Databricks Unity Catalog Edge Deployment —

Unity Catalog Edge

Databricks Unity Catalog Edge Deployment Data Governance Catalog Schema Access Control Lineage Audit Model Registry MLflow Delta Lake IoT Factory Retail

Unity Catalog Setup

=== Unity Catalog Configuration ===

SQL — Create Catalog and Schema

CREATE CATALOG IF NOT EXISTS production;

CREATE SCHEMA IF NOT EXISTS production.ml_models;

CREATE SCHEMA IF NOT EXISTS production.feature_store;

CREATE SCHEMA IF NOT EXISTS production.edge_metrics;

-- Grant permissions

GRANT USE CATALOG ON CATALOG production TO `data-engineers`;

GRANT USE SCHEMA ON SCHEMA production.ml_models TO `ml-engineers`;

GRANT SELECT ON SCHEMA production.feature_store TO `ml-engineers`;

GRANT CREATE TABLE ON SCHEMA production.edge_metrics TO `edge-devices`;

-- Row-level security

CREATE FUNCTION production.ml_models.region_filter(region STRING)

RETURN IF(IS_ACCOUNT_GROUP_MEMBER('global-admin'), true, region = current_user_region());

ALTER TABLE production.ml_models.predictions

SET ROW FILTER production.ml_models.region_filter ON (region);

-- Column masking

CREATE FUNCTION production.ml_models.mask_pii(val STRING)

RETURN IF(IS_ACCOUNT_GROUP_MEMBER('pii-access'), val, '***MASKED***');

ALTER TABLE production.feature_store.customers

ALTER COLUMN email SET MASK production.ml_models.mask_pii;

Python — Unity Catalog with MLflow

import mlflow

เนื้อหาเกี่ยวข้อง — ดูเพิ่มเติมเรื่อง Tailscale Mesh Stream Processing

from mlflow.models import infer_signature

mlflow.set_registry_uri("databricks-uc")

# Register model in Unity Catalog

with mlflow.start_run():

model = train_model(X_train, y_train)

signature = infer_signature(X_train, model.predict(X_train))

mlflow.sklearn.log_model(

แนะนำเพิ่มเติม — สัญญาณเทรดรายวัน XM Signal

model,

artifact_path="model",

registered_model_name="production.ml_models.edge_detector",

signature=signature,

input_example=X_train[:5],

)

# Set model alias

from mlflow import MlflowClient

client = MlflowClient()

client.set_registered_model_alias(

name="production.ml_models.edge_detector",

alias="champion",

version=3

)

from dataclasses import dataclass

@dataclass

class CatalogObject:

full_name: str

เนื้อหาเกี่ยวข้อง — ดูเพิ่มเติมเรื่อง Several การใช — คู่มือฉบับสมบูรณ์ 2026

object_type: str

owner: str

access: str

lineage: str

objects = [

CatalogObject("production.ml_models.edge_detector", "Model", "ml-team", "ml-engineers: USE", "training_data → model"),

CatalogObject("production.feature_store.user_features", "Table", "data-team", "ml-engineers: SELECT", "raw_events → features"),

CatalogObject("production.edge_metrics.predictions", "Table", "edge-service", "analysts: SELECT", "edge_device → metrics"),

CatalogObject("production.ml_models.preprocessing", "Function", "ml-team", "ml-engineers: EXECUTE", "feature_store → transform"),

]

แนะนำเพิ่มเติม — ระบบเทรดของ iCafeForex

Edge Deployment Pipeline

=== Edge Model Deployment ===

Deployment Pipeline

1. Train model in Databricks

2. Register in Unity Catalog

3. Set alias "champion"

4. Export to ONNX/TensorRT

5. Push to Edge Registry (Harbor/ECR)

6. Deploy to Edge devices via GitOps

7. Monitor metrics → send back to Unity Catalog

เนื้อหาเกี่ยวข้อง — ดูเพิ่มเติมเรื่อง Java Micronaut DNS Management — คู่มือฉบับสมบูรณ์ 2026

Edge Deployment Script

import mlflow

import onnx

import docker

client = MlflowClient()

model_version = client.get_model_version_by_alias(model_name, alias)

model_uri = f"models:/{model_name}@{alias}"

# Export to ONNX

model = mlflow.pyfunc.load_model(model_uri)

onnx_path = export_to_onnx(model, "model.onnx")

# Build edge container

docker_client = docker.from_env()

image = docker_client.images.build(

path="./edge-container",

tag=f"edge-model:{model_version.version}",

buildargs={"MODEL_PATH": onnx_path}

)

# Push to registry

docker_client.images.push(f"registry/edge-model:{model_version.version}")

# Deploy to edge devices

deploy_to_device(device, f"edge-model:{model_version.version}")

log_deployment(model_name, model_version.version, device)

@dataclass

เนื้อหาเกี่ยวข้อง — แนะนำให้อ่าน Snyk Code Security Identity Access Management

class EdgeDevice:

device_id: str

location: str

model_version: str

status: str

latency_ms: float

accuracy: float

last_sync: str

devices = [

EdgeDevice("edge-001", "Factory A - Line 1", "v3.2", "Running", 12.5, 0.965, "2 min ago"),

EdgeDevice("edge-002", "Factory A - Line 2", "v3.2", "Running", 14.2, 0.958, "1 min ago"),

EdgeDevice("edge-003", "Factory B - Line 1", "v3.1", "Updating", 0, 0, "Deploying v3.2"),

EdgeDevice("edge-004", "Retail Store 1", "v3.2", "Running", 18.7, 0.942, "5 min ago"),

EdgeDevice("edge-005", "Retail Store 2", "v3.2", "Offline", 0, 0, "Last seen 2h ago"),

]

if d.status == "Running":

เคล็ดลับ

• Namespace: ใช้ Catalog.Schema.Object จัดระเบียบ Data
• Alias: ใช้ Model Alias champion/challenger แทน Version Number
• Lineage: เปิด Lineage ติดตามที่มาข้อมูลอัตโนมัติ
• Monitor: ส่ง Edge Metrics กลับ Unity Catalog ทุก Inference
• Rollback: เตรียม Rollback Plan เสมอเมื่อ Deploy Model ใหม่

Unity Catalog คืออะไร

Unified Governance Databricks Data Assets Tables Views Functions Models Catalog Schema Object GRANT REVOKE Lineage Audit Delta Lake Sharing