Tag: resilience

A DelegatingHandler registered before AddStandardResilienceHandler runs once per logical request, not once per retry. Move it after the resilience handler, and make it idempotent, because the retry re-sends the same HttpRequestMessage. Measured on Microsoft.Extensions.Http.Resilience 10.10.0 and 8.10.0.

StackExchange.Redis 3.3.0 adds opt-in maintenance notifications (smart client handoffs) for Redis Enterprise, Redis Cloud and Azure Managed Redis: relaxed timeouts during migrations, topology re-reads, and a proactive move before an endpoint goes away. Here is how to turn it on and the SER010 error you will hit first.

Polly 8.8.0 adds EnableReloadsWithMonitor, so a resilience pipeline can hot-reload from a feature flag or remote config monitor that is not registered in DI. A probe shows the rebuild, and one trap: context.GetOptions still reads the DI monitor and returns defaults.

Microsoft.Extensions.AI 10.9.0 adds RoutingChatClient, OrderedFailoverChatClient, and SemanticRoutingChatClient. Verified against the real package: what fails over, what does not, and why MEAI001 breaks your build.

Aspire's AddServiceDefaults applies a standard resilience handler to every HttpClient. Calling AddStandardResilienceHandler again stacks a second one instead of replacing it. Here are the three real override paths, the -standard options name nobody documents, and the infinite timeout you inherit if you just remove it.

Use the Microsoft.Extensions.Http.Resilience handler for HttpClient calls, since it is Polly with HTTP-aware defaults and telemetry in one line. Reach for Polly's ResiliencePipeline directly only when you protect something that is not an HttpClient.