Takeaway: Azure Chaos Studio helps organizations validate software resilience by simulating outages, failovers, community disruptions, and infrastructure failures earlier than they impression manufacturing.
You don’t know with certainty that your software is resilient till that resilience is examined. Higher to study it isn’t by intentionally breaking it in a check atmosphere and watching the way it reacts, than by a failure in manufacturing. Azure Chaos Studio is our managed service for doing precisely that, safely and on goal.
At the moment, Azure Chaos Studio Workspaces is in public preview: a scenario-focused strategy that allows you to check the failure modes Azure clients really see in manufacturing. We’ve been exhausting at work making Workspaces simple to make use of, with broad fault help and named eventualities that mirror actual outages, as a substitute of remoted faults.
Why designing for resilience isn’t sufficient
Azure clients have invested in resilient design: multi-zone deployments, geo-redundant storage, automated database failover, retry logic, load-balanced entrance ends. Nevertheless, the true query is when an incident begins: when the failure arrives, do these mechanisms recuperate your software within the time you assumed they’d?
Actual outages don’t learn the structure diagram. A zone-redundant deployment can fail as a result of a well being probe was misconfigured years in the past. A database with automated failover can go away the appliance lifeless as a result of a connection string is difficult coded to a single area. Geo-redundant storage can briefly produce stale reads the appliance code by no means anticipated. These errors are widespread, and so they solely present up when the failure occurs.
Reliability and resiliency on Azure are a shared accountability. Microsoft is liable for the platform and the resilience constructed into Azure companies. Prospects are liable for configuring that resilience and the code that makes use of it. No layer makes up for a niche in one other. The one option to know whether or not your structure, configuration, and software logic will maintain up in manufacturing is to show they maintain below failure earlier than an outage assessments them for you.
How Chaos Studio Workspaces adjustments resilience testing
Chaos Studio is Azure’s managed chaos engineering service for validating how purposes behave below failure. By simulating managed disruptions throughout infrastructure, networking, databases, and software dependencies, it helps groups uncover resilience gaps earlier than clients expertise them. Chaos Studio Workspaces focuses on eventualities that match what occurs in manufacturing, so that you begin from an actual outage sample as a substitute of assembling particular person faults. You start with a named situation like Zone Down, DNS Outage, or SQL failover, already sequenced in opposition to the sources in a Workspace.
Most outages train two layers without delay. There’s the platform layer: did the service come again, did failover full inside your Restoration Time Goal, did visitors reroute. And there’s the appliance layer: did your code keep information integrity, choose up in-flight transactions, retry the precise issues, degrade gracefully. A chaos check that solely stops a Digital Machine (VM) tells you in regards to the platform layer. The eventualities in Chaos Studio Workspaces are designed to validate your entire stack.
Workspaces scale back the burden of getting began. The commonest purpose resilience testing stalls is that groups don’t know the place to start out. The Workspace is the brand new top-level useful resource: you level it at a subscription or useful resource group, and its managed identification discovers what’s in scope and recommends the eventualities that apply. These eventualities present up contained in the Workspace, able to configure and run, and a refresh, updates the suggestions every time your infrastructure adjustments.
A library of actual outage eventualities. Chaos Studio Workspaces ships with curated eventualities knowledgeable by patterns noticed in actual Azure incidents, so the patterns you check in opposition to are the patterns clients really expertise. Consider these as resilience templates, a quick path to the failure modes most groups want to check, and whenever you want one thing totally different, design your individual from the identical fault library.
Accessible at present:
Availability Zone Down: Digital Machine Scale Units (VMSS) shutdown with per-zone focusing on to validate cross-zone routing and restoration.
Availability Zone Down and Database failover: Compute Zone Down composed with Azure Database for PostgreSQL (Versatile Server) failover, to look at failover conduct in opposition to your configured restoration goals and application-side connection dealing with.
DNS Outage: a full DNS decision outage by way of NSG guidelines that block resolver visitors, to validate how your software behaves when title decision fails.
Microsoft Entra ID Outage: identity-provider failure that workouts authentication retry, token caching, and fallback paths.
Cache Stampede: Redis flush mixed with database restart and an App Service course of crash, to validate conduct below a cache-miss storm and the ensuing database surge. The App Service process-crash variant presently helps Home windows App Service plans.
Occasion-Pushed Messaging Disruption: Azure Service Bus and Occasion Hubs disable, to validate dead-letter dealing with and backpressure.
Behind each situation are granular API-level actions constructed for Workspaces:
Every situation composes the precise faults routinely. And when a curated situation doesn’t match your workload, you’ll be able to construct your individual. The brand new Situation Designer is a drag-and-drop expertise within the Azure portal for composing any of those faults right into a customized situation arranging steps, branches, and faults with the identical flexibility as traditional Chaos Studio experiments, now accessible straight inside Workspaces. Begin with a curated template, or design from scratch utilizing the complete fault library.
VM agent faults reminiscent of Central Processing Unit (CPU) and reminiscence strain additionally run in Workspaces. Every situation sequences the precise mixture of faults routinely, so working Zone Down + Database Failover doesn’t imply pondering by way of “shut down VMSS situations in zone 1, then force-failover the database major.” The library will continue to grow by means of public preview and into GA, with plans to discover extra eventualities over time, reminiscent of:
Storage account failover
Microsoft Azure SQL Managed Occasion failover
Microsoft Azure Entrance Door and Microsoft Azure Utility Gateway
Partial zone degradation
Microsoft Azure Kubernetes Service (AKS)-native pod chaos
Buyer-observed area down
That very same basis can also be related for AI purposes shifting into manufacturing. Copilots, brokers, retrieval-augmented technology pipelines, and inference endpoints might introduce new AI-specific failure modes, however they nonetheless depend on the identical Azure constructing blocks as different distributed purposes: compute, databases, caches, search indexes, identification, networking, messaging, and storage. Chaos Studio Workspaces can validate that basis at present by means of eventualities like Zone Down, Database Failover, DNS Outage, Cache Stampede, and Occasion-Pushed Messaging Disruption, whereas the catalog continues to evolve towards AI-specific behaviors reminiscent of retrieval drift, token throttling, and mannequin conduct shifts below load as extra insights are gathered fromworking carefully with clients constructing AI on Azure.
Situation reviews. When a run finishes, Chaos Studio Workspaces generates a structured drill report. It lays out what the situation injected, which sources it affected, how the restoration timeline performed out, which alerts have been attributable to the drill versus the conventional baseline, and the place the workload behaved in a different way than anticipated. The report reads like an inside post-incident evaluation, which makes it helpful each for the staff that ran the drill and for the leaders who need to see resilience being validated frequently. Groups can export it and connect it to alter tickets, audit proof, or service well being evaluations.
Bringing resilience testing into AI-powered operations
Alongside the product, we’re transport two methods to drive Chaos Studio from the instruments engineers already work in. The primary is the Chaos Studio Talent for GitHub Copilot: it walks you thru the entire loop in a dialog. Level a Workspace at a subscription, see the eventualities it recommends, run a drill, and get again a report of what really occurred, correlated in opposition to your Azure Monitor alerts.
The second is an Mannequin Context Protocol (MCP) server that exposes the identical Chaos Studio operations as typed instruments, so different assistants and autonomous brokers: Claude, Cursor, Codex, or your individual, can provision a Workspace, run a situation, and question the alerts round it and not using a individual within the loop. Each run in opposition to the identical Chaos Studio APIs and your individual Azure sign-in, and you’ll attempt them at present.
We’re transport this on day one for one purpose: When a buyer asks an AI assistant about Chaos Studio, the expertise must be formed by us, not improvised by a big language mannequin (LLM) studying our REST API. In our expertise, one of many hardest components of resilience testing is usually deciding to run the drill within the first place, and that call more and more lives within the chat instruments engineers already use, so this must stay there too.
The place that is headed: The Talent turning into a step inside automated operations flows on Microsoft Foundry, and one of many methods an Azure SRE agent validates its personal assumptions about how a workload fails. Attempt it and inform us what’s lacking; we’ll shut the gaps by means of public preview.
Get began
Azure Chaos Studio Workspaces is in public preview at present. Normal availability is presently focused for late 2026, topic to alter.
To begin:
Create a Workspace scoped to a subscription or useful resource group you need to check.
Let discovery populate the advisable eventualities for the sources it finds. Desire to construct your individual? Open the Situation Designer and compose a customized situation from the fault library, no scripting required.
Run your first drill. When you’ve by no means run a chaos experiment, run Zone Down. A full availability-zone failure surfaces how compute placement, database failover, DNS decision, and application-layer retry logic reply below stress. In case your workload recovers inside an appropriate time, you’ve gained proof about the way it responds to one of the vital widespread causes of prolonged cloud downtime. If it doesn’t, you’ve discovered the hole in your phrases as a substitute of your clients’.
Resilience isn’t one thing a single characteristic, a single redundancy mechanism, or a single structure determination gives you. It’s an engineering self-discipline, and the self-discipline requires verification. Azure Chaos Studio Workspaces is how we’re making that verification the default for Azure workloads, together with the AI workloads extra of our clients are placing into manufacturing.
