sparkstack/sparkstack.yaml.example at main · jlapenna/sparkstack · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
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
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
# Spark Services Orchestration - Operational & Overlay Network Configuration
# Copy this file to ~/.config/sparkstack/sparkstack.yaml and adjust values as needed.
# (If not present in ~/.config/sparkstack/, sparkstack will look in the project root)

# List of services to be enabled on the head node.
# Supported values: SparkRun, InferenceStack, RegistrySync, Monitoring, OpenClaw, Headscale
enabled_services:
  - SparkRun
  - InferenceStack
  - RegistrySync
  - Monitoring
  - OpenClaw
  - Headscale

# Overlay network settings (Headscale / Tailscale VPN setup)
overlay:
  # The routable control plane LAN IP or DNS name of the Headscale server.
  # If empty, the overlay network is considered unconfigured.
  headscale_server: ""

  # The port Headscale listens on (also used for the docker-compose port mapping).
  headscale_port: 8080

  # Reusable pre-auth key generated by sparkstack setup. Used by all
  # Tailscale sidecars to authenticate with Headscale.
  headscale_auth_key: ""

  # The Tailnet IP of the head node's sidecar container.
  head_tailnet_ip: ""

  # The Tailnet IP of the worker node sidecar.
  worker_tailnet_ip: ""

  # Pinned Tailscale client image version for sidecar containers.
  tailscale_version: "v1.82.5"

# Targets (SSH connection strings or Docker context names) where services should deploy.
# Leave empty to run the service locally on the Head node.
targets:
  spark: "ssh://localhost"
  openclaw: "ssh://localhost"
  monitoring: "ssh://localhost"

# Detached monitoring ownership settings.
# When set, sparkstack skips deploying local Prometheus/Grafana/Tempo and pushes to this host instead.
monitoring_host: ""

# Port configuration
ports:
  # Base/control port for the vLLM API Gateway
  vllm_port: 4000

  # Port offset where backend vLLM nodes start allocating ports
  backend_start_port: 8001

# Hardware constraints and Blackwell specification overrides
hardware:
  # Usable pool of system memory in GB after OS and driver overhead
  usable_spark_memory_gb: 121.0

  # Memory reserved for host OS, kernel, Docker daemon, SSH, and monitoring containers
  system_reserved_memory_gb: 12.0

  # NVIDIA Blackwell Resource Constraints (VRAM & Cache)
  max_vram_utilization: 0.95
  default_kv_cache_ceiling: 128000