hyper-stack-4j/cluster-config.yaml at master · dev4any1/hyper-stack-4j · 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
67
68
# hyper-stack-4j cluster configuration
# Copy and adjust per deployment

cluster:
  name: hyper-stack-4j-cluster
  seed-nodes:
    - 192.168.1.10:5701     # coordinator-1 (stable machine)
    - 192.168.1.11:5701     # coordinator-2 (stable machine)
  seed-node-count: 2        # never 1 — no SPOF
  backup-count: 2           # Hazelcast map replicas

coordinator:
  count: 2                  # leader + standby
  grpc-port: 9090
  http-port: 8080
  max-queue-depth: 1000
  max-batch-size: 8
  preemption-enabled: true
  preemption-strategy: ABORT  # ABORT | CHECKPOINT

scheduler:
  max-wait-ms: 50
  priority-weights:
    HIGH: 3
    NORMAL: 1
    LOW: 1

node:
  grpc-port: 9091
  device-id: 0              # CUDA device index
  vram-headroom-fraction: 0.10  # keep 10% VRAM free

kv-cache:
  gpu:
    capacity-fraction: 0.85
    eviction: LRU
  cpu:
    capacity-gb: 24
    eviction: LFU
  disk:
    capacity-gb: 200
    path: /mnt/kvcache
    eviction: LFU

health:
  probe-interval-ms: 5000
  vram-warning-threshold: 0.90
  vram-critical-threshold: 0.98
  circuit-breaker:
    failure-rate-threshold: 50
    sliding-window-size: 10
    wait-duration-seconds: 30

sampling:
  defaults:
    temperature: 0.7
    top-k: 50
    top-p: 0.9
    repetition-penalty: 1.1
    max-tokens: 512
  profiles:
    deterministic:
      temperature: 0.1
      greedy: true
    creative:
      temperature: 1.2
      top-k: 100
      top-p: 0.95