Current write sequence follows these steps(candidate supported):

• edit(candidate)
• commit

But it is required to use locking and also detect concurrent edits for shared datastores (using discard changes).

Expected canonical write for devices supporting only writable running capability:

• lock running or fail
• edit running (unlock on fail)
• unlock running

supporting only candidate:

• lock candidate
  • if lock fails, try to discard changes
    • if discard fails, fail
    • if discard succeeds, lock candidate
      • if lock fails, fail
• edit candidate (discard-changes + unlock on fail)
• commit
• unlock candidate

supporting both writable running and candidate:

• lock running or fail
• SAME AS FOR CANDIDATE ONLY
• unlock running