disagg_planner.yaml 4.55 KB
Newer Older
1
2
# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
# SPDX-License-Identifier: Apache-2.0
Alec's avatar
Alec committed
3

4
5
6
apiVersion: nvidia.com/v1alpha1
kind: DynamoGraphDeployment
metadata:
7
  name: vllm-disagg-planner
8
9
  annotations:
    nvidia.com/enable-grove: "false" # temporarily disable grove because current k8s connector does not work with grove
10
spec:
11
12
13
14
15
  envs:
    - name: DYNAMO_SERVICE_CONFIG
      value: '{"Prometheus":{"global":{"scrape_interval":"5s"},"scrape_configs":[{"job_name":"prometheus","static_configs":[{"targets":["localhost:9090"]}]},{"job_name":"frontend","static_configs":[{"targets":["vllm-disagg-planner-frontend:8000"]}]}]}}'
    - name: DYNAMO_NAMESPACE
      value: "vllm-disagg-planner"
16
17
    - name: PROMETHEUS_PORT
      value: "8000"
18
19
  services:
    Frontend:
20
      dynamoNamespace: vllm-disagg-planner
21
      componentType: frontend
22
      replicas: 1
23
24
      extraPodSpec:
        mainContainer:
25
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
    Planner:
      dynamoNamespace: vllm-disagg-planner
      envFromSecret: hf-token-secret
      componentType: planner
      replicas: 1
      livenessProbe:
        exec:
          command:
            - /bin/sh
            - -c
            - "exit 0"
        periodSeconds: 60
        timeoutSeconds: 30
        failureThreshold: 10
      readinessProbe:
        exec:
          command:
            - /bin/sh
            - -c
            - "exit 0"
        initialDelaySeconds: 60
        periodSeconds: 60
        timeoutSeconds: 30
        failureThreshold: 10
      pvc:
        create: false
52
        name: dynamo-pvc # Must be pre-created before deployment and SLA profiler must have been run
53
        mountPoint: /data
54
55
      extraPodSpec:
        mainContainer:
56
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
57
          workingDir: /workspace/components/planner/src/dynamo/planner
julienmancuso's avatar
julienmancuso committed
58
59
60
          command:
            - /bin/sh
            - -c
61
          args:
julienmancuso's avatar
julienmancuso committed
62
63
64
65
66
            - >-
              python3 -m planner_sla
              --environment=kubernetes
              --backend=vllm
              --adjustment-interval=60
67
              --profile-results-dir=/data/profiling_results
68
    Prometheus: # NOTE: this is set on Prometheus to ensure a service is created for the Prometheus component. This is a workaround and should be managed differently.
69
      dynamoNamespace: vllm-disagg-planner
70
      componentType: frontend
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
      replicas: 1
      envs:
        - name: PYTHONPATH
          value: "/workspace/components/planner/src"
      livenessProbe:
        exec:
          command:
            - /bin/sh
            - -c
            - "exit 0"
        periodSeconds: 60
        timeoutSeconds: 30
        failureThreshold: 10
      readinessProbe:
        exec:
          command:
            - /bin/sh
            - -c
            - "exit 0"
        initialDelaySeconds: 30
        periodSeconds: 60
        timeoutSeconds: 30
        failureThreshold: 10
      extraPodSpec:
        mainContainer:
96
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
Alec's avatar
Alec committed
97
          workingDir: /workspace/components/backends/vllm
98
99
100
          command:
            - /bin/sh
            - -c
101
          args:
102
            - "python3 -m dynamo.planner.prometheus"
103
    VllmDecodeWorker:
104
      dynamoNamespace: vllm-disagg-planner
105
      envFromSecret: hf-token-secret
106
      componentType: worker
107
      replicas: 2
108
109
      resources:
        limits:
110
          gpu: "1"
111
112
      extraPodSpec:
        mainContainer:
113
114
115
116
117
118
          startupProbe:
            httpGet:
              path: /health
              port: 9090
            periodSeconds: 10
            failureThreshold: 60
119
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
Alec's avatar
Alec committed
120
          workingDir: /workspace/components/backends/vllm
121
          command:
122
            - python3
123
          args:
124
125
126
127
            - -m
            - dynamo.vllm
            - --model
            - Qwen/Qwen3-0.6B
128
    VllmPrefillWorker:
129
      dynamoNamespace: vllm-disagg-planner
130
      envFromSecret: hf-token-secret
131
      componentType: worker
132
      replicas: 2
133
134
      resources:
        limits:
135
          gpu: "1"
136
137
      extraPodSpec:
        mainContainer:
138
139
140
141
142
143
          startupProbe:
            httpGet:
              path: /health
              port: 9090
            periodSeconds: 10
            failureThreshold: 60
144
          image: nvcr.io/nvidia/ai-dynamo/vllm-runtime:0.4.1
Alec's avatar
Alec committed
145
          workingDir: /workspace/components/backends/vllm
146
          command:
147
            - python3
148
          args:
149
150
151
152
153
            - -m
            - dynamo.vllm
            - --model
            - Qwen/Qwen3-0.6B
            - --is-prefill-worker