[8586] | 1 | %GENERIC cluster class definition
|
---|
[8578] | 2 | %
|
---|
| 3 | % Usage:
|
---|
[8586] | 4 | % cluster=generic('name','astrid',);
|
---|
| 5 | % cluster=generic('name','astrid','np',3);
|
---|
| 6 | % cluster=generic('name',oshostname(),'np',3,'login','username');
|
---|
[8578] | 7 |
|
---|
| 8 | classdef generic
|
---|
| 9 | properties (SetAccess=public)
|
---|
| 10 | % {{{1
|
---|
[8586] | 11 | name='';
|
---|
[8578] | 12 | login='';
|
---|
[8586] | 13 | np=1;
|
---|
[8578] | 14 | port=0;
|
---|
| 15 | interactive=1;
|
---|
| 16 | codepath=[issmtier() '/bin'];
|
---|
| 17 | executionpath=[issmdir() '/execution'];
|
---|
| 18 | valgrind=[issmtier() '/externalpackages/valgrind/install/bin/valgrind'];
|
---|
| 19 | valgrindlib=[issmtier() '/externalpackages/valgrind/install/lib/libmpidebug.so'];
|
---|
| 20 | valgrindsup=[issmtier() '/externalpackages/valgrind/issm.supp'];
|
---|
| 21 | %}}}
|
---|
| 22 | end
|
---|
| 23 | methods
|
---|
| 24 | function cluster=generic(varargin) % {{{1
|
---|
[8586] | 25 |
|
---|
| 26 | %use provided options to change fields
|
---|
| 27 | options=pairoptions(varargin{:});
|
---|
| 28 |
|
---|
| 29 | %get name
|
---|
| 30 | if ~exist(options,'name'), error('option ''name'' has not been provided'); end
|
---|
| 31 | cluster.name=getfieldvalue(options,'name');
|
---|
| 32 |
|
---|
[8578] | 33 | %initialize cluster using user settings if provided
|
---|
[8583] | 34 | if (exist([cluster.name '_settings'])==2), eval([cluster.name '_settings']); end
|
---|
[8578] | 35 |
|
---|
[8586] | 36 | %OK get other fields
|
---|
[8578] | 37 | for i=1:size(options.list,1),
|
---|
| 38 | fieldname=options.list{i,1};
|
---|
| 39 | fieldvalue=options.list{i,2};
|
---|
[8587] | 40 | if ismember(fieldname,properties('generic')),
|
---|
[8578] | 41 | cluster.(fieldname)=fieldvalue;
|
---|
| 42 | else
|
---|
| 43 | disp(['''' fieldname ''' is not a property of cluster generic']);
|
---|
| 44 | end
|
---|
| 45 | end
|
---|
| 46 | end
|
---|
| 47 | %}}}
|
---|
| 48 | function disp(cluster) % {{{1
|
---|
| 49 | % display the object
|
---|
| 50 | disp(sprintf('class ''%s'' object ''%s'' = ',class(cluster),inputname(1)));
|
---|
| 51 | disp(sprintf(' name: %s',cluster.name));
|
---|
| 52 | disp(sprintf(' login: %s',cluster.login));
|
---|
| 53 | disp(sprintf(' np: %i',cluster.np));
|
---|
| 54 | disp(sprintf(' port: %i',cluster.port));
|
---|
| 55 | disp(sprintf(' codepath: %s',cluster.codepath));
|
---|
| 56 | disp(sprintf(' executionpath: %s',cluster.executionpath));
|
---|
| 57 | disp(sprintf(' valgrind: %s',cluster.valgrind));
|
---|
| 58 | disp(sprintf(' valgrindlib: %s',cluster.valgrindlib));
|
---|
| 59 | disp(sprintf(' valgrindsup: %s',cluster.valgrindsup));
|
---|
| 60 | end
|
---|
| 61 | %}}}
|
---|
| 62 | function IsConsistent(cluster) % {{{1
|
---|
[8586] | 63 | if cluster.np<1
|
---|
| 64 | error(['number of processors should be at least 1']);
|
---|
[8578] | 65 | end
|
---|
| 66 | if isnan(cluster.np),
|
---|
| 67 | error('number of processors should not be NaN!');
|
---|
| 68 | end
|
---|
| 69 | end
|
---|
| 70 | %}}}
|
---|
| 71 | function BuildQueueScript(cluster,md) % {{{1
|
---|
| 72 |
|
---|
| 73 | %retrieve parameters
|
---|
| 74 | modelname=md.name;
|
---|
| 75 | solution_type=md.solution_type;
|
---|
[9565] | 76 | isvalgrind=md.debug.valgrind;
|
---|
| 77 | isgprof=md.debug.gprof;
|
---|
[8578] | 78 |
|
---|
| 79 | %open file for writing:
|
---|
| 80 | fid=fopen([modelname '.queue'],'w');
|
---|
| 81 |
|
---|
| 82 | %write instructions for launching a job on the cluster
|
---|
| 83 | fprintf(fid,'#!/bin/sh\n');
|
---|
[9565] | 84 | if ~isvalgrind,
|
---|
[8578] | 85 | if cluster.interactive
|
---|
[9561] | 86 | fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s ',...
|
---|
| 87 | cluster.np,cluster.codepath,EnumToString(solution_type),cluster.executionpath,modelname);
|
---|
[8578] | 88 | else
|
---|
[9561] | 89 | fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 90 | cluster.np,cluster.codepath,EnumToString(solution_type),cluster.executionpath,modelname,modelname,modelname);
|
---|
[8578] | 91 | end
|
---|
| 92 | else
|
---|
[9561] | 93 | %fprintf(fid,'LD_PRELOAD=%s mpiexec -np %i %s --leak-check=full --gen-suppressions=all --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',cluster.valgrindlib,cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution_type),cluster.executionpath,modelname,modelname,modelname);
|
---|
| 94 | fprintf(fid,'LD_PRELOAD=%s mpiexec -np %i %s --leak-check=full --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 95 | cluster.valgrindlib,cluster.np,cluster.valgrind,cluster.valgrindsup, cluster.codepath,EnumToString(solution_type),cluster.executionpath,modelname,modelname,modelname);
|
---|
[8578] | 96 | end
|
---|
| 97 |
|
---|
[9565] | 98 | if isgprof,
|
---|
[8578] | 99 | fprintf(fid,'\n gprof %s/issm.exe gmon.out > %s.performance',cluster.codepath,modelname);
|
---|
| 100 | end
|
---|
| 101 |
|
---|
| 102 | if ~md.io_gather,
|
---|
| 103 | %concatenate the output files:
|
---|
| 104 | fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
|
---|
| 105 | end
|
---|
| 106 |
|
---|
| 107 | %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 108 | if cluster.interactive,
|
---|
| 109 | fid=fopen([modelname '.errlog'],'w'); fclose(fid);
|
---|
| 110 | fid=fopen([modelname '.outlog'],'w'); fclose(fid);
|
---|
| 111 | end
|
---|
| 112 |
|
---|
| 113 | end
|
---|
| 114 | %}}}
|
---|
| 115 | function LaunchQueueJob(cluster,md,options)% {{{1
|
---|
| 116 |
|
---|
| 117 | %lauch command, to be executed via ssh
|
---|
| 118 | launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' md.runtimename ' && mkdir ' md.runtimename ...
|
---|
| 119 | ' && cd ' md.runtimename ' && mv ../' md.runtimename '.tar.gz ./ && tar -zxf ' md.runtimename '.tar.gz && source ' md.name '.queue '];
|
---|
| 120 |
|
---|
| 121 | if ~strcmpi(options.batch,'yes'),
|
---|
| 122 |
|
---|
| 123 | %compress the files into one zip.
|
---|
| 124 | compressstring=['tar -zcf ' md.runtimename '.tar.gz ' md.name '.bin ' md.name '.queue ' md.name '.petsc '];
|
---|
[9571] | 125 | if md.dakota_analysis,
|
---|
[8578] | 126 | compressstring=[compressstring md.name '.qmu.in'];
|
---|
| 127 | end
|
---|
| 128 | if cluster.interactive,
|
---|
| 129 | compressstring=[compressstring ' ' md.name '.errlog ' md.name '.outlog '];
|
---|
| 130 | end
|
---|
| 131 | system(compressstring);
|
---|
| 132 |
|
---|
| 133 | disp('uploading input file and queueing script');
|
---|
| 134 | issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[md.runtimename '.tar.gz']});
|
---|
| 135 |
|
---|
| 136 | disp('launching solution sequence on remote cluster');
|
---|
| 137 | issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
|
---|
| 138 | else
|
---|
| 139 | disp('batch mode requested: not launching job interactively');
|
---|
| 140 | disp('launch solution sequence on remote cluster by hand');
|
---|
| 141 | end
|
---|
| 142 |
|
---|
| 143 | end %}}}
|
---|
| 144 | function Download(cluster,md)% {{{1
|
---|
| 145 |
|
---|
| 146 | %some check
|
---|
| 147 | if isempty(md.runtimename),
|
---|
| 148 | error('supply runtime name for results to be loaded!');
|
---|
| 149 | end
|
---|
| 150 |
|
---|
| 151 | %Figure out the directory where all the files are in:
|
---|
| 152 | directory=[cluster.executionpath '/' md.runtimename '/'];
|
---|
| 153 |
|
---|
| 154 | %What packages are we picking up from remote cluster
|
---|
| 155 | packages={[md.name '.outlog'],[md.name '.errlog']};
|
---|
[9571] | 156 | if md.dakota_analysis,
|
---|
[8578] | 157 | packages{end+1}=[md.name '.qmu.err'];
|
---|
| 158 | packages{end+1}=[md.name '.qmu.out'];
|
---|
| 159 | if isfield(md.qmu_params,'tabular_graphics_data'),
|
---|
| 160 | if md.qmu_params.tabular_graphics_data==true,
|
---|
| 161 | packages{end+1}='dakota_tabular.dat';
|
---|
| 162 | end
|
---|
| 163 | end
|
---|
| 164 | else
|
---|
| 165 | packages{end+1}=[md.name '.outbin'];
|
---|
| 166 | end
|
---|
| 167 |
|
---|
| 168 | %copy files from cluster to present directory
|
---|
| 169 | issmscpin(cluster.name, cluster.login, cluster.port, directory, packages);
|
---|
| 170 | end %}}}
|
---|
| 171 | end
|
---|
| 172 | end
|
---|