Stratara.Mediator
4.4.2
Prefix Reserved
dotnet add package Stratara.Mediator --version 4.4.2
NuGet\Install-Package Stratara.Mediator -Version 4.4.2
<PackageReference Include="Stratara.Mediator" Version="4.4.2" />
<PackageVersion Include="Stratara.Mediator" Version="4.4.2" />
<PackageReference Include="Stratara.Mediator" />
paket add Stratara.Mediator --version 4.4.2
#r "nuget: Stratara.Mediator, 4.4.2"
#:package Stratara.Mediator@4.4.2
#addin nuget:?package=Stratara.Mediator&version=4.4.2
#tool nuget:?package=Stratara.Mediator&version=4.4.2
Stratara.Mediator
Derived. The behaviour described here is specified under
openspec/specs/. Those specifications are the source; this page explains and illustrates them.
License: MIT.
In-process mediator with DI-resolved handlers and pipeline behaviors. Drop-in replacement for MediatR-style routing without the runtime cost of MethodInfo.Invoke — uses a typed wrapper cache and direct DI dispatch.
Quick start
services.AddMediator()
.AddCommandHandlersFromAssemblyContaining<Program>()
.AddQueryHandlersFromAssemblyContaining<Program>()
.AddPipelineBehaviorWithResult(typeof(LoggingBehavior<,>))
.AddPipelineBehavior(typeof(LoggingBehavior<>));
// Optional: wrap in authorization decorator
services.AddAuthorizingMediator<MyAuthorizationProvider>();
Every dispatch is traced as a span named Handle <RequestType>. The spans come from the
Stratara.Application activity source — subscribe to it and they appear; register your own
OpenTelemetry Tracer and the mediator uses that instead. Nothing has to be registered for the
mediator to start.
IMediator is registered scoped. Resolve it from a scope (a request scope in ASP.NET Core, or
an explicit IServiceProvider.CreateScope() in a console host) — resolving it from the root
provider throws.
What's in the box
IMediator.HandleAsync<TResult>(IRequest<TResult>, CancellationToken)— routes queries and commands-with-result toIQueryHandler<TRequest, TResult>through any registeredIPipelineBehavior<TRequest, TResult>chain.IMediator.HandleAsync<TRequest>(TRequest, CancellationToken)— routes void commands toICommandHandler<TRequest>through any registeredIPipelineBehavior<TRequest>chain.AuthorizingMediatordecorator — checks[RequireRole]attributes viaIAuthorizationProviderand[RequirePermission]attributes viaIPermissionResolver(both AND) on the request's runtime type before delegating to the inner mediator. Its startup validator fails fast when a permission-guarded type is registered without an authorizing mediator or without a resolver, so a guard can never be silently skipped.BucketLockPool— concurrency primitive that serialisesIAggregateScopedCommanddispatch per bucket id. Used by message-bus consumers (e.g. theMediatorCommandWorkerinStratara.Outbox.RabbitMQ) to keep aggregate writes single-writer.
Pipeline behavior contract
Behaviors run outer-to-inner in DI registration order:
public sealed class LoggingBehavior<TRequest, TResult> : IPipelineBehavior<TRequest, TResult>
where TRequest : IRequest<TResult>
{
public async Task<TResult> HandleAsync(
TRequest request, Func<Task<TResult>> next, CancellationToken cancellationToken)
{
// before
var result = await next();
// after
return result;
}
}
Tenant isolation
AddStrataraTenantIsolation() registers a pipeline behavior that enforces tenant isolation at the
mediator entrance — before the handler runs — for any request that opts in by implementing the
ITenantScopedRequest marker. Requests that do not implement the marker pass through untouched.
public sealed record GetCustomerQuery(Guid CustomerId, Guid TenantId)
: IQuery<CustomerDto>, ITenantScopedRequest;
services
.AddStrataraValidation() // validation stays outermost
.AddStrataraTenantIsolation(); // then tenant isolation
The behavior compares the request's TenantId (the data owner) against the ambient session's
data-owner tenant (SessionContext.TenantId), not the actor tenant (SessionContext.ActorTenantId).
A request whose payload names a different tenant than the established session subject is rejected with
TenantAccessDeniedException (translated to HTTP 403 on ASP.NET hosts that register
AddStrataraProblemDetails() from Stratara.ServiceDefaults.AspNetCore, or 401 for a caller that is not authenticated; surfaced through the
message-failure path on workers).
Default vs. strict mode
TenantIsolationMode.Default— enforces only the subject match. A privileged cross-tenant operation (actor tenant ≠ data-owner tenant) passes, because the calling endpoint is expected to have promoted the session's data-owner tenant to the target before dispatch.TenantIsolationMode.Strict— additionally routes every cross-tenant operation through anICrossTenantAuthorizer. Stratara registers a deny-all default (viaTryAdd), so strict mode rejects all cross-tenant access until you register your own authorizer that grants it:
services.AddStrataraTenantIsolation(o => o.Mode = TenantIsolationMode.Strict);
services.AddScoped<ICrossTenantAuthorizer, PlatformAdminCrossTenantAuthorizer>();
internal sealed class PlatformAdminCrossTenantAuthorizer(IHttpContextAccessor http)
: ICrossTenantAuthorizer
{
public ValueTask<bool> IsCrossTenantAllowedAsync(SessionContext session, CancellationToken ct) =>
ValueTask.FromResult(http.HttpContext?.User.IsInRole("PlatformAdmin") ?? false);
}
The behavior runs both in-process (queries via
IMediatorat the endpoint, whereHttpContextis available) and worker-side (commands dispatched through the outbox, where there is noHttpContext). AnICrossTenantAuthorizerthat needs request-role state should be applied on the in-process path; the worker path must base its decision on theSessionContextalone.
Dependencies
Stratara.Abstractions— forIMediator/IRequest/ICommand/IQuery/IPipelineBehaviorcontracts, plusITenantScopedRequest/ICrossTenantAuthorizer/TenantAccessDeniedException.Stratara.Diagnostics— log-event IDs for the tenant-isolation behavior.Microsoft.Extensions.DependencyInjection.Abstractions.Microsoft.Extensions.Logging.Abstractions.OpenTelemetry.Api— emits anActivityper dispatch under theStratara.Applicationsource.
No EF Core, no message bus, no event sourcing. Library-safe.
| Product | Versions Compatible and additional computed target framework versions. |
|---|---|
| .NET | net10.0 is compatible. net10.0-android was computed. net10.0-browser was computed. net10.0-ios was computed. net10.0-maccatalyst was computed. net10.0-macos was computed. net10.0-tvos was computed. net10.0-windows was computed. |
-
net10.0
- JetBrains.Annotations (>= 2026.2.0)
- Microsoft.Extensions.DependencyInjection.Abstractions (>= 10.0.11)
- Microsoft.Extensions.Hosting.Abstractions (>= 10.0.11)
- Microsoft.Extensions.Logging.Abstractions (>= 10.0.11)
- Microsoft.Extensions.Options (>= 10.0.11)
- OpenTelemetry.Api (>= 1.18.0)
- Stratara.Abstractions (>= 4.4.2)
- Stratara.Diagnostics (>= 4.4.2)
NuGet packages (3)
Showing the top 3 NuGet packages that depend on Stratara.Mediator:
| Package | Downloads |
|---|---|
|
Stratara.Validation
Vendor-neutral request validation for the Stratara framework — a mediator pipeline behavior that runs IValidator<T> implementations before the handler and throws an aggregated StrataraValidationException on failure. No FluentValidation dependency; an optional adapter is shipped separately. |
|
|
Stratara.Outbox.RabbitMQ
Outbox-pattern command and event dispatch for the Stratara event-sourced stack — RabbitMQ IMessageBus implementation, retry worker, mediator command worker, and Redis-coordinated projection-replay state. Azure Service Bus support ships as the sibling Stratara.Outbox.AzureServiceBus package. |
|
|
Stratara.Infrastructure
Infrastructure glue for the Stratara framework — authorization decorators, configuration providers, and DI composition helpers that wire Mediator, Outbox, Identity, and EF Core into a hosted app. |
GitHub repositories
This package is not used by any popular GitHub repositories.
| Version | Downloads | Last Updated |
|---|---|---|
| 4.4.2 | 47 | 10/2/2026 |
| 4.4.1 | 384 | 9/29/2026 |
| 4.4.0 | 179 | 9/27/2026 |
| 4.3.1 | 163 | 9/25/2026 |
| 4.3.0 | 431 | 9/23/2026 |
| 4.2.0 | 195 | 9/18/2026 |
| 4.1.1 | 134 | 9/16/2026 |
| 4.1.0 | 135 | 9/16/2026 |
| 4.0.4 | 968 | 9/14/2026 |
| 4.0.3 | 224 | 9/3/2026 |
| 4.0.2 | 655 | 9/3/2026 |
| 4.0.1 | 328 | 9/2/2026 |
| 4.0.0 | 683 | 8/31/2026 |
| 4.0.0-preview.1 | 89 | 8/31/2026 |
| 3.4.0 | 176 | 8/28/2026 |
| 3.3.0 | 378 | 8/25/2026 |
| 3.2.3 | 165 | 8/22/2026 |
| 3.2.2 | 777 | 8/14/2026 |
| 3.2.1 | 416 | 8/2/2026 |
| 3.2.0 | 167 | 7/18/2026 |
A release about a flag that cost too much. Every host sharing the replay coordination state over Redis
asked Redis, with a blocking round trip on the calling thread, whether a replay was active — and the
Orleans execution model asked on every hot path, eight to ten times per command on a host with three
projections, inside grain turns. On a two-core host that was enough to starve the thread pool and time
the Redis connection out, surfacing as failed catch-ups, lagging read models and failed grain
registrations although no replay ever ran. The answer now comes from memory. No migration is needed;
one optional setting is new.
### Fixed
- **No path waits on Redis to learn whether a replay is active.** The Redis-backed replay state answers
`IsReplayActive` from a field. A change made in the host itself is seen there at once; a change made
elsewhere — a replay that began or ended in another host, or a marking that lapsed — is announced over
a channel beside the replay-request channel and seen within its latency, and every host re-reads the
marking every `ProjectionReplay:RefreshSeconds` regardless, as the safety net for a lost announcement.
The Orleans execution model's command dispatcher, bundle dispatcher, store readers and drain, and the
bus dispatchers, make no Redis call for it any more. Nothing changes for a host without Redis, whose
state was in process already, nor for a host that registers its own `IProjectionReplayState`.
### Added
- **`ProjectionReplayOptions.RefreshSeconds`** (default 5), read from the `ProjectionReplay` section
like the lease. It bounds how long a host answers from a replay marking that changed elsewhere where the
announcement was lost. Refused when the host starts at zero or below, and at or above
`LeaseSeconds`, with an `OptionsValidationException` naming `ProjectionReplay:RefreshSeconds`.
- **Log events `104_015` to `104_018`.** A host that cannot refresh the replay marking from Redis says
so once at Warning (`104_015`) and keeps its last answer; `104_016` at Information says it reads the
marking again. A host that cannot subscribe to the marking's announcements says so once at Warning
(`104_017`), sees a change on the refresh period only, and `104_018` says it is subscribed again.
### Changed
- **A host that cannot reach Redis keeps its last answer about the replay.** Before, every dispatch and
every catch-up on such a host failed with the connection's exception. Now they proceed on what the host
last learned — initially that no replay is active — and the loss is logged once. The replay stays
correct regardless: on the Orleans execution model the readers are paused by the rebuild and the
replay themselves, and apply the store from the beginning once more when the replay ends.
- **Rolling upgrade.** A host on a release before this one announces nothing when it starts or ends a
replay, so a host on this release sees that replay at the refresh period rather than at once. The
marking, its key and its lease are unchanged, so the two releases share one replay state.