Clean up and update regression test script and related files #202
Conversation
|
|
||
| else | ||
|
|
||
| if [[ $i =~ ufs.s2s ]]; then | ||
| d=$( cmp ${RTPWD}/${CNTLMED_DIR}/$i ${RUNDIR}/$i | wc -l ) |
There was a problem hiding this comment.
Can we leave this file and add extra netcdf file comparison using nccmp, cprnc or compare_ncfile.py (need to check efficiency) when cmp comparison shows difference?
There was a problem hiding this comment.
ufs.s2s.* is still being compared because it is in LIST_FILES of tests/tests/. It's just that CNTLMED_DIR no longer exists.
Let me look into these other comparison methods. Where can I find info on compare_ncfile.py?
|
On dell: /u/emc.glopara/bin/compare_ncfile.py
…On Wed, Oct 7, 2020 at 8:35 AM Minsuk Ji ***@***.***> wrote:
***@***.**** commented on this pull request.
------------------------------
In tests/rt_utils.sh
<#202 (comment)>
:
>
else
- if [[ $i =~ ufs.s2s ]]; then
- d=$( cmp ${RTPWD}/${CNTLMED_DIR}/$i ${RUNDIR}/$i | wc -l )
ufs.s2s.* is still being compared because it is in LIST_FILES of
tests/tests/. It's just that CNTLMED_DIR no longer exists.
Let me look into these other comparison methods. Where can I find info on
compare_ncfile.py?
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub
<#202 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AI7D6TIKKU6V3E6EZBFEGODSJROAPANCNFSM4R7AJ5GA>
.
|
|
usage: /gpfs/dell1/ptmp/Jun.Wang/fv3_parallel/compare_ncfile.py file1 file2
On Wed, Oct 7, 2020 at 8:37 AM Jun Wang - NOAA Federal <[email protected]>
wrote:
… On dell: /u/emc.glopara/bin/compare_ncfile.py
On Wed, Oct 7, 2020 at 8:35 AM Minsuk Ji ***@***.***> wrote:
> ***@***.**** commented on this pull request.
> ------------------------------
>
> In tests/rt_utils.sh
> <#202 (comment)>
> :
>
> >
> else
>
> - if [[ $i =~ ufs.s2s ]]; then
> - d=$( cmp ${RTPWD}/${CNTLMED_DIR}/$i ${RUNDIR}/$i | wc -l )
>
> ufs.s2s.* is still being compared because it is in LIST_FILES of
> tests/tests/. It's just that CNTLMED_DIR no longer exists.
>
> Let me look into these other comparison methods. Where can I find info on
> compare_ncfile.py?
>
> —
> You are receiving this because you commented.
> Reply to this email directly, view it on GitHub
> <#202 (comment)>,
> or unsubscribe
> <https://github.com/notifications/unsubscribe-auth/AI7D6TIKKU6V3E6EZBFEGODSJROAPANCNFSM4R7AJ5GA>
> .
>
|
|
sorry, my ptmp file is gone.
/u/emc.glopara/bin/compare_ncfile.py file1 file2
On Wed, Oct 7, 2020 at 8:38 AM Jun Wang - NOAA Federal <[email protected]>
wrote:
… usage: /gpfs/dell1/ptmp/Jun.Wang/fv3_parallel/compare_ncfile.py file1 file2
On Wed, Oct 7, 2020 at 8:37 AM Jun Wang - NOAA Federal ***@***.***>
wrote:
> On dell: /u/emc.glopara/bin/compare_ncfile.py
>
> On Wed, Oct 7, 2020 at 8:35 AM Minsuk Ji ***@***.***>
> wrote:
>
>> ***@***.**** commented on this pull request.
>> ------------------------------
>>
>> In tests/rt_utils.sh
>> <#202 (comment)>
>> :
>>
>> >
>> else
>>
>> - if [[ $i =~ ufs.s2s ]]; then
>> - d=$( cmp ${RTPWD}/${CNTLMED_DIR}/$i ${RUNDIR}/$i | wc -l )
>>
>> ufs.s2s.* is still being compared because it is in LIST_FILES of
>> tests/tests/. It's just that CNTLMED_DIR no longer exists.
>>
>> Let me look into these other comparison methods. Where can I find info
>> on compare_ncfile.py?
>>
>> —
>> You are receiving this because you commented.
>> Reply to this email directly, view it on GitHub
>> <#202 (comment)>,
>> or unsubscribe
>> <https://github.com/notifications/unsubscribe-auth/AI7D6TIKKU6V3E6EZBFEGODSJROAPANCNFSM4R7AJ5GA>
>> .
>>
>
|
|
On Orion, comparing a single file ufs.s2s.cpl.r.2013-04-01-43200.nc (2.4G) took nccmp about 7.6 seconds as opposed to 0.6 seconds by cmp. I think we should just use nccmp because:
From 3 and 4 above, unless they become a module at some point in time, they will have to reside in a person's directory, which is not a good idea. Let me know your thoughts. |
|
Just to be clear, you tested two files which cmp shows are different (giving 0.6secs) but the same two files using nccmp -d shows they are the same (and it takes 7.6s). I tried the same two sized files using my install of cprnc on Hera and got about 9secs, so it seems nccmp -d is slightly faster and doesn't have the porting issue. |
|
Minsuk, I think you need to load python module in order to use
compare_ncfile.py. For us we may need to use the netcdf comparison utility
as a utility replacing cmp. But for larger files, for a C768 atm files, it
takes several hours to compare one file ~6Gb compressed files. My
experience is that the compare_ncfile.py is pretty fast.
…On Wed, Oct 7, 2020 at 1:27 PM Minsuk Ji ***@***.***> wrote:
@junwang-noaa <https://github.com/junwang-noaa> @DeniseWorthen
<https://github.com/DeniseWorthen>
On Orion, comparing a single file ufs.s2s.cpl.r.2013-04-01-43200.nc
(2.4G) took nccmp about 7.6 seconds as opposed to 0.6 seconds by cmp.
I think we should just use nccmp because:
1. nccmp is available as a module on Hera and Orion
2. We are just comparing 1 or 2 ufs.s2s.* files (and even that as a
fallback), so maybe efficiency does not matter too much?
3. I have had some problem with cprnc on Orion due to shared libraries
(libnetcdff.so.6, libpthread-2.17.so, libc-2.17.so) not being found
4. compare_ncfile.py requires netCDF4 python module, which at this
point I am not sure where to find on Hera and Orion
From 3 and 4 above, unless they become a module at some point in time,
they will have to reside in a person's directory, which is not a good idea.
Let me know your thoughts.
—
You are receiving this because you were mentioned.
Reply to this email directly, view it on GitHub
<#202 (comment)>,
or unsubscribe
<https://github.com/notifications/unsubscribe-auth/AI7D6TIQNNJOMTPZKQAYYRTSJSQIDANCNFSM4R7AJ5GA>
.
|
|
We are creating new baselines in PR #200, but we are leaving the comparisons of the ufs.s2s.cpl<> files commented out until we have implemented the additional check if ufs.s2s.cpl<> shows a difference. Is that correct? |
That is also my understanding. Does it mean that this PR #202 will have to generate a new baseline, correct? |
|
Yes, I think that is right since the files don't get copied to the baseline unless they're in the list of files. |
|
I think this PR is no longer needed since Rahul and Dusan's weather PR (ufs-community/ufs-weather-model#217) is also making rt.sh related changes, and already running S2S based on new baselines. I propose we open a new PR in the weather repo later to implement nccmp. |
* add link_mpi=dbg for debug test
* add dbg_mt in debug option when OPENMP is on
* update log files for hera, orion, wcoss cray and dell; skip-ci
* update hera.gnu log file
* Set DEBUG_LINKMPI=OFF for cheyenne.intel and gaea.intel; fix post library dependencies for cheyenne.{intel,gnu} gnumake config
* Regression test log for cheyenne.intel
Co-authored-by: Jun Wang <[email protected]>
Co-authored-by: Dom Heinzeller <[email protected]>
Description of changes
Clean up and update rt.sh and related files
Specific notes
Issues Fixed (include github issue #): #201
Are changes expected to change answers?
Specific changes:
Testing performed:
Hashes used for testing: