Context Navigation

source: issm/trunk/src/m/classes/clusters/generic.m@ 20500

Last change on this file since 20500 was 20500, checked in by Mathieu Morlighem, 9 years ago
merged trunk-jpl and trunk for revision 20497
File size: 11.8 KB

Rev	Line
[8586]	1	%GENERIC cluster class definition
[8578]	2	%
	3	% Usage:
[8586]	4	% cluster=generic('name','astrid','np',3);
	5	% cluster=generic('name',oshostname(),'np',3,'login','username');
[8578]	6
	7	classdef generic
[13395]	8	properties (SetAccess=public)
	9	% {{{
	10	name='';
	11	login='';
	12	np=1;
	13	port=0;
	14	interactive=1;
	15	codepath=[issmdir() '/bin'];
[13975]	16	etcpath=[issmdir() '/etc'];
[13395]	17	executionpath=[issmdir() '/execution'];
	18	valgrind=[issmdir() '/externalpackages/valgrind/install/bin/valgrind'];
	19	valgrindlib=[issmdir() '/externalpackages/valgrind/install/lib/libmpidebug.so'];
	20	valgrindsup=[issmdir() '/externalpackages/valgrind/issm.supp'];
[17806]	21	verbose=1;
[18301]	22	shell='/bin/sh';
[13395]	23	%}}}
	24	end
	25	methods
	26	function cluster=generic(varargin) % {{{
[8586]	27
[19105]	28	%Change the defaults if ispc
	29	if ispc,
	30	cluster.codepath = [issmdir() '\bin'];
	31	cluster.etcpath = [issmdir() '\etc'];
	32	cluster.executionpath = [issmdir() '\execution'];
	33	end
	34
[13395]	35	%use provided options to change fields
	36	options=pairoptions(varargin{:});
[8586]	37
[13395]	38	%get name
	39	cluster.name=getfieldvalue(options,'name',oshostname());
[8586]	40
[13395]	41	%initialize cluster using user settings if provided
	42	if (exist([cluster.name '_settings'])==2), eval([cluster.name '_settings']); end
[8578]	43
[13395]	44	%OK get other fields
	45	cluster=AssignObjectFields(pairoptions(varargin{:}),cluster);
	46	end
	47	%}}}
	48	function disp(cluster) % {{{
	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));
[20500]	57	disp(sprintf(' etcpath: %s',cluster.etcpath));
[13395]	58	disp(sprintf(' valgrind: %s',cluster.valgrind));
	59	disp(sprintf(' valgrindlib: %s',cluster.valgrindlib));
	60	disp(sprintf(' valgrindsup: %s',cluster.valgrindsup));
[17806]	61	disp(sprintf(' verbose: %s',cluster.verbose));
[18301]	62	disp(sprintf(' shell: %s',cluster.shell));
[13395]	63	end
	64	%}}}
	65	function md = checkconsistency(cluster,md,solution,analyses) % {{{
	66	if cluster.np<1
	67	md = checkmessage(md,['number of processors should be at least 1']);
	68	end
	69	if isnan(cluster.np),
	70	md = checkmessage(md,'number of processors should not be NaN!');
	71	end
	72	end
	73	%}}}
[20500]	74	function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota) % {{{
[8578]	75
[13395]	76	%write queuing script
[20500]	77	%what is the executable being called?
	78	executable='issm.exe';
	79	if isdakota,
	80	version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3));
	81	if (version>=6),
	82	executable='issm_dakota.exe';
	83	end
	84	end
	85
[13395]	86	if ~ispc(),
[12706]	87
[13395]	88	fid=fopen([modelname '.queue'],'w');
[18301]	89	fprintf(fid,'#!%s\n',cluster.shell);
[13395]	90	if ~isvalgrind,
	91	if cluster.interactive
[16137]	92	if IssmConfig('_HAVE_MPI_'),
[20500]	93	fprintf(fid,'mpiexec -np %i %s/%s %s %s %s \n',cluster.np,cluster.codepath,executable,EnumToString(solution),[cluster.executionpath '/' dirname],modelname);
[13395]	94	else
[20500]	95	fprintf(fid,'%s/%s %s %s %s ',cluster.codepath,executable,EnumToString(solution),[cluster.executionpath '/' dirname],modelname);
[13395]	96	end
	97	else
[16137]	98	if IssmConfig('_HAVE_MPI_'),
[20500]	99	fprintf(fid,'mpiexec -np %i %s/%s %s %s %s 2> %s.errlog >%s.outlog &',cluster.np,cluster.codepath,executable,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
[13395]	100	else
[20500]	101	fprintf(fid,'%s/%s %s %s %s 2> %s.errlog >%s.outlog &',cluster.codepath,executable,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
[13395]	102	end
	103	end
	104	elseif isgprof,
	105	fprintf(fid,'\n gprof %s/issm.exe gmon.out > %s.performance',cluster.codepath,modelname);
	106	else
	107	%Add --gen-suppressions=all to get suppression lines
	108	fprintf(fid,'LD_PRELOAD=%s \\\n',cluster.valgrindlib);
	109	if ismac,
[16137]	110	if IssmConfig('_HAVE_MPI_'),
[20500]	111	fprintf(fid,'mpiexec -np %i %s --leak-check=full --error-limit=no --dsymutil=yes --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
[13395]	112	cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname], modelname,modelname,modelname);
	113	else
[20500]	114	fprintf(fid,'%s --leak-check=full --dsymutil=yes --error-limit=no --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
[13395]	115	cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname], modelname,modelname,modelname);
	116	end
	117	else
[16137]	118	if IssmConfig('_HAVE_MPI_'),
[20500]	119	fprintf(fid,'mpiexec -np %i %s --leak-check=full --error-limit=no --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
[13395]	120	cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
	121	else
[20500]	122	fprintf(fid,'%s --leak-check=full --error-limit=no --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
[13395]	123	cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
	124	end
	125	end
	126	end
	127	if ~io_gather, %concatenate the output files:
	128	fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
	129	end
	130	fclose(fid);
[12706]	131
[13395]	132	else % Windows
[12706]	133
[13395]	134	fid=fopen([modelname '.bat'],'w');
	135	fprintf(fid,'@echo off\n');
[15396]	136
[20500]	137	if IssmConfig('_HAVE_PETSC_MPI_'),
	138	warning('parallel runs not allowed yet in Windows. Defaulting to 1 cpus');
	139	cluster.np=1;
	140	end
[17806]	141
[15396]	142	if cluster.np>1,
	143	fprintf(fid,'"C:\\Program Files\\MPICH2\\bin\\mpiexec.exe" -n %i "%s/issm.exe" %s ./ %s ',cluster.np,cluster.codepath,EnumToString(solution),modelname);
[13395]	144	else
[15396]	145	fprintf(fid,'"%s/issm.exe" %s ./ %s ',cluster.codepath,EnumToString(solution),modelname);
[13395]	146	end
	147	fclose(fid);
	148	end
[8578]	149
[13395]	150	%in interactive mode, create a run file, and errlog and outlog file
	151	if cluster.interactive,
	152	fid=fopen([modelname '.errlog'],'w'); fclose(fid);
	153	fid=fopen([modelname '.outlog'],'w'); fclose(fid);
	154	end
	155	end
	156	%}}}
[20500]	157	function BuildQueueScriptMultipleModels(cluster,dirname,modelname,solution,dirnames,modelnames,nps) % {{{
	158
	159	%some checks:
	160	if isempty(modelname), error('BuildQueueScriptMultipleModels error message: need a non empty model name!');end
	161
	162	%what is the executable being called?
	163	executable='issm_slr.exe';
	164
	165	if ispc(), error('BuildQueueScriptMultipleModels not support yet on windows machines');end;
	166
	167	%write queuing script
	168	fid=fopen([modelname '.queue'],'w');
	169
	170	fprintf(fid,'#!%s\n',cluster.shell);
	171
	172	%number of cpus:
	173	mpistring=sprintf('mpiexec -np %i ',cluster.np);
	174
	175	%executable:
	176	mpistring=[mpistring sprintf('%s/%s ',cluster.codepath,executable)];
	177
	178	%solution name:
	179	mpistring=[mpistring sprintf('%s ',EnumToString(solution))];
	180
	181	%execution directory and model name:
	182	mpistring=[mpistring sprintf('%s/%s %s',cluster.executionpath,dirname,modelname)];
	183
	184	%inform main executable of how many icecaps, glaciers and earth models are being run:
	185	mpistring=[mpistring sprintf(' %i ',length(dirnames))];
	186
	187	%icecaps, glaciers and earth location, names and number of processors associated:
	188	for i=1:length(dirnames),
	189	mpistring=[mpistring sprintf(' %s/%s %s %i ',cluster.executionpath,dirnames{i},modelnames{i},nps{i})];
	190	end
	191
	192	%log files:
	193	if ~cluster.interactive,
	194	mpistring=[mpistring sprintf('2> %s.errlog> %s.outlog',modelname,modelname)];
	195	end
	196
	197	%write this long string to disk:
	198	fprintf(fid,mpistring);
	199	fclose(fid);
	200
	201	%in interactive mode, create a run file, and errlog and outlog file
	202	if cluster.interactive,
	203	fid=fopen([modelname '.errlog'],'w'); fclose(fid);
	204	fid=fopen([modelname '.outlog'],'w'); fclose(fid);
	205	end
	206	end
	207	%}}}
[13395]	208	function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{
[12706]	209
[13395]	210	%write queuing script
	211	if ~ispc(),
[12706]	212
[13395]	213	fid=fopen([modelname '.queue'],'w');
	214	fprintf(fid,'#!/bin/sh\n');
	215	if ~isvalgrind,
	216	if cluster.interactive
	217	fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s ',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname);
	218	else
	219	fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s 2> %s.errlog >%s.outlog ',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
	220	end
	221	elseif isgprof,
	222	fprintf(fid,'\n gprof %s/kriging.exe gmon.out > %s.performance',cluster.codepath,modelname);
	223	else
	224	%Add --gen-suppressions=all to get suppression lines
	225	fprintf(fid,'LD_PRELOAD=%s \\\n',cluster.valgrindlib);
	226	fprintf(fid,'mpiexec -np %i %s --leak-check=full --suppressions=%s %s/kriging.exe %s %s 2> %s.errlog >%s.outlog ',...
	227	cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
	228	end
	229	if ~io_gather, %concatenate the output files:
	230	fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
	231	end
	232	fclose(fid);
[11527]	233
[13395]	234	else % Windows
[8578]	235
[13395]	236	fid=fopen([modelname '.bat'],'w');
	237	fprintf(fid,'@echo off\n');
	238	if cluster.interactive
	239	fprintf(fid,'"%s/issm.exe" %s "%s" %s ',cluster.codepath,EnumToString(solution),[cluster.executionpath '/' modelname],modelname);
	240	else
	241	fprintf(fid,'"%s/issm.exe" %s "%s" %s 2> %s.errlog >%s.outlog',...
	242	cluster.codepath,EnumToString(solution),[cluster.executionpath '/' modelname],modelname,modelname,modelname);
	243	end
	244	fclose(fid);
	245	end
[8578]	246
[13395]	247	%in interactive mode, create a run file, and errlog and outlog file
	248	if cluster.interactive,
	249	fid=fopen([modelname '.errlog'],'w'); fclose(fid);
	250	fid=fopen([modelname '.outlog'],'w'); fclose(fid);
	251	end
	252	end
	253	%}}}
[20500]	254	function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{
[8578]	255
[15396]	256	if ~ispc,
[18301]	257
[15396]	258	%compress the files into one zip.
	259	compressstring=['tar -zcf ' dirname '.tar.gz '];
	260	for i=1:numel(filelist),
	261	compressstring = [compressstring ' ' filelist{i}];
	262	end
	263	if cluster.interactive,
	264	compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog '];
	265	end
	266	system(compressstring);
[11527]	267
[17806]	268	if cluster.verbose, disp('uploading input file and queueing script'); end
[15396]	269	issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']});
[20500]	270	end
	271	end %}}}
	272	function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
[11527]	273
[20500]	274	if ~ispc,
	275
	276	%figure out what shell extension we will use:
	277	if isempty(strfind(cluster.shell,'csh')),
	278	shellext='sh';
	279	else
	280	shellext='csh';
	281	end
	282
[17806]	283	if cluster.verbose, disp('launching solution sequence on remote cluster'); end
[20500]	284
	285	if ~isempty(restart)
	286	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && cd ' dirname ' && source ' modelname '.queue '];
	287	else
	288	if ~batch,
	289	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
	290	' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && source ' modelname '.queue '];
	291	else
	292	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
	293	' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz '];
	294	end
	295	end
[15396]	296	issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
	297	else
	298	system([modelname '.bat']);
	299	end
[20500]	300
[13395]	301	end %}}}
	302	function Download(cluster,dirname,filelist)% {{{
[11527]	303
[13395]	304	if ispc(),
[12706]	305	%do nothing
	306	return;
[8578]	307	end
	308
[12706]	309	%copy files from cluster to current directory
	310	directory=[cluster.executionpath '/' dirname '/'];
	311	issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist);
[11527]	312	end %}}}
[8578]	313	end
	314	end

Note: See TracBrowser for help on using the repository browser.

Download in other formats: