It is doable, I think. Basically you need to walk the internal list of
open files
looking for a name match. This is probably complicated if mpio is being used
because that list is duplicated on every processor.
As an alternative, you could implement wrapper around nc_create and nc_open
that call the regular nc_create/nc_open, but also keep a list of the
names of the
path arguments.
=Dennis Heimbigner
Unidata
On 1/22/2021 8:35 PM, Hernan Arango wrote:
Hi Lucas,
Thank you for your answer. However, what I am specifically asking is:
Given ONLY the NetCDF dataset filename, is there a function that can
be used in a Fortran code (or a C-binding routine) to inquire if the
filename is open (with a logical true or false result). If true,
return the one or more identifiers (vector) associated with the
opening of that specific filename. Notice that It is possible that
such filename has been opened more than once and not closed.
I assume that this is a very difficult question because I don’t have
an idea how to code such function. Maybe the developers of NetCDF can
tell us if it is possible or not.
Cheers, Hernan
*From: *Lucas Villa Real <lucasvr@xxxxxxxxxx>
*Date: *Friday, January 22, 2021 at 10:17 PM
*To: *Hernan Arango <arango@xxxxxxxxxxxxxxxxxx>
*Cc: *"netcdfgroup@xxxxxxxxxxxxxxxx" <netcdfgroup@xxxxxxxxxxxxxxxx>
*Subject: *Re: [netcdfgroup] How to inquire if a NetCDF file is
already open for reading or writing?
----- Original message -----
From: Hernan Arango <arango@xxxxxxxxxxxxxxxxxx>
Sent by: "netcdfgroup" <netcdfgroup-bounces@xxxxxxxxxxxxxxxx>
To: "netcdfgroup@xxxxxxxxxxxxxxxx" <netcdfgroup@xxxxxxxxxxxxxxxx>
Cc:
Subject: [EXTERNAL] [netcdfgroup] How to inquire if a NetCDF file
is already open for reading or writing?
Date: Fri, Jan 22, 2021 6:01 PM
Hello,
I am wondering if there is a function to determine if a NetCDF is
already open for reading or writing based on the filename? I am
trying to avoid open too many files. I know that there are ways
in Unix to increase the number of open files, but that’s not what
I am asking. In parallel I/O running on lots of processes, there
is a need to avoid opening too many files. I am using both the
standard NetCDF library and the PIO library that is the basis of
the SCORPIO library (https://e3sm.org/scorpio-parallel-io-library
<https://urldefense.proofpoint.com/v2/url?u=https-3A__e3sm.org_scorpio-2Dparallel-2Dio-2Dlibrary&d=DwMGaQ&c=jf_iaSHvJObTbx-siA1ZOg&r=yZCLoPFNF2xM1OIVBISlXxZvERMVKNT1d5PSrOJfnc8&m=QnVlDcV5suiBIhpLT3T5oRt9rSpedaNzD5PPQIaVwoY&s=9fcAm4IjIU6e3FbfD7JzZm28BtCM1zMwV3G7jQ-Bc0A&e=>).
Therefore, I have integer file ID (ncid) and file descriptors of
TYPE (file_desc_t).
I am looking for something like:
ncid=is_nc_open(‘my_file.nc’)
or
fileDesc=is_pio_open(iosystem, my_paralle_file.nc’)
Thank you, H
Hi Hernan,
If you're on Linux, then one option is to scan the symlinks under
/proc/self/fd looking for a target whose name matches the file name
you're looking for.
I'm not sure if ncids have a 1:1 correspondence with HDF5 file ids; if
they do, then perhaps you could also look into H5Fget_obj_count(ncid,
H5F_OBJ_FILE) and/or H5Fget_obj_count(H5F_OBJ_ALL, H5F_OBJ_FILE). See
https://support.hdfgroup.org/HDF5/doc/RM/RM_H5F.html#File-GetObjCount
<https://support.hdfgroup.org/HDF5/doc/RM/RM_H5F.html#File-GetObjCount>
for more details on that API.
Best regards,
Lucas
_______________________________________________
NOTE: All exchanges posted to Unidata maintained email lists are
recorded in the Unidata inquiry tracking system and made publicly
available through the web. Users who post to any of the lists we
maintain are reminded to remove any personal information that they
do not want to be made public.
netcdfgroup mailing list
netcdfgroup@xxxxxxxxxxxxxxxx
For list information or to unsubscribe, visit:
https://www.unidata.ucar.edu/mailing_lists/
