Stratara.ServiceDefaults 4.1.1

• net10.0

  • Destructurama.Attributed (>= 5.3.0)
  • Microsoft.Extensions.Http.Resilience (>= 10.9.0)
  • Microsoft.Extensions.ServiceDiscovery (>= 10.9.0)
  • OpenTelemetry.Exporter.OpenTelemetryProtocol (>= 1.18.0)
  • OpenTelemetry.Extensions.Hosting (>= 1.18.0)
  • OpenTelemetry.Instrumentation.EntityFrameworkCore (>= 1.18.0-beta.1)
  • OpenTelemetry.Instrumentation.Http (>= 1.18.0)
  • OpenTelemetry.Instrumentation.Runtime (>= 1.18.0)
  • RabbitMQ.Client.OpenTelemetry (>= 1.0.0-rc.2)
  • Serilog.Extensions.Hosting (>= 10.0.0)
  • Serilog.Formatting.Compact (>= 3.0.0)
  • Serilog.Settings.Configuration (>= 10.0.1)
  • Serilog.Sinks.Async (>= 2.1.0)
  • Serilog.Sinks.Console (>= 6.1.1)
  • Serilog.Sinks.File (>= 7.0.0)
  • Serilog.Sinks.OpenTelemetry (>= 4.2.0)
  • Stratara.Shared (>= 4.1.1)

Correctness fixes for the Orleans execution model ahead of production use. Recorded commands can no longer
reach a bus drain, a durable timer fires once however long its handler runs, and heavy commands are
specified as the exception to the one-writer-per-aggregate promise they always were. The execution-model
reset removes only the host's own checkpoints, and three store-level fixes land for the partition
counter, its backfill and a checkpoint's first write. No schema change; **stop every bus outbox worker
before upgrading hosts that record commands through the execution model**, and register
`AddStrataraIntentStore` on the silos that run the drain.

### Changed

- **Orleans: the execution-model reset removes only the checkpoints of the host's own projections and
 sagas.** It removed every checkpoint in the read store, so a reset in one deployment made another
 deployment sharing the read store rebuild its projections unasked. The reset now removes the
 checkpoints of the projections and sagas registered in the composition it is resolved from, and the
 report counts only those. A checkpoint of a projection the host no longer registers stays; nothing
 reads it. Resolve the reset from the host's own composition — one that registers no store reader
 removes no checkpoint.
- **The documentation states when a read store can be shared.** Checkpoints are keyed by consumer, not
 by deployment: deployments sharing a read store need distinct projection names, and at most one of
 them runs `AddStrataraSagaGrains`.
- **Orleans: heavy commands are specified as running outside their aggregate's turn and order.** They
 always ran in their own bounded pool beside the aggregate's other commands; the specification promised
 otherwise. Their hand-over no longer holds the per-aggregate send order either: a command dispatched after
 a heavy command for the same aggregate no longer waits for it, and two due heavy commands on one aggregate
 no longer hold up the drain's resume pass. Where a heavy command and another command on the same aggregate
 both append, the store's version check refuses the later one, which is resumed like any failing command.
- **Orleans: a recorded command is stored under a kind of its own.** No bus outbox drain reads it any more.
 Commands recorded under 4.1.0 are still resumed; stop every bus outbox worker before upgrading hosts that
 record commands.

### Fixed

- **Orleans: the outbox drain no longer publishes recorded commands to the bus.** It chose between resuming
 recorded commands and publishing stored commands by whether the execution model's dispatcher was
 registered on its own silo, and the migration guide registers the dispatcher on the API host and the drain
 on the silos. Such a silo handed every recorded command — kept ones included — to the bus: commands ran
 twice, kept commands came back, the attempt bound did not hold. The drain now resumes recorded commands
 wherever an intent store is registered and warns (`117_111`) where recorded commands exist but none is.
 Register `AddStrataraIntentStore` on the silos that run the drain.
- **Orleans: a durable timer fires once however long its handler runs.** A handler that outlasted the
 reminder call's response timeout was started again by the next tick. A tick that arrives while the same
 timer's handler runs now does nothing, and a tick confirms its timer is still registered before firing.

- **Orleans: two writers of the same first checkpoint no longer fail one of them.** When an
 activation and its successor overlap during a failover, both can find no checkpoint row and both
 insert it; the one that lost to the key threw. It now updates the row the other inserted.
- **Orleans: the partition counter works on a model that does not use snake-case names.** The
 interceptor advanced the counter with SQL naming `partition_position` and its columns literally, so
 a write context mapping the counter under other names failed every append. It now goes through the
 context's model, as the native reader already did.
- **Orleans: the partition-counter backfill positions old entries in the order they were appended.**
 It ordered each partition's history by bucket first, so a projection or saga reading across
 aggregates saw the facts of one bucket before earlier facts of another. Entries are now positioned
 in sequence order, as the event-sourcing-store specification already promised. A store backfilled
 under 4.1.0 keeps the positions it was given; the backfill does not reposition entries.