sample_schism_hbv3.yml

resource_group: schism_rg # the resource group containing the batch account
job_name: simulation_scenario # job name, will be used to name the pool and the job
batch_account_name: mybatchaccount # batch account name
storage_account_name: mybatchassociatedstore # storage account name
storage_container_name: test # this is mounted to $AZ_BATCH_MOUNTS_DIR/<<storage_container_name>> in addition to batch container which is mounted to $AZ_BATCH_MOUNTS_DIR/batch
study_copy_flags: --recursive --preserve-symlinks --exclude-regex ".*out.*nc"
study_dir: simulations/my_scenario # path to simulation directory within storage_container_name
setup_dirs: # supplementary directories copied alongside study_dir
 - hrrr # meteorological forcing data directory
vm_size: Standard_HB120rs_v3 #Standard_D16s_v3 # "Standard_D8as_v6" #"standard_hc44rs" #"Standard_HB120rs_v2" # this is the VM size for the pool, e.g. "Standard_D16s_v3"
num_hosts: 2 # number of compute nodes in the pool
# num_cores: 240 # optional; defaults to num_hosts * cores_per_node from vm_core_map
# HBv3 has 120 physical cores, so num_cores = 2 * 120 = 240 by default
num_scribes: 10 # number of SCHISM scribes; use fewer for GEN_TVD-VL output variant

# HBv3 (Standard_HB120rs_v3): 120 cores AMD EPYC Milan-X, 448 GiB, 3D V-Cache, NDR200 InfiniBand
# Milan-X has significantly larger L3 cache than Rome/Milan which benefits dense-mesh simulations.
# This template is specifically tuned for HBv3 with MVAPICH2; for HBv2 use sample_schism_mvapich2_2024apr.yml.
vm_size: "Standard_HB120rs_v3"

node_type: 'TargetDedicatedNodes' # 'TargetDedicatedNodes' or 'TargetLowPriorityNodes' (spot/preemptible)
task_slots_per_node: 1
max_retry_count: 3 # retry a task up to this many times if it fails
delete_after_mins: 1000 # job and nodes auto-deleted after this many idle minutes

mpi_command: |
  cd sflux; rm -f *.nc; python3 make_links_full.py; cd ../;
  rm -f *.nc;
  echo "------running make_links";
  python make_links.py;
  echo "------sflux links made";
  ls | head ;
  cd ../; 
  echo "------checking param.nml link";
  ls -l param.nml;
  echo "------checking param.nml content";
  head -10 param.nml
  current_date=$(date);
  if [ -f $AZ_BATCH_TASK_ID.state.txt ]; then
    echo $current_date >> $AZ_BATCH_TASK_ID.state.txt;
    echo "Restarting from previous run: $current_date";
    iterations=$($SCHISM_SCRIPTS_HOME/batch/get_iterations_from_fluxout.sh);
    echo "flux.out @ $iterations. Copying hotstart_\d+_$iterations\.nc files";
    azcopy copy --include-regex="hotstart_\d+_$iterations\.nc" --recursive "https://{storage_account_name}.blob.core.windows.net/{storage_container_name}/{study_dir}/outputs/?{sas}" . || true;
    echo "Generating and linking hotstart for $iterations";
    $SCHISM_SCRIPTS_HOME/batch/generate_and_link_hotstart.sh $iterations;
    $SCHISM_SCRIPTS_HOME/batch/update_param_for_restart.sh param.nml;
  else
    echo $current_date > $AZ_BATCH_TASK_ID.state.txt;
  fi
  echo "------checking files with ls";
  ls;
  echo "------starting SCHISM";
  mpirun {mpi_opts} pschism_PREC_EVAP_GOTM_TVD-VL {num_scribes}

  # GEN_TVD-VL variant includes GEN tracer transport (e.g. for salmon/nutrient simulations)
  # Replace with pschism_PREC_EVAP_GOTM_TVD-VL if GEN tracers are not needed
  #mpirun {mpi_opts} pschism_PREC_EVAP_GOTM_GEN_TVD-VL {num_scribes}

# template — Alma Linux 8.7 HPC image with MVAPICH2 optimized for HBv3 (Standard_HB120rs_v3)
# Use this when targeting HBv3 nodes with their larger on-chip L3 cache.
template_name: "alma810_mvapich2_202505290_hbv3" # this is the template name for the pool. Found in azure_dms_batch/dmsbatch/templates