fix: diag manager ensemble support when using the same diag/field/dat…

…a yaml (#1632)

data_override/include/data_override.inc

@@ -599,6 +599,10 @@ subroutine read_table_yaml(data_table)
     if (index(trim(filename), "ens_") .ne. 0) then
       if (file_exists(filename) .and. file_exists("data_table.yaml")) &
         call mpp_error(FATAL, "Both data_table.yaml and "//trim(filename)//" exists, pick one!")
+
+      !< If the end_* file does not exist, revert back to "data_table.yaml"
+      !! where every ensemble is using the same yaml
+      if (.not. file_exists(filename)) filename = "data_table.yaml"
     endif
 
     file_id = open_and_parse_file(trim(filename))

diag_manager/fms_diag_yaml.F90

@@ -36,7 +36,7 @@ module fms_diag_yaml_mod
                            time_diurnal, time_power, time_none, r8, i8, r4, i4, DIAG_NOT_REGISTERED, &
                            middle_time, begin_time, end_time, MAX_STR_LEN
 use yaml_parser_mod, only: open_and_parse_file, get_value_from_key, get_num_blocks, get_nkeys, &
-                           get_block_ids, get_key_value, get_key_ids, get_key_name
+                           get_block_ids, get_key_value, get_key_ids, get_key_name, missing_file_error_code
 use fms_yaml_output_mod, only: fmsYamlOutKeys_type, fmsYamlOutValues_type, write_yaml_from_struct_3, &
                                yaml_out_add_level2key, initialize_key_struct, initialize_val_struct
 use mpp_mod,         only: mpp_error, FATAL, NOTE, mpp_pe, mpp_root_pe, stdout
@@ -394,8 +394,13 @@ subroutine diag_yaml_object_init(diag_subset_output)
   if (index(trim(yamlfilename), "ens_") .ne. 0) then
     if (file_exists(yamlfilename) .and. file_exists("diag_table.yaml")) &
       call mpp_error(FATAL, "Both diag_table.yaml and "//trim(yamlfilename)//" exists, pick one!")
+    !< If the end_* file does not exist, revert back to "diag_table.yaml"
+    !! where every ensemble is using the same yaml
+    if (.not. file_exists(yamlfilename)) yamlfilename = "diag_table.yaml"
   endif
   diag_yaml_id = open_and_parse_file(trim(yamlfilename))
+  if (diag_yaml_id .eq. missing_file_error_code) &
+    call mpp_error(FATAL, "The "//trim(yamlfilename)//" is not present and it is required!")
 
   call diag_get_value_from_key(diag_yaml_id, 0, "title", diag_yaml%diag_title)
   call get_value_from_key(diag_yaml_id, 0, "base_date", diag_yaml%diag_basedate)

field_manager/field_manager.F90

@@ -618,6 +618,10 @@ subroutine read_field_table_yaml(nfields, table_name)
 if (index(trim(filename), "ens_") .ne. 0) then
   if (file_exists(filename) .and. file_exists(tbl_name)) &
     call mpp_error(FATAL, "Both "//trim(tbl_name)//" and "//trim(filename)//" exists, pick one!")
+
+  !< If the end_* file does not exist, revert back to tbl_name
+  !! where every ensemble is using the same yaml
+  if (.not. file_exists(filename)) filename = tbl_name
 endif
 
 if (.not. file_exists(trim(filename))) then

parser/yaml_parser.F90

@@ -39,6 +39,7 @@ module yaml_parser_mod
 private
 
 public :: open_and_parse_file
+public :: missing_file_error_code
 public :: get_num_unique_blocks
 public :: get_unique_block_ids
 public :: get_block_name
@@ -52,6 +53,8 @@ module yaml_parser_mod
 !public :: clean_up
 !> @}
 
+integer, parameter :: missing_file_error_code = 999
+
 !> @brief Dermine the value of a key from a keyname
 !> @ingroup yaml_parser_mod
 interface get_value_from_key
@@ -253,7 +256,7 @@ function open_and_parse_file(filename) &
 
    inquire(file=trim(filename), EXIST=yaml_exists)
    if (.not. yaml_exists) then
-      file_id = 999
+      file_id = missing_file_error_code
       call mpp_error(NOTE, "The yaml file:"//trim(filename)//" does not exist, hopefully this is your intent!")
       return
    end if

test_fms/data_override/test_data_override_ensembles.sh

@@ -86,14 +86,36 @@ data_table:
    fieldname_in_model: runoff
    override_file:
    - fieldname_in_file: runoff
-     file_name: INPUT/runoff.daitren.clim.1440x1080.v20180328_ens_02.nc
+     file_name: INPUT/runoff.daitren.clim.1440x1080.v20180328_ens_01.nc
      interp_method: none
    factor: 1.0
 _EOF
 
     test_expect_failure "test_data_override with both data_table.yaml and data_table.ens_xx.yaml files" '
       mpirun -n 12 ../test_data_override_ongrid_${KIND}
       '
+
+cat <<_EOF > input.nml
+&data_override_nml
+  use_data_table_yaml = .True.
+/
+
+&test_data_override_ongrid_nml
+  test_case = 6
+  write_only = .False.
+/
+
+&ensemble_nml
+   ensemble_size = 2
+/
+_EOF
+
+rm -rf INPUT/.
+rm -rf data_table.ens_01.yaml data_table.ens_02.yaml
+test_expect_success "test_data_override with two ensembles, same yaml file (${KIND})" '
+      mpirun -n 12 ../test_data_override_ongrid_${KIND}
+      '
+
 rm -rf INPUT
 fi
 test_done

test_fms/data_override/test_data_override_ongrid.F90

@@ -55,6 +55,7 @@ program test_data_override_ongrid
 integer, parameter                         :: scalar = 3
 integer, parameter                         :: weight_file = 4
 integer, parameter                         :: ensemble_case = 5
+integer, parameter                         :: ensemble_same_yaml = 6
 integer                                    :: test_case = ongrid
 integer                                    :: npes
 integer, allocatable                       :: pelist(:)
@@ -82,7 +83,7 @@ program test_data_override_ongrid
 call mpp_get_current_pelist(pelist)
 
 select case (test_case)
-case (ensemble_case)
+case (ensemble_case, ensemble_same_yaml)
   call set_up_ensemble_case()
 end select
 
@@ -93,7 +94,7 @@ program test_data_override_ongrid
 call mpp_get_data_domain(Domain, is, ie, js, je)
 
 select case (test_case)
-case (ensemble_case)
+case (ensemble_case, ensemble_same_yaml)
   ! Go back to the full pelist
   call mpp_set_current_pelist(pelist)
 end select
@@ -108,7 +109,7 @@ program test_data_override_ongrid
     call generate_scalar_input_file ()
   case (weight_file)
     call generate_weight_input_file ()
-  case (ensemble_case)
+  case (ensemble_case, ensemble_same_yaml)
     call generate_ensemble_input_file()
   end select
 
@@ -117,7 +118,7 @@ program test_data_override_ongrid
 
 else
   select case (test_case)
-  case (ensemble_case)
+  case (ensemble_case, ensemble_same_yaml)
     !< Go back to the ensemble pelist
     call mpp_set_current_pelist(pelist_ens)
   end select
@@ -134,7 +135,7 @@ program test_data_override_ongrid
     call scalar_test()
   case (weight_file)
     call weight_file_test()
-  case (ensemble_case)
+  case (ensemble_case, ensemble_same_yaml)
     call ensemble_test()
     call mpp_set_current_pelist(pelist)
   end select
@@ -709,26 +710,34 @@ subroutine ensemble_test()
   real(lkind)                                :: expected_result  !< Expected result from data_override
   type(time_type)                            :: Time             !< Time
   real(lkind), allocatable, dimension(:,:)   :: runoff           !< Data to be written
+  integer                                    :: scale_fac        !< Scale factor to use when determining
+                                                                 !! the expected answer
+  logical :: sucessful !< .True. if the data_override was sucessful
 
   allocate(runoff(is:ie,js:je))
 
+  scale_fac = ensemble_id
+  if (test_case .eq. ensemble_same_yaml) scale_fac = 1
+
   runoff = 999._lkind
   !< Run it when time=3
   Time = set_date(1,1,4,0,0,0)
-  call data_override('OCN','runoff',runoff, Time)
+  call data_override('OCN','runoff',runoff, Time, override=sucessful)
+  if (.not. sucessful) call mpp_error(FATAL, "The data was not overriden correctly")
   !< Because you are getting the data when time=3, and this is an "ongrid" case, the expected result is just
-  !! equal to the data at time=3, which is 3+ensemble_id.
-  expected_result = 3._lkind + real(ensemble_id,kind=lkind)
+  !! equal to the data at time=3, which is 3+scale_fac.
+  expected_result = 3._lkind + real(scale_fac,kind=lkind)
   call compare_data(Domain, runoff, expected_result)
 
   !< Run it when time=4
   runoff = 999._lkind
   Time = set_date(1,1,5,0,0,0)
-  call data_override('OCN','runoff',runoff, Time)
-  !< You are getting the data when time=4, the data at time=3 is 3+ensemble_id. and at time=5 is 4+ensemble_id.,
+  call data_override('OCN','runoff',runoff, Time, override=sucessful)
+  if (.not. sucessful) call mpp_error(FATAL, "The data was not overriden correctly")
+  !< You are getting the data when time=4, the data at time=3 is 3+scale_fac. and at time=5 is 4+scale_fac.,
   !! so the expected result is the average of the 2 (because this is is an "ongrid" case and there
   !! is no horizontal interpolation).
-  expected_result = (3._lkind + real(ensemble_id,kind=lkind) + 4._lkind + real(ensemble_id,kind=lkind)) / 2._lkind
+  expected_result = (3._lkind + real(scale_fac,kind=lkind) + 4._lkind + real(scale_fac,kind=lkind)) / 2._lkind
   call compare_data(Domain, runoff, expected_result)
 
   deallocate(runoff)

test_fms/diag_manager/test_ens_runs.F90

@@ -25,7 +25,8 @@ program test_ens_runs
                               register_diag_field, diag_manager_init, diag_manager_end, register_static_field, &
                               diag_axis_init
   use time_manager_mod, only: time_type, operator(+), JULIAN, set_time, set_calendar_type, set_date
-  use mpp_mod,          only: FATAL, mpp_error, mpp_npes, mpp_pe, mpp_get_current_pelist, mpp_set_current_pelist
+  use mpp_mod,          only: FATAL, mpp_error, mpp_npes, mpp_pe, mpp_get_current_pelist, mpp_set_current_pelist, &
+                              input_nml_file
   use fms2_io_mod,      only: FmsNetcdfFile_t, open_file, close_file, read_data, get_dimension_size, &
                               set_filename_appendix, get_instance_filename
   use ensemble_manager_mod, only: get_ensemble_size, ensemble_manager_init
@@ -47,7 +48,18 @@ program test_ens_runs
   character(len=10) :: text           !< The filename appendix
   integer :: expected_ntimes
 
+  integer                            :: io_status       !< Status after reading the namelist
+
+  !< Configuration parameters
+  logical :: using_ens_yaml = .true.      !< Indicates whether or not ensembles yamls are used in the test
+
+  namelist / test_ens_runs_nml / using_ens_yaml
+
   call fms_init
+
+  read (input_nml_file, test_ens_runs_nml, iostat=io_status)
+  if (io_status > 0) call mpp_error(FATAL,'=>test_ens_runs: Error reading input.nml')
+
   call ensemble_manager_init
   npes = mpp_npes()
   if (npes .ne. 2) &
@@ -69,6 +81,7 @@ program test_ens_runs
     ensemble_id = 2
     call mpp_set_current_pelist((/1/))
     expected_ntimes = 24
+    if (.not. using_ens_yaml) expected_ntimes = 48
   endif
 
   write( text,'(a,i2.2)' ) 'ens_', ensemble_id
@@ -105,6 +118,7 @@ subroutine check_output()
     real, allocatable     :: var_data(:) !< Buffer to read variable data to
     integer               :: j           !< For looping
     character(len=255)    :: filename    !< Name of the diag file
+    integer               :: scale_fac   !< Scale factor to use when determining the expected answer
 
     call get_instance_filename("test_ens.nc", filename)
     if (.not. open_file(fileobj, filename, "read")) &
@@ -116,9 +130,12 @@ subroutine check_output()
     allocate(var_data(var_size))
     var_data = -999.99
 
+    scale_fac = 1
+    if (using_ens_yaml) scale_fac = ensemble_id
+
     call read_data(fileobj, "var0", var_data)
     do j = 1, var_size
-      if (var_data(j) .ne. real(j * ensemble_id))&
+      if (var_data(j) .ne. real(j * scale_fac))&
         call mpp_error(FATAL, "The variable data for var1 at time level:"//&
                               string(j)//" is not the correct value!")
     enddo

test_fms/diag_manager/test_ens_runs.sh

@@ -77,10 +77,10 @@ cat <<_EOF > diag_table.yaml
 title: test_diag_manager_01
 base_date: 2 1 1 0 0 0
 diag_files:
-- file_name: test_0days
+- file_name: test_ens
   time_units: days
   unlimdim: time
-  freq: 0 days
+  freq: 1 hours
   varlist:
   - module: ocn_mod
     var_name: var0
@@ -93,5 +93,24 @@ test_expect_failure "Running diag_manager with both diag_table.yaml and diag_tab
   mpirun -n 2 ../test_ens_runs
 '
 
+# Both ensembles have the same diag table yaml
+cat <<_EOF > input.nml
+&diag_manager_nml
+  use_modern_diag = .True.
+/
+
+&ensemble_nml
+   ensemble_size = 2
+/
+&test_ens_runs_nml
+  using_ens_yaml = .false.
+/
+_EOF
+rm -rf diag_table.ens_01.yaml diag_table.ens_02.yaml
+my_test_count=`expr $my_test_count + 1`
+test_expect_success "Running diag_manager with 2 ensembles, both ensembles have the same yaml (test $my_test_count)" '
+  mpirun -n 2 ../test_ens_runs
+'
+
 fi
 test_done

test_fms/field_manager/test_field_manager2.sh

@@ -148,6 +148,33 @@ _EOF
   rm -rf field_table.yaml
 
   test_expect_success "field manager test with 2 ensembles" 'mpirun -n 2 ./test_field_table_read'
+
+cat <<_EOF > input.nml
+&field_manager_nml
+  use_field_table_yaml = .true.
+/
+&test_field_table_read_nml
+  test_case = 2
+/
+&ensemble_nml
+   ensemble_size = 2
+/
+_EOF
+
+  cat <<_EOF > field_table.yaml
+field_table:
+- field_type: tracer
+  modlist:
+  - model_type: atmos_mod
+    varlist:
+    - variable: radon
+    - variable: radon2
+    - variable: radon3
+      longname: bad radon!
+_EOF
+
+rm -rf field_table.ens_01.yaml field_table.ens_02.yaml
+test_expect_success "field manager test with 2 ensembles same yaml" 'mpirun -n 2 ./test_field_table_read'
 fi
 
 test_done

test_fms/field_manager/test_field_table_read.F90

@@ -53,6 +53,7 @@ program test_field_table_read
 character(len=10) :: text
 integer, parameter :: default_test = 0
 integer, parameter :: ensemble_test = 1
+integer, parameter :: ensemble_same_yaml_test = 2
 
 ! namelist parameters
 integer :: test_case = default_test
@@ -69,7 +70,7 @@ program test_field_table_read
 
 nfields_expected = 4
 select case (test_case)
-case (ensemble_test)
+case (ensemble_test, ensemble_same_yaml_test)
   if (npes .ne. 2) &
     call mpp_error(FATAL, "test_field_table_read:: this test requires 2 PEs!")
 
@@ -78,16 +79,16 @@ program test_field_table_read
   if (ens_siz(1) .ne. 2) &
     call mpp_error(FATAL, "This test requires 2 ensembles")
 
+  nfields_expected = 3
   if (mpp_pe() .eq. 0) then
     !PEs 0 is the first ensemble
     ensemble_id = 1
     call mpp_set_current_pelist((/0/))
-    nfields_expected = 3
   else
     !PEs 1 is the second ensemble
     ensemble_id = 2
     call mpp_set_current_pelist((/1/))
-    nfields_expected = 4
+    if (test_case .eq. ensemble_test) nfields_expected = 4
   endif
 
   write( text,'(a,i2.2)' ) 'ens_', ensemble_id