Stratara.Diagnostics 3.1.4

• net10.0

  • Microsoft.Extensions.Logging.Abstractions (>= 10.0.8)
  • OpenTelemetry.Api (>= 1.15.3)

### Added

- **Command-workload isolation (heavy-command lane)** — long-running commands can now be routed to a
 dedicated worker lane so they cannot starve interactive commands. Mark a command with the new
 `Stratara.Abstractions.Mediator.IHeavyCommand` marker and the `ICommandOutboxDispatcher`
 automatically publishes it to a separate heavy-command topic (`IMessagingIdentifier.HeavyCommandTopic` /
 `HeavyCommandSubscription`, configurable under `Messaging:HeavyCommand`, defaulting to `heavy-command` /
 `heavy-command-subscription`). Run a dedicated heavy-command worker with the new
 `services.AddHeavyCommandWorker(degreeOfParallelism?)` extension, or the
 `builder.AddHeavyCommandWorkerServices(degreeOfParallelism?)` host composite — in the same process as
 the interactive worker (two lanes) or in a separately scaled host. Each worker's degree of parallelism
 is configurable per lane. `IMessagingIdentifier` gains `HeavyCommandTopic`, `HeavyCommandSubscription`,
 and the `GetCommandTopic(Type)` / `GetCommandSubscription(Type)` routing helpers. The interactive lane
 (`AddMediatorWorker()`) is unchanged and remains the default; commands not marked heavy keep their
 existing routing. If a heavy command is dispatched while no heavy worker is bound, the publish is
 rejected and the command is preserved in the outbox until a heavy-command worker comes online — it is
 never dropped. Works over both the RabbitMQ and Azure Service Bus message buses (Azure Service Bus
 requires the heavy-command topic/subscription to be provisioned, like the existing command topic). New
 log-event ID `105_005` (`CommandWorkerLaneStarted`) in `Stratara.Diagnostics`.
- **Observability metrics across the worker pipeline** (`Stratara.Diagnostics`) — the shared
 `Stratara.Service` meter now publishes throughput and latency instruments so operators can see how the
 event-sourcing pipeline is behaving instead of flying blind on a single counter. New instruments:
 `event_source.events.appended` (counter, tagged by `event.type` / `aggregate.type`),
 `outbox.published` (counter, tagged by `outbox.kind` = `command` / `event`), `command.duration`
 (histogram, ms, tagged by `request.type` / `outcome`), `projection.events.processed` (counter) +
 `projection.bundle.duration` (histogram, ms), `saga.events.processed` (counter) +
 `saga.bundle.duration` (histogram, ms), and `saga.inflight` (up/down counter). They are recorded by the
 event source, command worker, projection worker, saga worker, and outbox worker respectively. Because
 projections and sagas are real-time bus subscribers without a persisted checkpoint, these report
 **throughput and latency**, not consumer lag. No configuration is required — point any OpenTelemetry
 metrics exporter at the `Stratara.Service` meter.
- **Operational health checks for the event store and outbox** (`Stratara.EventSourcing.EntityFrameworkCore`) —
 two opt-in readiness checks added to any `IHealthChecksBuilder`: `AddEventStoreHealthCheck()` verifies
 the write-side database is reachable, and `AddOutboxHealthCheck(degradedThreshold?, unhealthyThreshold?)`
 reports the pending outbox backlog (exposed under the `pending` data key) and escalates to
 `Degraded` / `Unhealthy` when the backlog crosses the supplied thresholds. Both are tagged `ready` by
 default (so they map to a readiness endpoint, not liveness) and require the Stratara write store to be
 registered. The write-store DbContext is now also resolvable as a scoped `IWriteDbContext` service to
 support these checks.
- **Polly-backed mediator resilience behavior** (`Stratara.Resilience`) — an opt-in pipeline behavior
 wraps the in-process dispatch of a request marked with the new
 `Stratara.Abstractions.Resilience.IResilientRequest` in the named Polly pipeline the request selects
 (`ResiliencePipelineName`). Register it with the new `AddStrataraResilienceBehavior()` (after
 `AddStrataraValidation()` / `AddStrataraTenantIsolation()` so the retry wraps the handler, not the
 guards); requests without the marker are unaffected. A new built-in pipeline
 `ResilienceNames.ConcurrencyConflict` retries **only** on
 `Stratara.Abstractions.Persistence.ConcurrencyConflictException` (5 attempts, short exponential
 backoff) so a handler that re-reads and re-applies on an optimistic-concurrency clash succeeds without
 bespoke retry loops; it is registered by `AddResiliencePipelines()` alongside the existing message-bus
 and dispatcher pipelines. Only mark handlers that are safe to re-run (idempotent or concurrency-guarded).