[Gluster-users] fchown() and write() returning ENOENT error

[Sabyasachi Ruj]

Dan had these questions for GlusterFS:

So I guess the first thing we want to know about the setup
is whether or not posix locks can be considered barriers.

i.e. assuming that SQLite always holds a SHARED posix lock on a
particular range when reading the db file, and an EXCLUSIVE
lock on the same range when writing, what is the answer to
the following:

 When SQLite reads from the file, is it guaranteed that
 it will see all writes made to the file before the current
 lock (be it SHARED or EXCLUSIVE) was taken?

 Are locks also barriers for unlink()/access() calls? i.e.
 if a process holding an EXCLUSIVE lock unlinks a journal
 file, and then later on a process grabs a SHARED lock
 and uses access() to check for the existence of the
 journal file, is it guaranteed that it will find that there
 is no such file (i.e. will it see the results of the unlink()
 call).

 Does it affect either of the above when (or even if) fsync()
 is called by the process doing the writing or unlinking?