A secure cloud load-balancing system routes traffic across server clusters, with analytics dashboards below.
Azure Application Gateway now has two public-preview features aimed at teams that outgrew simple path-based routing: advanced routing rules and advanced load balancing. They can be used separately or together. The Azure Updates entry is dated November 17, 2026. The underlying Microsoft Tech Community post, by product managers Jay Soni and Jordi Rey, ran on November 12. One independent write-up gives the announcement date as November 11, 2026, so the exact day varies by source. The substance does not.

A secure cloud load-balancing system routes traffic across server clusters, with analytics dashboards below. The division of labor​

The two features split the work cleanly:

  • Advanced routing rules decide which requests qualify for a route.
  • Advanced load balancing decides how traffic is spread across backend pools, and across servers inside each pool.

Application Gateway is a Layer 7 service. Microsoft's Learn overview describes it as a web traffic load balancer that routes on HTTP attributes such as URL paths and host headers, not just IP and port. This update is about web requests, not general TCP/UDP balancing.

Advanced routing rules​

Microsoft says a rule can match on five request attributes:

  • URL path
  • HTTP header
  • query string
  • HTTP method
  • client IP address

Matches can be exact. Path, header and query-string conditions also accept RE2 regular expressions. Microsoft's example pattern matches /api/orders and everything beneath it with a single expression.

A rule can chain up to five conditions, joined with AND. Rules are evaluated in priority order, and the first match decides the destination. A request that matches nothing goes to the configured default backend pool. A rule can also carry a rewrite rule set.

Microsoft Learn adds more detail. The routing-rules page lists Advanced as a rule type next to Basic and Path-based, chosen when you want decisions based on HTTP request attributes. The documentation also gives these limits:

  • 100 advanced routing maps per gateway
  • 100 advanced rules per map
  • 10,000 condition sets per gateway
  • 5 values per condition
  • 128 characters per condition or regex value

Documented method values are GET, HEAD, POST, PUT, DELETE, CONNECT, OPTIONS and PATCH. Method matching is exact and case-sensitive. Path matching normalizes and decodes the URL before it compares.

Exclusions that matter​

Learn's routing-rules page flags these limits:

  • Advanced rules are in preview.
  • They are not supported for Multisite listeners.
  • They are not supported for the Layer 4 TCP/TLS proxy.

Per the Learn documentation, advanced rules attach only to a Basic frontend listener. Many production gateways host several sites on Multisite listeners, so this one will rule out some deployments before anyone writes a condition.

Two other documented behaviors are worth knowing:

  • WAF policies are evaluated before routing, so blocked requests never reach the routing engine.
  • WAF association is at gateway or listener level, not per advanced rule.

Learn also separates the Application Gateway rule's priority from the priorities used inside an advanced routing map. Check which priority field you are editing.

Advanced load balancing​

Selection happens at two levels.

Across backend pools:

  • Weighted distribution splits traffic by relative weight, such as 80/20.
  • Priority-based failover keeps traffic on a primary pool. It moves to the next pool when the primary's share of healthy servers falls below a configurable Minimum Healthy Percent.

Within a pool:

  • Round robin spreads requests evenly.
  • Weighted round robin sends more requests to larger servers.
  • Least connections sends each request to the healthy server with the fewest active connections, so a server stuck on long-running requests gets less new traffic.

The algorithm is set per pool.

Scenarios Microsoft proposes​

Microsoft lists several configurations:

  • Canary or blue-green releases. Route on an x-release: canary header, or send 10% of traffic to a green pool and raise its weight over time. Pulling a pool out is a weight change, not a redeployment.
  • Read/write split. GETs go to read replicas or cache-heavy nodes. POST, PUT and DELETE go to the primary write tier.
  • Client-aware routing. Mobile clients go to an optimized backend, based on a header or a User-Agent pattern.
  • Internal and partner traffic. Corporate or partner IP ranges go to dedicated pools. Microsoft pairs this with WAF protection and proper authentication and authorization.
  • Multi-tenant SaaS. Tenant hostnames map to premium or shared pools.
  • API versioning. Route on an api-version query string or header while clients keep a stable URL.
  • AI endpoints. Reserved Provisioned Throughput Unit (PTU) endpoints go in a priority 1 pool. Pay-as-you-go endpoints go in priority 2. Traffic spills over only when the healthy share of reserved endpoints drops below the threshold.

These are possible designs, not built-in guarantees. Routing by client IP or tenant hostname does not provide authorization or tenant isolation. Microsoft itself says to layer WAF and authentication on top. The AI-spillover pattern comes with no published performance, cost or reliability figures.

The caveats​

  1. API version. Both features need API version 2026-03-01 or later. Older templates, SDKs and automation may need updating.
  2. Weighted distribution ignores health. During preview, weighted distribution does not check pool health. To remove a pool from rotation, set its weight to 0. Do not treat a weighted split as health-aware failover. Priority-based failover is the mode driven by the healthy-percentage threshold.
  3. No SLA. Microsoft says previews carry no service-level agreement and are not recommended for production workloads.

A separate community write-up advises verifying rule priority, default destinations and failure behavior, including the weighted health limitation. That is common-sense advice rather than Microsoft guidance.

A practical test plan​

This is my own suggested approach, not Microsoft's procedure:

  • Confirm the listener is a Basic frontend listener, not Multisite.
  • Confirm your deployment tooling targets API version 2026-03-01 or later.
  • Test a rule's priority order with requests that deliberately match more than one rule.
  • Send a request that matches no rule and confirm the default pool, HTTP settings and optional rewrite apply.
  • For weighted splits, take a backend down and confirm what happens, since the gateway will not remove it for you.
  • For priority failover, test degraded states around the Minimum Healthy Percent threshold.
  • Send a deliberately unauthorized request to any IP- or tenant-based route and confirm the application still rejects it.

Bottom line​

This is a networking feature for Azure administrators, not a Windows change. The preview makes Application Gateway more expressive for canary releases, API versioning and AI capacity management. The listener restriction, the API version requirement and the missing health checks on weighted splits all argue for testing in a non-production gateway for now.

 

References

  1. (In preview) Public Preview: Advanced load balancing in Azure Application Gateway Azure Updates 2026-11-17T20:00:32Z
  2. Azure Application Gateway request routing rules configuration | Microsoft Learn learn.microsoft.com
  3. What is Azure Application Gateway | Microsoft Learn learn.microsoft.com