[Discuss] Gluster startup, small-files performance

[Richard Pieri]

Bill Bogstad wrote:
> I wasn't saying that  POSIX doesn't require fsync() to force data to
> disk.   I was saying that POSIX says nothing about if ANY data has to
> actually be saved to stable storage if fsync()/close() has NOT been called.

Mea culpa. I misunderstood your point. You are correct: POSIX says
nothing about what to do with data if fsync() and close() aren't called.

The programmers' assumption likely derives from the sync or flush daemon
which historically calls sync() every 30 seconds. Reliance on the sync
daemon is fine on traditional, POSIX compliant file systems with honest
hardware and assuming that durability isn't a priority. Not so much on
cluster file systems that use their own flush mechanisms to keep writes
ordered properly.

-- 
Rich P.