AI 热点

Databricks 开放 Genie ZeroOps Pipeline Agent 公开测试

Unite.AI更新于 11月20日 06:27

Databricks Opens Public Beta for Genie ZeroOps Pipeline Agent

AI 导读

Databricks 于 2026 年 11 月 19 日开放 Genie ZeroOps 公开测试,这款数据管道智能体可自动检测故障、调查原因并准备修复方案。它能汇集日志、源代码、表元数据与血缘信息,使用 SQL 定位根因,并在隔离沙箱中通过单元测试和 dry run 验证修复。ZeroOps 未经批准不会更改生产代码或数据,但会持续使用大语言模型处理所选资产,当前监控范围为作业、管道和表。

推荐理由

AI 评分 69

公开测试展示了从故障检测到受控修复的完整流程,也明确列出持续 AI 处理、共享权限与外部通知的企业部署取舍。

主题

标签

正文

Databricks opened Genie ZeroOps to public beta on November 19, 2026, after private-preview testing of its autonomous data-pipeline agent. The company says ZeroOps observed more than 80,000 real customer issues during that limited preview.

ZeroOps detects failures, investigates their causes and prepares fixes. The beta adds shared inboxes, notifications that summarize each issue and recommendation, custom instructions for company-specific workflows, and one-click pull-request creation for Git-backed jobs.

Private-preview testing covered hundreds of companies, and Databricks reports that more than half of the issues required organization-specific knowledge to triage, rank and resolve properly. That finding drove features that let teams identify known flaky sources, preferred fixes and naming conventions, or filter expected failures and lower-priority notifications.

From Detection to Reviewed Fixes

ZeroOps watches for explicit job and pipeline failures as well as tables that stop updating or arrive with unexpectedly few rows, problems that may not trigger a conventional job failure. It gathers logs, source code, table metadata and lineage, runs SQL queries to identify root causes, and prioritizes findings by downstream impact. Proposed fixes are tested with unit tests and dry runs in an isolated sandbox before anyone applies them.

Investigations start automatically when an issue is detected. Built-in access to run history, data-quality signals and Unity Catalog lineage, which maps dependencies between assets, connects code diagnostics to the state of the data. Alerts can go to Slack, Microsoft Teams, PagerDuty or generic webhooks; email destinations are unsupported.

Approval and Data Controls

ZeroOps does not change production code or data without approval. The documented guardrails require an explicit user instruction and confirmation before it pushes a Git-backed fix; the push adds a commit without force-pushing or rewriting branch history. Naming a repository’s default branch can bypass pull-request review unless branch protection blocks direct pushes.

Databricks says the agent uses Anthropic and OpenAI models on Databricks-hosted endpoints inside its security perimeter. Selecting assets authorizes continuous, autonomous large-language-model processing until the feature is disabled, including repository contents for Git-backed jobs. Before an external notification includes a finding’s title or summary, the agent checks for sensitive information such as names and email addresses and masks what it finds. Copies stored in outside systems fall outside Databricks’ retention and deletion controls.

A configuration can be shared with up to 25 individual workspace users, but not with groups or service principals. Collaborators can work on its findings, yet shared users see code excerpts, suggested fixes and sample data rows even without access to the underlying assets. Sharing does not grant access to those assets.

Setup Requirements and Beta Scope

An administrator must enable Genie ZeroOps from the workspace Previews page. Users then select assets and need OWNER, CAN MANAGE or CAN VIEW permission on them. The setup process backfills findings from the previous 24 hours before reporting new issues. Investigating a Git repository requires a credential with read access; creating a pull-request branch requires write access.

Databricks introduced ZeroOps on June 16, 2026 with planned private-preview support for jobs, pipelines, tables and machine-learning workloads. Current documentation states that monitoring covers jobs, pipelines and tables, not models or apps, and that performance and cost regressions are outside its detection scope. The beta operates at workspace rather than account level, is not HIPAA compliant, and requires separately billed data-quality monitoring to detect late-arriving or incomplete table data.

ZeroOps automates investigation and fix preparation while leaving production authority with the customer. The beta’s central enterprise trade-off is less manual diagnosis in exchange for continuous AI processing of selected operational data, making sharing visibility, external notifications and Git review part of deployment rather than separate administrative details.

来源

Unite.AI