343 lines
9.2 KiB
YAML
343 lines
9.2 KiB
YAML
# ============================================================
|
|
# Velocity-OS — K3s Deployments
|
|
# All services in velocity-os namespace.
|
|
# GPU: RTX 6000 Blackwell 96GB VRAM — MIG partitioned.
|
|
# MIG slice 0 (48GB): SGLang LLM inference (core-api)
|
|
# MIG slice 1 (48GB): ComfyUI media generation (media-engine)
|
|
# ============================================================
|
|
|
|
---
|
|
# ── PostgreSQL (StatefulSet for stable identity) ─────────────
|
|
apiVersion: apps/v1
|
|
kind: StatefulSet
|
|
metadata:
|
|
name: postgres
|
|
namespace: velocity-os
|
|
labels:
|
|
app: postgres
|
|
tier: database
|
|
spec:
|
|
serviceName: postgres
|
|
replicas: 1
|
|
selector:
|
|
matchLabels:
|
|
app: postgres
|
|
template:
|
|
metadata:
|
|
labels:
|
|
app: postgres
|
|
tier: database
|
|
spec:
|
|
containers:
|
|
- name: postgres
|
|
image: ${ECR_REGISTRY}/postgres:15-alpine
|
|
ports:
|
|
- containerPort: 5432
|
|
env:
|
|
- name: POSTGRES_DB
|
|
valueFrom:
|
|
secretKeyRef:
|
|
name: velocity-secrets
|
|
key: POSTGRES_DB
|
|
- name: POSTGRES_USER
|
|
valueFrom:
|
|
secretKeyRef:
|
|
name: velocity-secrets
|
|
key: POSTGRES_USER
|
|
- name: POSTGRES_PASSWORD
|
|
valueFrom:
|
|
secretKeyRef:
|
|
name: velocity-secrets
|
|
key: POSTGRES_PASSWORD
|
|
- name: PGDATA
|
|
value: /var/lib/postgresql/data/pgdata
|
|
resources:
|
|
requests:
|
|
memory: "1Gi"
|
|
cpu: "500m"
|
|
limits:
|
|
memory: "2Gi"
|
|
cpu: "1000m"
|
|
volumeMounts:
|
|
- name: postgres-data
|
|
mountPath: /var/lib/postgresql/data
|
|
livenessProbe:
|
|
exec:
|
|
command: [pg_isready, -U, velocity]
|
|
initialDelaySeconds: 30
|
|
periodSeconds: 10
|
|
readinessProbe:
|
|
exec:
|
|
command: [pg_isready, -U, velocity]
|
|
initialDelaySeconds: 5
|
|
periodSeconds: 5
|
|
volumes:
|
|
- name: postgres-data
|
|
persistentVolumeClaim:
|
|
claimName: pvc-postgres-data
|
|
|
|
---
|
|
# ── Redis (session cache, future queue) ──────────────────────
|
|
apiVersion: apps/v1
|
|
kind: Deployment
|
|
metadata:
|
|
name: redis
|
|
namespace: velocity-os
|
|
labels:
|
|
app: redis
|
|
spec:
|
|
replicas: 1
|
|
selector:
|
|
matchLabels:
|
|
app: redis
|
|
template:
|
|
metadata:
|
|
labels:
|
|
app: redis
|
|
spec:
|
|
containers:
|
|
- name: redis
|
|
image: ${ECR_REGISTRY}/redis:7-alpine
|
|
ports:
|
|
- containerPort: 6379
|
|
resources:
|
|
requests:
|
|
memory: "256Mi"
|
|
cpu: "100m"
|
|
limits:
|
|
memory: "512Mi"
|
|
cpu: "250m"
|
|
args: ["--maxmemory", "400mb", "--maxmemory-policy", "allkeys-lru"]
|
|
|
|
---
|
|
# ── Core API (FastAPI) ────────────────────────────────────────
|
|
apiVersion: apps/v1
|
|
kind: Deployment
|
|
metadata:
|
|
name: core-api
|
|
namespace: velocity-os
|
|
labels:
|
|
app: core-api
|
|
tier: backend
|
|
spec:
|
|
replicas: 2
|
|
selector:
|
|
matchLabels:
|
|
app: core-api
|
|
strategy:
|
|
type: RollingUpdate
|
|
rollingUpdate:
|
|
maxUnavailable: 0
|
|
maxSurge: 1
|
|
template:
|
|
metadata:
|
|
labels:
|
|
app: core-api
|
|
tier: backend
|
|
spec:
|
|
# MIG slice 0: SGLang LLM inference
|
|
# The core-api pod requests MIG slice via resource limit
|
|
runtimeClassName: nvidia
|
|
containers:
|
|
- name: core-api
|
|
image: ${ECR_REGISTRY}/velocity-os/core:latest
|
|
ports:
|
|
- containerPort: 8443
|
|
envFrom:
|
|
- secretRef:
|
|
name: velocity-secrets
|
|
- configMapRef:
|
|
name: velocity-config
|
|
resources:
|
|
requests:
|
|
memory: "1Gi"
|
|
cpu: "500m"
|
|
# RTX 6000 Blackwell MIG 3g.48gb (SGLang slice)
|
|
nvidia.com/mig-3g.48gb: "1"
|
|
limits:
|
|
memory: "2Gi"
|
|
cpu: "1000m"
|
|
nvidia.com/mig-3g.48gb: "1"
|
|
volumeMounts:
|
|
- name: asset-store
|
|
mountPath: /opt/assets
|
|
- name: model-cache
|
|
mountPath: /opt/models
|
|
readOnly: true
|
|
livenessProbe:
|
|
httpGet:
|
|
path: /health
|
|
port: 8443
|
|
initialDelaySeconds: 20
|
|
periodSeconds: 15
|
|
readinessProbe:
|
|
httpGet:
|
|
path: /health
|
|
port: 8443
|
|
initialDelaySeconds: 10
|
|
periodSeconds: 5
|
|
volumes:
|
|
- name: asset-store
|
|
persistentVolumeClaim:
|
|
claimName: pvc-asset-store
|
|
- name: model-cache
|
|
persistentVolumeClaim:
|
|
claimName: pvc-model-cache
|
|
|
|
---
|
|
# ── WebOS (Nginx static + React) ─────────────────────────────
|
|
apiVersion: apps/v1
|
|
kind: Deployment
|
|
metadata:
|
|
name: webos
|
|
namespace: velocity-os
|
|
labels:
|
|
app: webos
|
|
tier: frontend
|
|
spec:
|
|
replicas: 2
|
|
selector:
|
|
matchLabels:
|
|
app: webos
|
|
strategy:
|
|
type: RollingUpdate
|
|
rollingUpdate:
|
|
maxUnavailable: 0
|
|
maxSurge: 1
|
|
template:
|
|
metadata:
|
|
labels:
|
|
app: webos
|
|
tier: frontend
|
|
spec:
|
|
containers:
|
|
- name: webos
|
|
image: ${ECR_REGISTRY}/velocity-os/webos:latest
|
|
ports:
|
|
- containerPort: 80
|
|
resources:
|
|
requests:
|
|
memory: "128Mi"
|
|
cpu: "100m"
|
|
limits:
|
|
memory: "256Mi"
|
|
cpu: "250m"
|
|
livenessProbe:
|
|
httpGet:
|
|
path: /health.txt
|
|
port: 80
|
|
initialDelaySeconds: 5
|
|
periodSeconds: 10
|
|
|
|
---
|
|
# ── Media Engine (Dream Weaver Gateway) ──────────────────────
|
|
apiVersion: apps/v1
|
|
kind: Deployment
|
|
metadata:
|
|
name: media-engine
|
|
namespace: velocity-os
|
|
labels:
|
|
app: media-engine
|
|
tier: ai
|
|
spec:
|
|
replicas: 1
|
|
selector:
|
|
matchLabels:
|
|
app: media-engine
|
|
template:
|
|
metadata:
|
|
labels:
|
|
app: media-engine
|
|
tier: ai
|
|
spec:
|
|
# MIG slice 1: ComfyUI media generation
|
|
runtimeClassName: nvidia
|
|
containers:
|
|
- name: media-engine
|
|
image: ${ECR_REGISTRY}/velocity-os/media-engine:latest
|
|
ports:
|
|
- containerPort: 8290
|
|
envFrom:
|
|
- secretRef:
|
|
name: velocity-secrets
|
|
- configMapRef:
|
|
name: velocity-config
|
|
resources:
|
|
requests:
|
|
memory: "2Gi"
|
|
cpu: "1000m"
|
|
# RTX 6000 Blackwell MIG 3g.48gb (ComfyUI slice)
|
|
nvidia.com/mig-3g.48gb: "1"
|
|
limits:
|
|
memory: "4Gi"
|
|
cpu: "2000m"
|
|
nvidia.com/mig-3g.48gb: "1"
|
|
volumeMounts:
|
|
- name: model-cache
|
|
mountPath: /opt/models
|
|
readOnly: true
|
|
- name: asset-store
|
|
mountPath: /opt/assets
|
|
livenessProbe:
|
|
httpGet:
|
|
path: /health
|
|
port: 8290
|
|
initialDelaySeconds: 30
|
|
periodSeconds: 30
|
|
volumes:
|
|
- name: model-cache
|
|
persistentVolumeClaim:
|
|
claimName: pvc-model-cache
|
|
- name: asset-store
|
|
persistentVolumeClaim:
|
|
claimName: pvc-asset-store
|
|
|
|
---
|
|
# ── DB Init Job (runs once: schema apply + seed) ─────────────
|
|
apiVersion: batch/v1
|
|
kind: Job
|
|
metadata:
|
|
name: db-init
|
|
namespace: velocity-os
|
|
labels:
|
|
app: db-init
|
|
spec:
|
|
# Never auto-restart; operator re-runs manually if needed
|
|
backoffLimit: 0
|
|
template:
|
|
metadata:
|
|
labels:
|
|
app: db-init
|
|
spec:
|
|
restartPolicy: Never
|
|
initContainers:
|
|
# Wait for postgres to be ready before running init
|
|
- name: wait-for-postgres
|
|
image: ${ECR_REGISTRY}/postgres:15-alpine
|
|
command: [sh, -c, "until pg_isready -h postgres -U $(POSTGRES_USER); do echo waiting...; sleep 2; done"]
|
|
envFrom:
|
|
- secretRef:
|
|
name: velocity-secrets
|
|
containers:
|
|
- name: db-init
|
|
image: ${ECR_REGISTRY}/velocity-os/core:latest
|
|
command:
|
|
- sh
|
|
- -c
|
|
- |
|
|
echo "=== Applying schemas ==="
|
|
psql $DATABASE_URL -f /app/db/schema.sql
|
|
psql $DATABASE_URL -f /app/db/schema_addendum.sql
|
|
psql $DATABASE_URL -f /app/db/schema_comms.sql
|
|
psql $DATABASE_URL -f /app/db/schema_crm_canonical.sql
|
|
psql $DATABASE_URL -f /app/oracle/schema_oracle.sql
|
|
psql $DATABASE_URL -f /app/oracle/schema_extension_v2.sql
|
|
echo "=== Seeding synthetic CRM v2 ==="
|
|
python /app/scripts/seed_synthetic_crm.py
|
|
echo "=== DB init complete ==="
|
|
envFrom:
|
|
- secretRef:
|
|
name: velocity-secrets
|
|
- configMapRef:
|
|
name: velocity-config
|