# auto_aiperf.sh 配置文件（bash 语法）

# 基础环境
NAMESPACE="ai-inference"
RELEASE_NAME="infernex"
CHART_DIR="${CHART_DIR:-/path/to/InferNex/charts/infernex}"
PROJECT_DIR="${PROJECT_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")/../.." && pwd)}"
ARTIFACTS_DIR="${PROJECT_DIR}/artifacts"
BASE_RESULT_DIR_NAME="Qwen_Qwen3-8B-openai-chat-fixed_schedule"

# 轮询与等待
POLL_INTERVAL_SECONDS=5
POST_UNINSTALL_WAIT_SECONDS=20
PODS_READY_TIMEOUT_SECONDS=900
LOOP_INTERVAL_SECONDS=10

# 进程匹配与网关
AIPERF_PROCESS_PATTERN="aiperf profile"
GATEWAY_SERVICE_NAME="inference-gateway-istio"
GATEWAY_SERVICE_PORT="80"

# decode 指标采集脚本
PYTHON_BIN="python3"
DECODE_COLLECTOR_SCRIPT="${PROJECT_DIR}/scripts/perf/collect_kv_transfer_metrics.py"
DECODE_COST_IDLE_TIMEOUT=10

# aiperf 命令（bash 数组）
AIPERF_CMD=(
  aiperf profile
  --input-file traces/toolagent_trace.jsonl
  --custom-dataset-type mooncake_trace
  --model Qwen/Qwen3-8B
  --endpoint-type chat
  --synthesis-max-isl 15000
  --extra-inputs ignore_eos:true
  --isl-block-size 512
  --streaming
)