Hello,
When I run my NEMO 5.0 config with XIOS 2, it runs fine.
I’m trying to switch to XIOS 3. My main motivation is to have less writers, as the cluster admins always complain I have too many disk access.
I copied the XIOS XML files from the new example config X3_ORCA2_ICE_PISCES. In a small test, I modified iodef.xml and I use 4 XIOS processes in 1 pool (Opool): “tgatherer”, “ugatherer”, “twriter”, “uwriter”. I modified file_def_nemo-oce.xml and indicated to use the right gatherer and writer, e.g.
<file id=“file11” name_suffix=“_grid_T” mode=“write” writer=“twriter” gatherer=“tgatherer” using_server2=“false” description=“ocean T grid variables”>
When running my config, the model hangs forever in iom_init, in particular in iom_setatt, in the calls to set_mooring(…). I don’t want to use anything related to moorings, so I commented the 3 related calls. As a side-node, I don’t know if I could have disabled the calls to set_moorings(…) from a namelist option that I forgot to put to .false., or if they are always there. In that case, it could make sense someday to move these 3 hard-coded vectors of longitude,latitude (TAO, RAMA and PIRATA moorings) out of iom.F90.
With these calls commented out, the model now hangs forever during the call to xios_close_context_definition()
I don’t know if it’s related to these set_mooring() that I don’t have anymore, or what to try out next.
Some notes:
- I tried XIOS in attached and detached mode. I don’t know what the XIOS option “using_server2” is, but I tried both true and false (in iodef.xml). In file_def_nemo-oce.xml I kept the value at “false” (as in the X3_ORCA2_PISCES_ICE example).
- In iodef.xml, I tried changing info_level from 0 to 9 but still XIOS does not tell me anything at all. At execution, "xios_server_??.err and .out files are created, but they are all empty
- Compiling XIOS in debug mode doesn’t give me any info either when I kill the model. I just exits without any message (either from NEMO or from XIOS)
- I’m working with gcc / gfortran. Given another topic on the forum [intel] Model hangs on initialization phase at
xios_close_context_definition
call - Versions - NEMO Community Ocean Model , I tried changing the version (gfortran 12.2.0-14 → 13.3.0) but the result is the same. I don’t think it’s a compiler bug as in that forum thread, but rather a deadlock in XIOS due to some wrong setting in my config - I tried AMM12 with XIOS3, but it also hangs forever during initialization. I did not yet start to add write(,) in the code where exactly is hangs, because I’m pretty convince it’s the same error.
Thank you !
Luc