Change flush routine to the following. This will reduce the window for when a scan is required to within 5 minutes of a cluster allocation assuming no flushes have occurred.

  1. clear scheduled flush (if necessary)
  2. opportunistic flush (for whatever reason)
  3. mark clean
  4. no intervening metadata changes
  5. mark dirty
  6. flush
  7. schedule flush (in 5 minutes)
  8. metadata changes