Add OpenSearch esrally benchmark for Kubernetes to PKB.

perfkitbenchmarker/data/cluster/swap_encryption_daemonset.yaml.j2

@@ -13,9 +13,13 @@ data:
     # Installs measurement tools, verifies the swap device, then writes /tmp/pkb_ready.
     set -euo pipefail
 
-    echo "[pkb] Installing measurement tools..."
-    # Only tools needed for Phase 1 (raw-device fio) and Phase 2 (CPU/I/O overhead).
-    # Workload benchmarks (redis, opensearch, kernel-build) run in separate pods.
+    echo "[pkb] Installing benchmark measurement tools..."
+    # Phase 1 tools: fio (raw-device I/O), cryptsetup/mdadm (dm-crypt inspection),
+    # sysstat (vmstat/pidstat), nvme-cli (NVMe telemetry).
+    # Phase 2 tools: stress-ng (CPU/I/O overhead), cgroup-tools (cgroup v1 guard),
+    # util-linux (taskset/ionice).
+    # Phase 3b tools: gcc/make/bc/flex/bison/libelf-dev/libssl-dev (kernel build).
+    # Workload benchmarks (redis, opensearch) run in separate PKB benchmark pods.
     PKB_APT_OK=0
     for _attempt in 1 2 3; do
       DEBIAN_FRONTEND=noninteractive apt-get update -qq 2>&1 || true
@@ -25,15 +29,30 @@ data:
         mdadm \
         sysstat \
         nvme-cli \
+        stress-ng \
+        cgroup-tools \
+        util-linux \
+        gcc \
+        make \
+        bc \
+        flex \
+        bison \
+        libelf-dev \
+        libssl-dev \
         2>&1 && PKB_APT_OK=1 && break
       echo "[pkb] apt-get attempt $_attempt failed, retrying in 15s..." >&2
       sleep 15
     done
-    if [ "$PKB_APT_OK" != "1" ] || ! command -v fio >/dev/null 2>&1; then
-      echo "[pkb] FATAL: fio not installed after 3 attempts" >&2
+    if [ "$PKB_APT_OK" != "1" ] || \
+       ! command -v fio >/dev/null 2>&1 || \
+       ! command -v stress-ng >/dev/null 2>&1 || \
+       ! command -v make >/dev/null 2>&1; then
+      echo "[pkb] FATAL: critical tools (fio, stress-ng, make) not installed after 3 attempts" >&2
       exit 1
     fi
-    echo "[pkb] fio installed: $(fio --version 2>&1 | head -1)"
+    echo "[pkb] fio: $(fio --version 2>&1 | head -1)"
+    echo "[pkb] stress-ng: $(stress-ng --version 2>&1 | head -1)"
+    echo "[pkb] gcc: $(gcc --version 2>&1 | head -1)"
 
     echo "[pkb] Verifying swap device is active..."
     PKB_SWAP_FOUND=0
@@ -53,7 +72,7 @@ data:
       exit 1
     fi
 
-    echo "[pkb] Measurement tools ready. Writing ready sentinel."
+    echo "[pkb] Benchmark tools ready. Writing ready sentinel."
     touch /tmp/pkb_ready
     exec sleep infinity
 ---

perfkitbenchmarker/data/container/kubernetes_opensearch_esrally/esrally_job.yaml.j2

@@ -0,0 +1,80 @@
+# opensearch-benchmark Kubernetes Job
+# Rendered by kubernetes_opensearch_esrally_benchmark.py Run().
+#
+# Template variables:
+#   k8s_namespace      - Kubernetes namespace (default: 'default')
+#   race_id            - Unique test-execution identifier string
+#   target_hosts       - OpenSearch Service host:port
+#   esrally_track      - opensearch-benchmark workload name, e.g. geonames
+#   esrally_challenge  - test-procedure name, e.g. append-no-conflicts
+#   esrally_version    - opensearch-benchmark pip package version, e.g. 1.7.0
+#
+# The Job installs opensearch-benchmark at runtime, runs the test in
+# benchmark-only pipeline against the OpenSearch headless Service, then
+# prints test_execution.json between PKB sentinel markers so Run() can
+# parse the results.
+---
+apiVersion: batch/v1
+kind: Job
+metadata:
+  name: esrally-pkb
+  namespace: {{ k8s_namespace | default('default') }}
+  labels:
+    pkb: "true"
+spec:
+  backoffLimit: 0
+  template:
+    metadata:
+      labels:
+        pkb: "true"
+    spec:
+      nodeSelector:
+        pkb_nodepool: clients
+      restartPolicy: Never
+      containers:
+      - name: esrally
+        image: python:3.11-slim
+        resources:
+          requests:
+            memory: "4Gi"
+            cpu: "2"
+          limits:
+            memory: "8Gi"
+            cpu: "4"
+        env:
+        - name: BENCHMARK_HOME
+          value: /var/benchmark
+        command:
+        - bash
+        - -c
+        - |
+          set -euo pipefail
+          export DEBIAN_FRONTEND=noninteractive
+          echo "[pkb] Installing system dependencies (git)..."
+          apt-get update -qq && apt-get install -y --no-install-recommends git > /dev/null
+          echo "[pkb] Installing opensearch-benchmark {{ esrally_version }}..."
+          pip install opensearch-benchmark=={{ esrally_version }} --quiet
+
+          echo "[pkb] Running opensearch-benchmark (workload={{ esrally_track }}, test-procedure={{ esrally_challenge }})..."
+          opensearch-benchmark execute-test \
+            --pipeline=benchmark-only \
+            --target-host="{{ target_hosts }}" \
+            --workload="{{ esrally_track }}" \
+            --test-procedure="{{ esrally_challenge }}" \
+            --workload-params="number_of_replicas:0" \
+            --test-execution-id="{{ race_id }}"
+
+          echo "[pkb] Test complete. Locating test_execution.json..."
+          # Use -path in find (not grep) to avoid non-zero exit with pipefail.
+          TEST_JSON=$(find /root /var/benchmark \
+            -path "*{{ race_id }}*" -name "test_execution.json" 2>/dev/null | head -1)
+          if [ -z "$TEST_JSON" ]; then
+            echo "[pkb-error] test_execution.json not found. Listing all benchmark files:"
+            find /root /var/benchmark -maxdepth 8 -type f 2>/dev/null | sort | head -50
+            exit 1
+          fi
+          echo "[pkb] Found: $TEST_JSON"
+          echo "=== PKB_RACE_JSON_BEGIN ==="
+          cat "$TEST_JSON"
+          echo ""
+          echo "=== PKB_RACE_JSON_END ==="

perfkitbenchmarker/data/container/kubernetes_opensearch_esrally/opensearch.yaml.j2

@@ -0,0 +1,94 @@
+# OpenSearch single-node StatefulSet + headless Service (clusterIP: None)
+# Rendered by kubernetes_opensearch_esrally_benchmark.py Prepare().
+#
+# Template variables:
+#   k8s_namespace      - Kubernetes namespace (default: 'default')
+#   opensearch_version - OpenSearch image tag, e.g. 2.13.0
+#   opensearch_port    - HTTP port, e.g. 9200
+#
+# Nodes in the 'servers' nodepool are selected via the pkb_nodepool label
+# which PKB sets automatically during nodepool creation.
+# An initContainer sets vm.max_map_count=262144 which OpenSearch requires.
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: opensearch
+  namespace: {{ k8s_namespace | default('default') }}
+  labels:
+    app: opensearch
+    pkb: "true"
+spec:
+  selector:
+    app: opensearch
+  ports:
+  - name: http
+    port: {{ opensearch_port }}
+    targetPort: {{ opensearch_port }}
+  clusterIP: None
+---
+apiVersion: apps/v1
+kind: StatefulSet
+metadata:
+  name: opensearch
+  namespace: {{ k8s_namespace | default('default') }}
+  labels:
+    app: opensearch
+    pkb: "true"
+spec:
+  # No volumeClaimTemplates: OpenSearch data lives on ephemeral pod
+  # storage, giving each PKB run a clean index (intentional).
+  serviceName: opensearch
+  replicas: 1
+  selector:
+    matchLabels:
+      app: opensearch
+  template:
+    metadata:
+      labels:
+        app: opensearch
+        pkb: "true"
+    spec:
+      nodeSelector:
+        pkb_nodepool: servers
+      tolerations:
+      - operator: Exists
+      initContainers:
+      - name: sysctl
+        image: busybox:1.36
+        command:
+        - sysctl
+        - -w
+        - vm.max_map_count=262144
+        securityContext:
+          privileged: true
+      containers:
+      - name: opensearch
+        image: opensearchproject/opensearch:{{ opensearch_version }}
+        ports:
+        - name: http
+          containerPort: {{ opensearch_port }}
+        env:
+        - name: discovery.type
+          value: single-node
+        - name: DISABLE_SECURITY_PLUGIN
+          value: "true"
+        - name: OPENSEARCH_JAVA_OPTS
+          value: "-Xms4g -Xmx4g"
+        - name: bootstrap.memory_lock
+          value: "false"
+        resources:
+          requests:
+            memory: "8Gi"
+            cpu: "2"
+          limits:
+            memory: "16Gi"
+            cpu: "4"
+        readinessProbe:
+          httpGet:
+            path: /_cluster/health
+            port: {{ opensearch_port }}
+          initialDelaySeconds: 30
+          periodSeconds: 10
+          timeoutSeconds: 5
+          failureThreshold: 12