databricks Platform · Updated Jun 12, 2026

Faster Inference Tables for Custom Model Serving Endpoints

Enhanced telemetry for real-time data ingestion in model serving.

What it does

Inference tables for custom model serving endpoints now utilize endpoint telemetry to quickly ingest request and response payloads into Unity Catalog tables. This feature allows for configurable sampling and ensures that existing inference tables remain unaffected. The telemetry data can be used for root cause analysis, monitoring endpoint health, and compliance reporting through standard SQL queries.

Why it matters

This update significantly improves the speed and efficiency of data ingestion for model serving, enabling organizations to gain real-time insights into model performance and usage. By leveraging telemetry data, data teams can enhance monitoring, improve compliance, and optimize model performance, ultimately driving better business decisions and outcomes.

Key capabilities

Real-time Data Ingestion

Telemetry data is ingested into Unity Catalog tables within seconds.

Configurable Sampling

Users can configure sampling options for telemetry data collection.

Root Cause Analysis

Persisted telemetry data aids in performing root cause analysis and monitoring.

Compliance Support

Standard SQL queries can be used to meet compliance requirements.

Prerequisites

Workspace must be enabled for Unity Catalog.

Permissions for USE CATALOG, USE SCHEMA, CREATE TABLE, and MODIFY on the destination Unity Catalog.

An existing custom model serving endpoint or permissions to create one.

Workspace must be in a supported region.

Recommended use cases

Monitoring model performance in real-time.Conducting root cause analysis for inference failures.Ensuring compliance with data governance standards.Optimizing model serving based on telemetry insights.

Limitations · remember

!Schema evolution on the target table is not supported.

!Only managed Delta tables are supported; external storage is not.

!The table location must be in the same region as the workspace.

!Only ASCII letters, digits, and underscores are supported in table names.

Architecture

Custom Model Serving Endpoint → AI Gateway → Telemetry Data Ingestion → Unity Catalog

Get started

Instrument Model Code

Add logging to your model to capture telemetry data.

Prepare Unity Catalog

Ensure a catalog and schema are ready to receive telemetry data.

Enable Telemetry

Select 'Enable inference tables and telemetry' in the endpoint UI.

Verify Data Flow

Check the Unity Catalog tables to ensure data is being ingested.

Query Telemetry Data

Use SQL to query the telemetry data for insights.

Key takeaway

The new inference tables enhance real-time data ingestion and monitoring capabilities for custom model serving, driving better insights and compliance.

2026-06-12-faster-inference-tables-for-custom-model-serving-endpoints Generated by the Databricks Updates Agent · browse all updates