runners

mist_infra/arc/forgejo/deploy.sh

@@ -2,30 +2,57 @@
 set -euo pipefail
 
 # ==========================================================
-# Deploy Forgejo act_runner to Kubernetes
+# Deploy Forgejo act_runner + KEDA autoscaling to Kubernetes
 # ==========================================================
 
 NAMESPACE="forgejo-runners"
 SCRIPT_DIR="$(dirname "$0")"
 
-echo "=== Deploying Forgejo Runner ==="
+echo "=== Deploying Forgejo Runner with KEDA Autoscaling ==="
 
-# Prompt for token if not set in deployment.yaml
-TOKEN="RPAjk4Jdc42By5vSxnULPPPrjU0goPLQIiKgwOIo"
+RUNNER_TOKEN="RPAjk4Jdc42By5vSxnULPPPrjU0goPLQIiKgwOIo"
+
+# Forgejo API token — create at https://git.juntekim.com/user/settings/applications
+# Needs: read:repository scope (to list repos and query action tasks)
+if [[ -z "${FORGEJO_API_TOKEN:-}" ]]; then
+  read -rsp "Forgejo API token (for metrics exporter): " FORGEJO_API_TOKEN
   echo
+fi
+
+# Install KEDA if not already installed
+if ! kubectl get crd scaledobjects.keda.sh &>/dev/null; then
+  echo "=== Installing KEDA ==="
+  helm repo add kedacore https://kedacore.github.io/charts
+  helm repo update kedacore
+  helm install keda kedacore/keda --namespace keda --create-namespace
+  echo "✅ KEDA installed"
+else
+  echo "✅ KEDA already installed"
+fi
 
 kubectl create namespace "$NAMESPACE" --dry-run=client -o yaml | kubectl apply -f -
 
+# Runner registration token (used by act_runner to register with Forgejo)
 kubectl create secret generic forgejo-runner-secret \
   --namespace "$NAMESPACE" \
-  --from-literal=token="$TOKEN" \
+  --from-literal=token="$RUNNER_TOKEN" \
+  --dry-run=client -o yaml | kubectl apply -f -
+
+# Forgejo API token (used by metrics exporter to query job queue)
+kubectl create secret generic forgejo-api-secret \
+  --namespace "$NAMESPACE" \
+  --from-literal=token="$FORGEJO_API_TOKEN" \
   --dry-run=client -o yaml | kubectl apply -f -
 
 kubectl apply -f "$SCRIPT_DIR/deployment.yaml"
+kubectl apply -f "$SCRIPT_DIR/metrics-exporter.yaml"
+kubectl apply -f "$SCRIPT_DIR/keda-scaledobject.yaml"
 
 echo
-echo "✅ Forgejo runner deployed"
+echo "✅ Forgejo runner deployed with KEDA autoscaling"
 echo
 echo "Next steps:"
-echo "- kubectl get pods -n $NAMESPACE"
-echo "- Check runner appears at: https://git.juntekim.com/-/admin/runners"
+echo "  kubectl get pods -n $NAMESPACE"
+echo "  kubectl get scaledobject -n $NAMESPACE"
+echo "  kubectl logs -n $NAMESPACE deploy/forgejo-metrics-exporter"
+echo "  Check runners at: https://git.juntekim.com/-/admin/runners"

mist_infra/arc/forgejo/deployment.yaml

@@ -4,19 +4,6 @@ metadata:
   name: forgejo-runners
 ---
 apiVersion: v1
-kind: PersistentVolumeClaim
-metadata:
-  name: forgejo-runner-data
-  namespace: forgejo-runners
-spec:
-  accessModes:
-    - ReadWriteOnce
-  storageClassName: rook-ceph-block
-  resources:
-    requests:
-      storage: 1Gi
----
-apiVersion: v1
 kind: Secret
 metadata:
   name: forgejo-runner-secret
@@ -26,12 +13,13 @@ stringData:
   token: "RPAjk4Jdc42By5vSxnULPPPrjU0goPLQIiKgwOIo"
 ---
 apiVersion: apps/v1
-kind: Deployment
+kind: StatefulSet
 metadata:
   name: forgejo-runner
   namespace: forgejo-runners
 spec:
-  replicas: 1
+  replicas: 3
+  serviceName: forgejo-runner
   selector:
     matchLabels:
       app: forgejo-runner
@@ -40,6 +28,15 @@ spec:
       labels:
         app: forgejo-runner
     spec:
+      affinity:
+        podAntiAffinity:
+          preferredDuringSchedulingIgnoredDuringExecution:
+            - weight: 100
+              podAffinityTerm:
+                labelSelector:
+                  matchLabels:
+                    app: forgejo-runner
+                topologyKey: kubernetes.io/hostname
       initContainers:
         - name: register
           image: gitea/act_runner:latest
@@ -51,7 +48,7 @@ spec:
                 act_runner register --no-interactive \
                   --instance https://git.juntekim.com \
                   --token "RPAjk4Jdc42By5vSxnULPPPrjU0goPLQIiKgwOIo" \
-                  --name mist-runner \
+                  --name "$(hostname)" \
                   --labels "self-hosted,linux,x64"
               else
                 echo "Runner already registered, skipping."
@@ -81,7 +78,13 @@ spec:
           volumeMounts:
             - name: runner-data
               mountPath: /data
-      volumes:
-        - name: runner-data
-          persistentVolumeClaim:
-            claimName: forgejo-runner-data
+  volumeClaimTemplates:
+    - metadata:
+        name: runner-data
+      spec:
+        accessModes:
+          - ReadWriteOnce
+        storageClassName: rook-ceph-block
+        resources:
+          requests:
+            storage: 1Gi

mist_infra/arc/forgejo/keda-scaledobject.yaml

@@ -0,0 +1,21 @@
+apiVersion: keda.sh/v1alpha1
+kind: ScaledObject
+metadata:
+  name: forgejo-runner-scaler
+  namespace: forgejo-runners
+spec:
+  scaleTargetRef:
+    apiVersion: apps/v1
+    kind: StatefulSet
+    name: forgejo-runner
+  minReplicaCount: 1   # keep at least 1 runner registered
+  maxReplicaCount: 100  # tune based on cluster capacity, not node count
+  pollingInterval: 15  # seconds between KEDA polls
+  cooldownPeriod: 60   # seconds before scaling down after jobs clear
+  triggers:
+    - type: metrics-api
+      metadata:
+        # 1 replica per pending job (capped at maxReplicaCount)
+        targetValue: "1"
+        url: "http://forgejo-metrics-exporter.forgejo-runners.svc.cluster.local:8080"
+        valueLocation: "value"

mist_infra/arc/forgejo/metrics-exporter.yaml

@@ -0,0 +1,146 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: forgejo-metrics-script
+  namespace: forgejo-runners
+data:
+  server.py: |
+    #!/usr/bin/env python3
+    """
+    Polls the Forgejo API for waiting action tasks and exposes the count as JSON.
+    KEDA metrics-api scaler reads: { "value": <count> }
+    """
+    import http.server
+    import json
+    import os
+    import threading
+    import time
+    import urllib.request
+
+    FORGEJO_URL = os.environ["FORGEJO_URL"].rstrip("/")
+    TOKEN = "69eccfc51f720c21c615cfd5caa422fb02f0ab43" 
+    REFRESH = int(os.environ.get("REFRESH_INTERVAL", "20"))
+
+    _state = {"value": 0}
+    _lock = threading.Lock()
+
+
+    def fetch(path):
+        req = urllib.request.Request(
+            f"{FORGEJO_URL}/api/v1{path}",
+            headers={"Authorization": f"token {TOKEN}", "Accept": "application/json"},
+        )
+        with urllib.request.urlopen(req, timeout=10) as r:
+            return json.loads(r.read())
+
+
+    def count_pending_jobs():
+        total = 0
+        page = 1
+        while True:
+            repos = fetch(f"/repos/search?limit=50&page={page}")
+            items = repos.get("data", [])
+            for repo in items:
+                owner = repo["owner"]["login"]
+                name = repo["name"]
+                try:
+                    result = fetch(
+                        f"/repos/{owner}/{name}/actions/tasks?status=waiting&limit=1"
+                    )
+                    total += result.get("total_count", 0)
+                except Exception:
+                    pass
+            if len(items) < 50:
+                break
+            page += 1
+        return total
+
+
+    def refresh_loop():
+        while True:
+            try:
+                count = count_pending_jobs()
+                with _lock:
+                    _state["value"] = count
+                print(f"pending jobs: {count}", flush=True)
+            except Exception as e:
+                print(f"refresh error: {e}", flush=True)
+            time.sleep(REFRESH)
+
+
+    threading.Thread(target=refresh_loop, daemon=True).start()
+
+
+    class Handler(http.server.BaseHTTPRequestHandler):
+        def do_GET(self):
+            with _lock:
+                val = _state["value"]
+            body = json.dumps({"value": val}).encode()
+            self.send_response(200)
+            self.send_header("Content-Type", "application/json")
+            self.end_headers()
+            self.wfile.write(body)
+
+        def log_message(self, *_):
+            pass
+
+
+    http.server.HTTPServer(("", 8080), Handler).serve_forever()
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: forgejo-metrics-exporter
+  namespace: forgejo-runners
+spec:
+  replicas: 1
+  selector:
+    matchLabels:
+      app: forgejo-metrics-exporter
+  template:
+    metadata:
+      labels:
+        app: forgejo-metrics-exporter
+    spec:
+      containers:
+        - name: exporter
+          image: python:3.12-alpine
+          command: ["python", "/scripts/server.py"]
+          env:
+            - name: FORGEJO_URL
+              value: https://git.juntekim.com
+            - name: FORGEJO_TOKEN
+              valueFrom:
+                secretKeyRef:
+                  name: forgejo-api-secret
+                  key: token
+            - name: REFRESH_INTERVAL
+              value: "20"
+          ports:
+            - containerPort: 8080
+          resources:
+            requests:
+              cpu: 10m
+              memory: 32Mi
+            limits:
+              cpu: 100m
+              memory: 64Mi
+          volumeMounts:
+            - name: scripts
+              mountPath: /scripts
+      volumes:
+        - name: scripts
+          configMap:
+            name: forgejo-metrics-script
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: forgejo-metrics-exporter
+  namespace: forgejo-runners
+spec:
+  selector:
+    app: forgejo-metrics-exporter
+  ports:
+    - port: 8080
+      targetPort: 8080