agg_router.sh 3.08 KB
Newer Older
1
#!/bin/bash
2
# SPDX-FileCopyrightText: Copyright (c) 2025-2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
3
# SPDX-License-Identifier: Apache-2.0
4
5
6
#
# Two aggregated workers behind a KV-aware router.
# GPUs: 2
7

8
9
10
11
12
set -e
trap 'echo Cleaning up...; kill 0' EXIT

SCRIPT_DIR="$(dirname "$(readlink -f "$0")")"
source "$SCRIPT_DIR/../../../common/launch_utils.sh"
13

14
15
# Parse command line arguments
ENABLE_OTEL=false
16
APPROX_MODE=false
17
18
19
20
21
22
while [[ $# -gt 0 ]]; do
    case $1 in
        --enable-otel)
            ENABLE_OTEL=true
            shift
            ;;
23
24
25
26
        --approx)
            APPROX_MODE=true
            shift
            ;;
27
28
29
30
        -h|--help)
            echo "Usage: $0 [OPTIONS]"
            echo "Options:"
            echo "  --enable-otel        Enable OpenTelemetry tracing"
31
            echo "  --approx             Enable approximate KV routing (no KV events)"
32
33
34
35
36
37
38
39
40
41
42
43
44
45
            echo "  -h, --help           Show this help message"
            echo ""
            echo "Note: System metrics are enabled by default on ports 8081 (worker-1), 8082 (worker-2)"
            exit 0
            ;;
        *)
            echo "Unknown option: $1"
            echo "Use --help for usage information"
            exit 1
            ;;
    esac
done

# Enable tracing if requested
46
TRACE_ARGS=()
47
48
49
50
if [ "$ENABLE_OTEL" = true ]; then
    export DYN_LOGGING_JSONL=true
    export OTEL_EXPORT_ENABLED=1
    export OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=${OTEL_EXPORTER_OTLP_TRACES_ENDPOINT:-http://localhost:4317}
51
    TRACE_ARGS+=(--enable-trace --otlp-traces-endpoint localhost:4317)
52
fi
53

54
55
MODEL="Qwen/Qwen3-0.6B"
HTTP_PORT="${DYN_HTTP_PORT:-8000}"
56
print_launch_banner "Launching Aggregated + KV Routing (2 GPUs)" "$MODEL" "$HTTP_PORT"
57

58
# run ingress
59
# dynamo.frontend accepts either --http-port flag or DYN_HTTP_PORT env var (defaults to 8000)
60
61
62
63
FRONTEND_ARGS=(--router-mode kv)
if [ "$APPROX_MODE" = true ]; then
    FRONTEND_ARGS+=(--no-kv-events)
fi
64
OTEL_SERVICE_NAME=dynamo-frontend \
65
python3 -m dynamo.frontend "${FRONTEND_ARGS[@]}" &
66
67

# run worker
68
69
70
71
72
73
74
75
# Build KV events args conditionally (only when not in approx mode)
KV_EVENTS_ARGS_1=()
KV_EVENTS_ARGS_2=()
if [ "$APPROX_MODE" = false ]; then
    KV_EVENTS_ARGS_1=(--kv-events-config '{"publisher":"zmq","topic":"kv-events","endpoint":"tcp://*:5557"}')
    KV_EVENTS_ARGS_2=(--kv-events-config '{"publisher":"zmq","topic":"kv-events","endpoint":"tcp://*:5558"}')
fi

76
OTEL_SERVICE_NAME=dynamo-worker-1 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_WORKER1:-8081} \
77
python3 -m dynamo.sglang \
78
79
  --model-path Qwen/Qwen3-0.6B \
  --served-model-name Qwen/Qwen3-0.6B \
80
81
82
  --page-size 16 \
  --tp 1 \
  --trust-remote-code \
83
  "${KV_EVENTS_ARGS_1[@]}" \
84
85
  --enable-metrics \
  "${TRACE_ARGS[@]}" &
86

87
OTEL_SERVICE_NAME=dynamo-worker-2 DYN_SYSTEM_PORT=${DYN_SYSTEM_PORT_WORKER2:-8082} \
88
CUDA_VISIBLE_DEVICES=1 python3 -m dynamo.sglang \
89
90
  --model-path Qwen/Qwen3-0.6B \
  --served-model-name Qwen/Qwen3-0.6B \
91
92
93
  --page-size 16 \
  --tp 1 \
  --trust-remote-code \
94
  "${KV_EVENTS_ARGS_2[@]}" \
95
  --enable-metrics \
96
97
98
99
  "${TRACE_ARGS[@]}" &

# Exit on first worker failure; kill 0 in the EXIT trap tears down the rest
wait_any_exit