Stratara 4.4.8

A fix for hosts on the Orleans execution model. A store reader that catches up a large backlog and a singleton
work that runs long now answer their keep-alive, so neither leaves reminder errors in the silo's log. A silo that
joins while a singleton work runs long on another silo no longer fails its start. No API change, no schema change
and no migration.

Fixed

• A busy store reader or singleton work answers its keep-alive. On the Orleans execution model, a projection or
  saga reader that caught up a backlog for longer than the runtime's response timeout made its keep-alive tick wait
  behind the catch-up, and Orleans logged Could not deliver reminder tick at Error, once per partition and
  keep-alive period. A singleton work whose run outlasted the timeout did the same. Nothing was lost, but a large
  catch-up left dozens of errors on the silo. The keep-alive and the poll now start a catch-up where none is
  running and do not wait for it. A partition is still read by one catch-up at a time, and a commit during a
  catch-up is still applied by it. A singleton work's keep-alive interleaves with its run, and runs still never
  overlap.
• A silo starts while a singleton work runs long. A silo that started while a singleton work it registers ran
  longer than the response timeout on another silo waited for that run in its own start. The start then failed
  with a TimeoutException. The call a starting silo makes to each singleton work and store reader now
  interleaves, so the start goes on.
• Operate the Orleans execution model says that a durable timer whose handler outlasts the response timeout
  still makes Orleans log Could not deliver reminder tick for that tick, and that the line says nothing about
  the timer. It also says that, outside a deploy, the line now points to an unreachable store reader or singleton
  work rather than a busy one.