Hi Thomas,
Rob Ross wrote:
> Also, I stand corrected -- you are right that there don't appear to be
> any calls to fsync() in netCDF (at least version 3.6.3 that I just
> looked through). That, obviously, surprises me.
It surprised me too, when I looked for the fsync() call in nc_sync().
The documented way to share access among processes to the same file is
to call nf90_open with the NF90_SHARE flag:
The NF90_SHARE flag is appropriate when one process may be writing the
dataset and one or more other processes reading the dataset
concurrently ...
but with no fsync() call and no call to fcntl using O_SYNC, it looks
like the documentation for the nc_sync() function and corresponding
nf90_sync function is inconsistent with the implementation.
Just out of curiosity, did you try using the NF90_SHARE flag on the
create/open calls in the writing and reading processes to see if that
makes any difference?
The lack of an fsync() call in the netCDF-3 library is not a recent
change. I checked older versions and it hasn't been there at least back
to version 3.4 released in 1998. I will have to do some research to see
whether avoiding fsync() was entirely a portability issue for use on
non-Unix systems, or whether there was some other intention. But for
now, I agree that it looks like calling nc_sync() ought to also call
fsync() on the underlying file descriptor. If nc_sync() did that, there
would be no need for a new nc_fsync() function.
Thanks for bringing this issue to our attention.
--Russ
