io_server_state_reader_mod Module Reference

Reads the IO server state that was stored in a NetCDF checkpoint file. More...

Data Types
interface	nc_get_vara_long
	ISO C binding for NetCDF get long scalar variable, required for retrieving long variables. More...
interface	nc_get_vars_text
	ISO C binding for NetCDF get text vars, required for 64 bit start, count & stride. More...
interface	nc_inq_dim
	ISO C binding for NetCDF inquire dimension, required for 64 bit dimension length. More...
interface	writer_field_manager_unserialise_state

Functions/Subroutines
subroutine, public	read_io_server_configuration (checkpoint_filename, io_xml_configuration, io_communicator_arg)
	Reads the IO server configuration, which is the XML configuration initially run with and stored in the checkpoint. Note that this will open, read the XML in and then close the file. More...
subroutine, public	reactivate_writer_federator_state (io_configuration, writer_entries, time_points)
	Reactivates the writer federator and everything beneath it (i.e. just not the writer field manager.) For memory reasons this explicitly reopens the checkpoint file, will read each individual byte code entry in & repackage before deallocating memory and moving onto the next entry. The file is then closed. More...
subroutine, public	reactivate_writer_field_manager_state (io_configuration, unserialise_writer_field_manager)
	Reactivates the writer field manager state from the checkpoint file, for memory reasons this will open the file, read in and deserialise the byte code before closing it. More...
subroutine	get_io_server_configuration (ncid, io_xml_configuration)
	Retrieves the IO server XML configuration from the checkpoint file. More...
subroutine	get_io_server_serialised_bytes (ncid, number_io_server, my_io_server_rank, base_key, raw_bytes)
	Retrieves some IO server serialised bytes which will make up the state of a specific facet. Note that this uses the ISO C bindings in order to support 64 bit starts, counts & strides along with 64 bit scalar long variable fields. More...
subroutine	restart_writer_state_from_checkpoint (writer_entries, raw_bytes)
	Restarts the writer state from a specific checkpoint byte data chunk of memory. More...
subroutine	restart_writer_state_timepoints (time_points, raw_bytes)
	Restarts the writer state timepoints held in the writer federator. More...
subroutine	restart_timeaveraged_state_from_checkpoint (raw_bytes)
	Will restart the time averaged manipulation state from the checkpoint file. More...
subroutine	restart_instantaneous_state_from_checkpoint (raw_bytes)
	Will restart the instantaneous manipulation state from the checkpoint file. More...
subroutine	restart_writer_field_manager_from_checkpoint (unserialise_writer_field_manager, raw_bytes)
	Will restart the field manager state from the checkpoint file. More...

Detailed Description

Reads the IO server state that was stored in a NetCDF checkpoint file.

Function/Subroutine Documentation

get_io_server_configuration()

subroutine io_server_state_reader_mod::get_io_server_configuration ( integer, intent(in)  ncid, character, dimension(:), intent(inout), allocatable  io_xml_configuration  )

private

Retrieves the IO server XML configuration from the checkpoint file.

Parameters

ncid	The NetCDF checkpoint file ID
io_xml_configuration	XML configuration is read from the checkpoint and placed into here

Definition at line 169 of file io_state_reader.F90.

    integer, intent(in) :: ncid
    character, dimension(:), allocatable, intent(inout) :: io_xml_configuration
 
    integer :: dim_id, var_id, dim_size
    logical :: found
 
    call check_netcdf_status(nf90_inq_dimid(ncid, "io_configuration_dim", dim_id), found)
    if (.not. found) return
    call check_netcdf_status(nf90_inquire_dimension(ncid, dim_id, len=dim_size))
 
    call check_netcdf_status(nf90_inq_varid(ncid, "io_configuration", var_id), found)
    if (.not. found) return
    allocate(io_xml_configuration(dim_size))
    call check_netcdf_status(nf90_get_var(ncid, var_id, io_xml_configuration, count=(/dim_size/)))

Here is the call graph for this function:

Here is the caller graph for this function:

get_io_server_serialised_bytes()

subroutine io_server_state_reader_mod::get_io_server_serialised_bytes ( integer, intent(in)  ncid, integer, intent(in)  number_io_server, integer, intent(in)  my_io_server_rank, character(len=*), intent(in)  base_key, character, dimension(:), intent(out), allocatable  raw_bytes  )

private

Retrieves some IO server serialised bytes which will make up the state of a specific facet. Note that this uses the ISO C bindings in order to support 64 bit starts, counts & strides along with 64 bit scalar long variable fields.

Parameters

ncid	The NetCDF checkpoint file ID
number_io_server	The number of IO servers
my_io_server_rank	My IO server rank
base_key	The base look up key to retrieve the state which is stored in the file
raw_bytes	The retrieved raw bytes as they sit in the file, this is allocated to hold the required data

Definition at line 193 of file io_state_reader.F90.

    integer, intent(in) :: ncid, number_io_server, my_io_server_rank
    character(len=*), intent(in) :: base_key
    character, dimension(:), allocatable, intent(out) :: raw_bytes
    
    integer :: dim_id, var_id
    logical :: found
    integer(kind=8) :: dim_size, serialised_range(2), number_serialised_entries
 
    integer(kind=c_int)        :: cncid, cdimid, cstatus, cvarid
    integer(kind=c_size_t)     :: cdlen
    character(len=256) :: tmpname
    integer(KIND=c_size_t), target :: cstart(1), ccounts(1)
    Integer(KIND=c_ptrdiff_t), target :: cstrides(1)
    type(c_ptr) :: cstartptr, ccountsptr, cstridesptr
 
    cncid=ncid
    cstartptr=c_loc(cstart)
    ccountsptr=c_loc(ccounts)
    cstridesptr=c_loc(cstrides)
 
    call check_netcdf_status(nf90_inq_dimid(ncid, trim(base_key)//"_dim", dim_id), found)
    if (.not. found) return
    
    cdimid=dim_id-1
    call check_netcdf_status(nc_inq_dim(cncid, cdimid, tmpname, cdlen))
    dim_size=cdlen
 
    call check_netcdf_status(nf90_inq_varid(ncid, trim(base_key)//"_directory", var_id), found)
    if (.not. found) return
    cvarid=var_id-1
    cstart(1)=my_io_server_rank
    if (my_io_server_rank .lt. number_io_server-1) then
      ccounts(1)=2
      call check_netcdf_status(nc_get_vara_long(cncid, cvarid, cstartptr, ccountsptr, serialised_range))
      if (serialised_range(2) .gt. dim_size) then
        call log_log(log_error, "Serialised entry beyond size in the file")
      end if
    else
      ccounts(1)=1
      call check_netcdf_status(nc_get_vara_long(cncid, cvarid, cstartptr, ccountsptr, serialised_range))
      serialised_range(2)=dim_size
    end if    
    number_serialised_entries=(serialised_range(2)-serialised_range(1)) + 1    
    call check_netcdf_status(nf90_inq_varid(ncid, trim(base_key), var_id), found)
    if (.not. found) return
    allocate(raw_bytes(number_serialised_entries))
 
    cvarid=var_id-1
    cstart=serialised_range(1)-1
    ccounts=number_serialised_entries
    cstrides(1)=1
    call check_netcdf_status(nc_get_vars_text(cncid, cvarid, cstartptr, ccountsptr, cstridesptr, raw_bytes))

Here is the call graph for this function:

Here is the caller graph for this function:

reactivate_writer_federator_state()

subroutine, public io_server_state_reader_mod::reactivate_writer_federator_state	(	type(io_configuration_type), intent(inout)	io_configuration,
		type(writer_type), dimension(:)	writer_entries,
		type(hashmap_type), intent(inout), volatile	time_points
	)

Reactivates the writer federator and everything beneath it (i.e. just not the writer field manager.) For memory reasons this explicitly reopens the checkpoint file, will read each individual byte code entry in & repackage before deallocating memory and moving onto the next entry. The file is then closed.

Parameters

io_configuration	The IO server configuration
writer_entries	The configured but empty writer entries to unpack the state into
time_points	Time points to unpack the state into

Definition at line 105 of file io_state_reader.F90.

    type(io_configuration_type), intent(inout) :: io_configuration
    type(writer_type), dimension(:) :: writer_entries
    type(hashmap_type), volatile, intent(inout) :: time_points
 
    integer :: ncid, ierr, i
    character, dimension(:), allocatable :: raw_bytes
 
    call check_netcdf_status(nf90_open(path = options_get_string(io_configuration%options_database, "checkpoint"), &
         mode = nf90_nowrite, ncid = ncid))
 
    do i=1, size(writer_entries)
      ! Note that the different writer entries are dealt with separately for memory reasons
      if (writer_entries(i)%include_in_io_state_write) then
        call get_io_server_serialised_bytes(ncid, io_configuration%number_of_io_servers, io_configuration%my_io_rank, &
             "serialised_writer_entry_"//trim(conv_to_string(i)), raw_bytes)
        call unserialise_writer_type(writer_entries(i), raw_bytes)
        deallocate(raw_bytes)
      end if
    end do
 
    call get_io_server_serialised_bytes(ncid, io_configuration%number_of_io_servers, io_configuration%my_io_rank, &
         "serialised_timeaveraged_manipulation", raw_bytes)
    call restart_timeaveraged_state_from_checkpoint(raw_bytes)
    deallocate(raw_bytes)
 
    call get_io_server_serialised_bytes(ncid, io_configuration%number_of_io_servers, io_configuration%my_io_rank, &
         "serialised_instantaneous_manipulation", raw_bytes)
    call restart_instantaneous_state_from_checkpoint(raw_bytes)
    deallocate(raw_bytes)
 
    call get_io_server_serialised_bytes(ncid, io_configuration%number_of_io_servers, io_configuration%my_io_rank, &
         "serialised_timepoints", raw_bytes)
    call restart_writer_state_timepoints(time_points, raw_bytes)
    deallocate(raw_bytes)
    
    call check_netcdf_status(nf90_close(ncid))

Here is the call graph for this function:

Here is the caller graph for this function:

reactivate_writer_field_manager_state()

subroutine, public io_server_state_reader_mod::reactivate_writer_field_manager_state	(	type(io_configuration_type), intent(inout)	io_configuration,
		procedure(writer_field_manager_unserialise_state)	unserialise_writer_field_manager
	)

Reactivates the writer field manager state from the checkpoint file, for memory reasons this will open the file, read in and deserialise the byte code before closing it.

Parameters

io_configuration	IO server configuration
unserialise_writer_field_manager	Procedure pointer to the unserialisation for the writer field manager

Definition at line 148 of file io_state_reader.F90.

    type(io_configuration_type), intent(inout) :: io_configuration
    procedure(writer_field_manager_unserialise_state) :: unserialise_writer_field_manager
 
    integer :: ncid, ierr
    character, dimension(:), allocatable :: raw_bytes
 
    call check_netcdf_status(nf90_open(path = options_get_string(io_configuration%options_database, "checkpoint"), &
         mode = nf90_nowrite, ncid = ncid))
 
    call get_io_server_serialised_bytes(ncid, io_configuration%number_of_io_servers, io_configuration%my_io_rank, &
         "serialised_writer_manager", raw_bytes)
    call restart_writer_field_manager_from_checkpoint(unserialise_writer_field_manager, raw_bytes)
    deallocate(raw_bytes)
 
    call check_netcdf_status(nf90_close(ncid))

Here is the call graph for this function:

Here is the caller graph for this function:

read_io_server_configuration()

subroutine, public io_server_state_reader_mod::read_io_server_configuration	(	character(len=string_length), intent(in)	checkpoint_filename,
		character, dimension(:), intent(inout), allocatable	io_xml_configuration,
		integer, intent(in)	io_communicator_arg
	)

Reads the IO server configuration, which is the XML configuration initially run with and stored in the checkpoint. Note that this will open, read the XML in and then close the file.

Parameters

checkpoint_filename	The checkpoint filename to open and read from
io_xml_configuration	XML configuration is read from the checkpoint and placed into here
io_communicator_arg	The MPI IO server communicator

Definition at line 71 of file io_state_reader.F90.

    character(len=STRING_LENGTH), intent(in) :: checkpoint_filename
    character, dimension(:), allocatable, intent(inout) :: io_xml_configuration
    integer, intent(in) :: io_communicator_arg
 
    integer :: ncid, number_io_server, my_io_server_rank, ierr
    integer :: dim_id, dim_size
    logical :: found
 
    call mpi_comm_rank(io_communicator_arg, my_io_server_rank, ierr)
    call mpi_comm_size(io_communicator_arg, number_io_server, ierr)
    call check_netcdf_status(nf90_open(path = checkpoint_filename, mode = nf90_nowrite, ncid = ncid))
    call check_netcdf_status(nf90_inq_dimid(ncid, "entries_directory_dim", dim_id), found)
    if (.not. found) then
      if (my_io_server_rank==0) then
        call log_log(log_warn, "Restarting the IO server fresh as the checkpoint file does not contain IO state")
      end if      
      return
    end if    
    call check_netcdf_status(nf90_inquire_dimension(ncid, dim_id, len=dim_size))    
 
    if (dim_size .ne. number_io_server) then
      call log_log(log_error, "Can not restart IO server with a different number of IO servers")
    end if
    call get_io_server_configuration(ncid, io_xml_configuration)
    call check_netcdf_status(nf90_close(ncid))

Here is the call graph for this function:

Here is the caller graph for this function:

restart_instantaneous_state_from_checkpoint()

subroutine io_server_state_reader_mod::restart_instantaneous_state_from_checkpoint ( character, dimension(:), allocatable  raw_bytes )

private

Will restart the instantaneous manipulation state from the checkpoint file.

Parameters

raw_byte

The serialised byte state to unpackage and restart from

Definition at line 312 of file io_state_reader.F90.

    character, dimension(:), allocatable :: raw_bytes
    
    if (.not. allocated(raw_bytes)) then
      call log_master_log(log_warn, "On restart no instantaneous state in checkpoint file")
      return
    end if
    call unserialise_instantaneous_state(raw_bytes)

Here is the call graph for this function:

Here is the caller graph for this function:

restart_timeaveraged_state_from_checkpoint()

subroutine io_server_state_reader_mod::restart_timeaveraged_state_from_checkpoint ( character, dimension(:), allocatable  raw_bytes )

private

Will restart the time averaged manipulation state from the checkpoint file.

Parameters

raw_byte

The serialised byte state to unpackage and restart from

Definition at line 300 of file io_state_reader.F90.

    character, dimension(:), allocatable :: raw_bytes
    
    if (.not. allocated(raw_bytes)) then
      call log_master_log(log_warn, "On restart no time averaged state in checkpoint file")
      return
    end if
    call unserialise_time_averaged_state(raw_bytes)

Here is the call graph for this function:

Here is the caller graph for this function:

restart_writer_field_manager_from_checkpoint()

subroutine io_server_state_reader_mod::restart_writer_field_manager_from_checkpoint ( procedure(writer_field_manager_unserialise_state)  unserialise_writer_field_manager, character, dimension(:), allocatable  raw_bytes  )

private

Will restart the field manager state from the checkpoint file.

Parameters

unserialise_writer_field_manager	The unserialise field manager procedure, done this way due to module ordering
raw_byte	The serialised byte state to unpackage and restart from

Definition at line 325 of file io_state_reader.F90.

    procedure(writer_field_manager_unserialise_state) :: unserialise_writer_field_manager
    character, dimension(:), allocatable :: raw_bytes
 
    if (.not. allocated(raw_bytes)) then
      call log_master_log(log_warn, "On restart no writer field manager state in checkpoint file")
      return
    end if
    call unserialise_writer_field_manager(raw_bytes)

Here is the call graph for this function:

Here is the caller graph for this function:

restart_writer_state_from_checkpoint()

subroutine io_server_state_reader_mod::restart_writer_state_from_checkpoint ( type(writer_type), dimension(:)  writer_entries, character, dimension(:), allocatable  raw_bytes  )

private

Restarts the writer state from a specific checkpoint byte data chunk of memory.

Parameters

writer_entries	The array of writer entries to fill in
raw_byte	The serialised byte state to unpackage and restart from

Definition at line 251 of file io_state_reader.F90.

    type(writer_type), dimension(:) :: writer_entries
    character, dimension(:), allocatable :: raw_bytes
 
    integer :: i, number_entries, current_point, byte_size
    
    if (.not. allocated(raw_bytes)) then
      call log_master_log(log_warn, "On restart no writer state in checkpoint file")
      return
    end if
    current_point=1
    number_entries=unpack_scalar_integer_from_bytedata(raw_bytes, current_point)
    if (number_entries .ne. size(writer_entries)) then
      call log_log(log_error, "On restart have a different number of configured entries than those in the checkpoint file")
    end if
    do i=1, size(writer_entries)
      if (writer_entries(i)%include_in_io_state_write) then
        byte_size=unpack_scalar_integer_from_bytedata(raw_bytes, current_point)
        call unserialise_writer_type(writer_entries(i), raw_bytes(current_point:current_point+byte_size-1))
        current_point=current_point+byte_size
      end if
    end do

Here is the call graph for this function:

restart_writer_state_timepoints()

subroutine io_server_state_reader_mod::restart_writer_state_timepoints ( type(hashmap_type), intent(inout), volatile  time_points, character, dimension(:), allocatable  raw_bytes  )

private

Restarts the writer state timepoints held in the writer federator.

Parameters

time_points	The timepoints hashmap which is filled in from the serialised version
raw_byte	The serialised byte state to unpackage and restart from

Definition at line 278 of file io_state_reader.F90.

    type(hashmap_type), volatile, intent(inout) :: time_points
    character, dimension(:), allocatable :: raw_bytes
 
    integer :: i, number_entries, current_point, byte_size, timestep_key
    real(kind=default_precision) :: r_value
 
    if (.not. allocated(raw_bytes)) then
      call log_master_log(log_warn, "On restart no writer state timepoints in checkpoint file")
      return
    end if
    current_point=1
    number_entries=unpack_scalar_integer_from_bytedata(raw_bytes, current_point)
    do i=1, number_entries
      timestep_key=unpack_scalar_integer_from_bytedata(raw_bytes, current_point)
      r_value=unpack_scalar_dp_real_from_bytedata(raw_bytes, current_point)
      call c_put_real(time_points, conv_to_string(timestep_key), r_value)
    end do

Here is the call graph for this function:

Here is the caller graph for this function: