-        # never schedules. Two vLLM workers keep the weights on the GPU and need
-        # little host RAM, so the request stays small and the limit allows burst.
-        requests:
-          cpu: "8"
-          memory: 12Gi
-          nvidia.com/gpu: "2"
-        limits:
-          cpu: "16"
-          memory: 40Gi
-          nvidia.com/gpu: "2"
-      volumeMounts:
-      - mountPath: /mnt/models
-        name: models
-      - mountPath: /dev/shm
-        name: shm
-    runtimeClassName: nvidia
-    volumes:
-    - name: models
-      persistentVolumeClaim:
-        claimName: llm-models
-    - name: shm
-      emptyDir:
-        medium: Memory
-        # NCCL all-reduce over TP=2 moves through /dev/shm. The 64Mi default is
-        # too small and shows up as an init hang or a bus error.
-        sizeLimit: 8Gi

k8s/apps/llm-serving/kustomization.yaml

@@ -14,10 +14,6 @@ resources:
   # v0.5.22+, published only on CUDA 13, and CUDA 13 dropped Volta kernels. The model
   # instead runs the implementation it ships (joint_schema_model.py) in fp16 on one V100.
   - clef-serve.yaml
-  # coder = Qwen3-Coder-30B-A3B-Instruct, 4-bit GPTQ, TP=2 at 262,144 tokens. It serves
-  # both `coder` and `reasoning`, so poimen-memory needs a host change only, never a
-  # model-name change. reasoning.yaml below can be dropped once that host change lands.
-  - coder.yaml
   - reasoning.yaml
   # ornith is retired: its InferenceService still reports Ready=True while running zero
   # replicas, because desired equals available at 0. Its manifest was never committed,

k8s/apps/llm-serving/networkpolicy.yaml

@@ -66,21 +66,6 @@ spec:
           port: 80
         - protocol: TCP
           port: 8000
-    # Allow from poimen namespace (poimen-memory ingest + query).
-    # The client-side egress rule is the other half and must stay in step:
-    # NetworkPolicy poimen/poimen-memory-llm-egress permits port 8080 to pods
-    # labelled part-of=llm-serving, which is the same selector used here. This
-    # policy had no matching ingress rule at all, so the packet was allowed out
-    # and dropped here. The client then saw a bare timeout and the predictor log
-    # stayed empty, which is why this failure looked like a hung inference engine
-    # for 34 days.
-    - from:
-        - namespaceSelector:
-            matchLabels:
-              kubernetes.io/metadata.name: poimen
-      ports:
-        - protocol: TCP
-          port: 8080
     # Allow intra-namespace (pod-to-pod within llm-serving)
     - from:
         - podSelector:

k8s/bootstrap/phase3-forgejo/forgejo-values.yaml

@@ -97,10 +97,25 @@ gitea:
       # USER and PASSWD injected via env vars below (GITEA__MAILER__USER, GITEA__MAILER__PASSWD)
 
 # Persistence (shared storage for repos)
+#
+# 60Gi, not the original 20Gi. The container registry stores its blobs here,
+# and a single candle-vllm sm_70 image is 14GB uncompressed. At 20Gi the volume
+# filled, and every blob upload failed with:
+#   Package registry API internal error: 500 write /data/tmp/package-upload/...:
+#   no space left on device
+# The push client surfaces that as a bare `received unexpected HTTP status: 500`
+# after retrying every layer, which looks like a network fault and is not one.
+#
+# Note for whoever expands this again: Longhorn's admission webhook checks the
+# PHYSICAL free space of the disk holding each replica, not the thin volume
+# size. A request for 100Gi was rejected while the disk had 23.5Gi free:
+#   disk ... does not have sufficient physical space for expansion:
+#   physical free space would drop below minimal: left=14596177920 < minimal=49931236556
+# 60Gi fit. Raise it in steps and watch the webhook message.
 persistence:
   enabled: true
   storageClass: longhorn
-  size: 20Gi
+  size: 60Gi
   accessModes:
     - ReadWriteOnce
 
@@ -134,10 +149,29 @@ ingress:
   # wildcard. The wildcard covers this host.
 
 # Resources
+#
+# The 1000m ceiling throttled Gitea badly. Measured on a running pod:
+#   nr_throttled   189528 of 1169014 CFS periods  (16.2%)
+#   throttled_usec 4424898602 us                   (~74 minutes lost)
+#   actual usage   951m against the 1000m limit
+# The visible symptom was slow Actions traffic: FetchTask took 3469.9ms and
+# repository compare endpoints took 3091-3811ms, so runners timed out and
+# retried, which added more load to an already saturated process.
+#
+# The database was never the bottleneck. A real action_runner UPDATE measured
+# 2.133ms with zero lock waits, so this is CPU starvation in Gitea itself.
+#
+# requests.cpu stays 200m and limits.cpu stays 1000m. The 3000m limit was a
+# temporary measure while the Actions stall was diagnosed; the real fix landed
+# elsewhere, so the limit is back to the original.
+#
+# Caution: this pod runs on talos-cp-3, a control-plane node. If Gitea
+# throttles again at 1000m, move the workload off the control plane rather
+# than raising the limit.
 resources:
   requests:
     cpu: 200m
-    memory: 512Mi
+    memory: 768Mi
   limits:
     cpu: 1000m
     memory: 2Gi

k8s/infra/longhorn/disk-scan-job.yaml

@@ -0,0 +1,90 @@
+# One-shot destructive health scan of a candidate Longhorn disk.
+#
+# WHY THIS IS A MANUAL JOB AND NOT AN ARGOCD HOOK
+#   It writes over the entire raw device. An ArgoCD PostSync hook would run it
+#   on every sync and destroy whatever the volume holds. Apply it by hand,
+#   once, on a disk that has no partitions and no mounts.
+#
+# WHY NOT TALOS
+#   Talos declares disks by mounting them (machine.disks in the machine
+#   config). It has no primitive for "run this command on the host", so a scan
+#   cannot live in the Talos config. This Job is the IaC-correct home.
+#
+# ORDER OF OPERATIONS
+#   1. Scan the raw device with this Job        <- the disk must be UNMOUNTED
+#   2. If it passes, add the disk to terraform.tfvars.local `longhorn_disks`
+#      with device /dev/disk/by-id/<wwid> and run `make apply-cp`
+#   3. Register it with Longhorn via the pattern in longhorn-add-disks-job.yaml
+#
+# What it does
+#   fio writes a CRC32C-verified pattern across the whole device, then reads
+#   every block back and compares. Any unreadable or mismatched sector fails
+#   the Job. This is a full write and read pass, not a sample.
+#
+# Expected duration
+#   2 TB on a 5400 rpm 2.5" drive at roughly 90 MB/s is about 6 h to write and
+#   6 h to verify. Budget 12 hours. Check progress with:
+#     kubectl -n llm-io logs job/disk-scan-cp1-sdb -f
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: disk-scan-cp1-sdb
+  namespace: llm-io
+  labels:
+    app: disk-scan
+spec:
+  backoffLimit: 0
+  # Do not let a stuck scan retry silently.
+  template:
+    metadata:
+      labels:
+        app: disk-scan
+    spec:
+      restartPolicy: Never
+      nodeSelector:
+        kubernetes.io/hostname: talos-cp-1
+      containers:
+        - name: scan
+          image: ubuntu:22.04
+          securityContext:
+            privileged: true
+          command:
+            - /bin/bash
+            - -c
+            - |
+              set -euo pipefail
+              apt-get update -qq
+              apt-get install -y -qq smartmontools fio
+
+              echo "=== device identity ==="
+              smartctl -i /dev/sdb | tail -12
+
+              echo "=== SMART before ==="
+              smartctl -H /dev/sdb | tail -2
+
+              echo "=== full write + verify (destructive) ==="
+              fio --name=devverify \
+                  --filename=/dev/sdb \
+                  --rw=write \
+                  --bs=1M \
+                  --iodepth=4 \
+                  --ioengine=libaio \
+                  --direct=1 \
+                  --verify=crc32c \
+                  --do_verify=1 \
+                  --verify_fatal=1 \
+                  --eta=always \
+                  --output-format=normal
+
+              echo "=== SMART after ==="
+              smartctl -H /dev/sdb | tail -2
+              smartctl -A /dev/sdb | awk '/Reallocated_Sector|Current_Pending|Offline_Uncorrect|UDMA_CRC/{print "  "$0}'
+
+              echo "=== SCAN PASSED ==="
+          volumeMounts:
+            - name: dev
+              mountPath: /dev
+      volumes:
+        - name: dev
+          hostPath:
+            path: /dev