Transient seg faults

[bueker]

Newbie to WRF. Version 4.5.

Successful compilation under WSL2 Ubuntu 22.04 (after a few days of permutations, of course).

Running 9 km dx=dy.
0.25 GFS initialization.
8 processors.
24-hour run.

(namelist.input attached. The rsl files don't show much out of the ordinary, except for what I saw below. No CFL issues, for example...)

Model runs about 9-10 hours from initialization, then seg faults:

$ tail -10 rsl.error.0002

#5 0x14523ca in ???
#6 0x487626 in ???
#7 0x408e77 in ???
#8 0x4082be in ???
#9 0x7ff027bced8f in __libc_start_call_main
at ../sysdeps/nptl/libc_start_call_main.h:58
#10 0x7ff027bcee3f in __libc_start_main_impl
at ../csu/libc-start.c:392
#11 0x4082f5 in ???
#12 0xffffffffffffffff in ???

Here's the odd thing: No seg fault occurs when I restart from an rst file written just before the crash!!

I could recompile with -D to find the crash line, but I am concerned the transient nature may limit the usefulness of this information.

Any ideas would be helpful! Apologies if I missed any similar issue in the forum archive.

 

[Ming Chen]

Hi,
Your namelist.input looks fine except that time-step = 20 is way too small for 9km grid interval. Pleas increase it to 45. Also, it is better to set radt =9 and turn on w_damping = 1.
if the model crashed after 9-10 hours of integration, it indicates that something is wrong in the physic/dynamics. if you save the wrfout file right before the model crash, can you find any weird values in the file?
It is quite uncommon that the model crashed but would continue to run after restarting. Please keep me updated if you get more information. Thanks in advance.

 

[bueker]

Thanks for your reply, Ming.

Yes, I had been at 4 km, I figured keeping the small time step would not be an issue, since most problems occur when CFL is violated. I also figured the smaller time step would mitigate the need for w-damping.

I will try the settings you recommend and examine the velocity fields right before the original crash, but I did not see anything out of the ordinary after a cursory look at a few fields.

I am hoping the new settings do the trick, otherwise I fear the possibility of a memory management issue.