Add exact production-trace real replay client

runs/frontier-fidelity-envelope-v1/run_qwen30_exact_trace_real_anchor.sh

@@ -0,0 +1,89 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}"
+REQUESTS_FILE="${REQUESTS_FILE:?REQUESTS_FILE is required}"
+TP="${TP:?TP is required}"
+MNS="${MNS:?MNS is required}"
+TRACE_LABEL="${TRACE_LABEL:?TRACE_LABEL is required}"
+SERVER_PORT="${SERVER_PORT:?SERVER_PORT is required}"
+VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
+MODEL_ROOT="${MODEL_ROOT:-/home/admin/cpfs/wjh/models/Qwen/Qwen3-30B-A3B}"
+SERVED_MODEL="qwen3-30b-exact-trace"
+SERVER_PID=""
+
+mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance" "${OUTPUT_ROOT}/results"
+exec > >(tee -a "${OUTPUT_ROOT}/logs/controller.log") 2>&1
+
+cleanup() {
+  if [[ -n "${SERVER_PID}" ]] && kill -0 "${SERVER_PID}" 2>/dev/null; then
+    kill -TERM -- "-${SERVER_PID}" 2>/dev/null || true
+    for _ in $(seq 1 30); do
+      kill -0 "${SERVER_PID}" 2>/dev/null || break
+      sleep 1
+    done
+    kill -KILL -- "-${SERVER_PID}" 2>/dev/null || true
+  fi
+  SERVER_PID=""
+}
+trap cleanup EXIT INT TERM
+
+IFS=',' read -r -a GPU_IDS <<< "${CUDA_VISIBLE_DEVICES:?fleet GPU allocation is required}"
+if [[ "${#GPU_IDS[@]}" -ne "${TP}" ]]; then
+  echo "ERROR: expected ${TP} GPUs, got ${CUDA_VISIBLE_DEVICES}" >&2
+  exit 1
+fi
+
+REQUEST_COUNT="$(wc -l < "${REQUESTS_FILE}")"
+echo "QWEN30_EXACT_TRACE_REAL_LAUNCH_ECHO host=$(hostname) gpus=${CUDA_VISIBLE_DEVICES} model=${MODEL_ROOT} runtime=vLLM-0.20.0+cu129 dtype=BF16 config=TP${TP}_MNS${MNS}_MBT8192 trace=${TRACE_LABEL} requests=${REQUEST_COUNT} source=${REQUESTS_FILE} arrivals=original600s input_output_prompt=exact prefix=on block=16 tpot_slo=150ms output=${OUTPUT_ROOT} expected_wall=12-35m hard_wall=3600s"
+date -u +"START_UTC=%Y-%m-%dT%H:%M:%SZ"
+sha256sum qwen30_exact_trace_client.py run_qwen30_exact_trace_real_anchor.sh \
+  > "${OUTPUT_ROOT}/provenance/source.sha256"
+sha256sum "${REQUESTS_FILE}" > "${OUTPUT_ROOT}/provenance/requests.sha256"
+sha256sum "${MODEL_ROOT}/config.json" > "${OUTPUT_ROOT}/provenance/model.sha256"
+nvidia-smi --query-gpu=index,name,uuid,driver_version --format=csv,noheader \
+  > "${OUTPUT_ROOT}/provenance/gpus.csv"
+
+export TOKENIZERS_PARALLELISM=false
+export VLLM_USE_V1=1
+export TORCH_CUDA_ARCH_LIST=9.0
+export HF_HUB_OFFLINE=1
+export TRANSFORMERS_OFFLINE=1
+ulimit -n 65536
+
+setsid "${VENV_ROOT}/bin/vllm" serve "${MODEL_ROOT}" \
+  --host 127.0.0.1 --port "${SERVER_PORT}" --served-model-name "${SERVED_MODEL}" \
+  --tensor-parallel-size "${TP}" --gpu-memory-utilization 0.92 \
+  --max-model-len 40960 --max-num-batched-tokens 8192 --max-num-seqs "${MNS}" \
+  --enable-prefix-caching --enable-chunked-prefill --no-enable-log-requests \
+  > "${OUTPUT_ROOT}/logs/server.log" 2>&1 &
+SERVER_PID=$!
+READY=0
+for _ in $(seq 1 120); do
+  if curl -fsS --max-time 2 "http://127.0.0.1:${SERVER_PORT}/v1/models" \
+    > "${OUTPUT_ROOT}/results/models.json" 2>/dev/null; then
+    READY=1
+    break
+  fi
+  if ! kill -0 "${SERVER_PID}" 2>/dev/null; then
+    tail -200 "${OUTPUT_ROOT}/logs/server.log"
+    exit 1
+  fi
+  sleep 3
+done
+if [[ "${READY}" -ne 1 ]]; then
+  tail -200 "${OUTPUT_ROOT}/logs/server.log"
+  exit 1
+fi
+
+"${VENV_ROOT}/bin/python" qwen30_exact_trace_client.py \
+  --port "${SERVER_PORT}" --requests-file "${REQUESTS_FILE}" \
+  --output "${OUTPUT_ROOT}/results/result.json" --tpot-slo-ms 150 \
+  --timeout-seconds 1800
+
+cleanup
+find "${OUTPUT_ROOT}" -type f ! -path '*/provenance/artifacts.sha256' -print0 \
+  | sort -z | xargs -0 sha256sum > "${OUTPUT_ROOT}/provenance/artifacts.sha256"
+date -u +"END_UTC=%Y-%m-%dT%H:%M:%SZ"
+echo "QWEN30_EXACT_TRACE_REAL_ANCHOR_COMPLETE"