Context Navigation

source: issm/trunk/src/m/classes/clusters/generic.m@ 22822

Last change on this file since 22822 was 22822, checked in by Mathieu Morlighem, 7 years ago
merged trunk-jpl and trunk for revision 22820
File size: 14.1 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	% {{{
[21341]	10	name = '';
	11	login = '';
	12	np = 1;
[22758]	13	npocean = 1;
[21341]	14	port = 0;
	15	interactive = 1;
	16	codepath = [IssmConfig('ISSM_PREFIX') '/bin'];
	17	etcpath = [issmdir() '/etc'];
	18	executionpath = [issmdir() '/execution'];
	19	valgrind = [issmdir() '/externalpackages/valgrind/install/bin/valgrind'];
	20	valgrindlib = [issmdir() '/externalpackages/valgrind/install/lib/libmpidebug.so'];
	21	valgrindsup = [issmdir() '/externalpackages/valgrind/issm.supp'];
	22	verbose = 1;
	23	shell = '/bin/sh';
[13395]	24	%}}}
	25	end
	26	methods
	27	function cluster=generic(varargin) % {{{
[8586]	28
[19105]	29	%Change the defaults if ispc
	30	if ispc,
	31	cluster.codepath = [issmdir() '\bin'];
	32	cluster.etcpath = [issmdir() '\etc'];
	33	cluster.executionpath = [issmdir() '\execution'];
	34	end
	35
[13395]	36	%use provided options to change fields
	37	options=pairoptions(varargin{:});
[8586]	38
[13395]	39	%get name
	40	cluster.name=getfieldvalue(options,'name',oshostname());
[8586]	41
[13395]	42	%initialize cluster using user settings if provided
	43	if (exist([cluster.name '_settings'])==2), eval([cluster.name '_settings']); end
[8578]	44
[13395]	45	%OK get other fields
	46	cluster=AssignObjectFields(pairoptions(varargin{:}),cluster);
	47	end
	48	%}}}
	49	function disp(cluster) % {{{
	50	% display the object
	51	disp(sprintf('class ''%s'' object ''%s'' = ',class(cluster),inputname(1)));
	52	disp(sprintf(' name: %s',cluster.name));
	53	disp(sprintf(' login: %s',cluster.login));
	54	disp(sprintf(' np: %i',cluster.np));
[22758]	55	disp(sprintf(' npocean: %i',cluster.npocean));
[13395]	56	disp(sprintf(' port: %i',cluster.port));
	57	disp(sprintf(' codepath: %s',cluster.codepath));
	58	disp(sprintf(' executionpath: %s',cluster.executionpath));
[20500]	59	disp(sprintf(' etcpath: %s',cluster.etcpath));
[13395]	60	disp(sprintf(' valgrind: %s',cluster.valgrind));
	61	disp(sprintf(' valgrindlib: %s',cluster.valgrindlib));
	62	disp(sprintf(' valgrindsup: %s',cluster.valgrindsup));
[17806]	63	disp(sprintf(' verbose: %s',cluster.verbose));
[18301]	64	disp(sprintf(' shell: %s',cluster.shell));
[13395]	65	end
	66	%}}}
	67	function md = checkconsistency(cluster,md,solution,analyses) % {{{
	68	if cluster.np<1
	69	md = checkmessage(md,['number of processors should be at least 1']);
	70	end
	71	if isnan(cluster.np),
	72	md = checkmessage(md,'number of processors should not be NaN!');
	73	end
	74	end
	75	%}}}
[21729]	76	function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{
[8578]	77
[13395]	78	%write queuing script
[20500]	79	%what is the executable being called?
	80	executable='issm.exe';
	81	if isdakota,
	82	version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3));
	83	if (version>=6),
	84	executable='issm_dakota.exe';
	85	end
	86	end
[21729]	87	if isoceancoupling,
	88	executable='issm_ocean.exe';
	89	end
[20500]	90
[13395]	91	if ~ispc(),
[12706]	92
[13395]	93	fid=fopen([modelname '.queue'],'w');
[18301]	94	fprintf(fid,'#!%s\n',cluster.shell);
[13395]	95	if ~isvalgrind,
	96	if cluster.interactive
[16137]	97	if IssmConfig('_HAVE_MPI_'),
[21341]	98	fprintf(fid,'mpiexec -np %i %s/%s %s %s %s \n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname);
[13395]	99	else
[21341]	100	fprintf(fid,'%s/%s %s %s %s ',cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname);
[13395]	101	end
	102	else
[16137]	103	if IssmConfig('_HAVE_MPI_'),
[21341]	104	fprintf(fid,'mpiexec -np %i %s/%s %s %s %s 2> %s.errlog >%s.outlog &',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname,modelname,modelname);
[13395]	105	else
[21341]	106	fprintf(fid,'%s/%s %s %s %s 2> %s.errlog >%s.outlog &',cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname,modelname,modelname);
[13395]	107	end
	108	end
	109	elseif isgprof,
	110	fprintf(fid,'\n gprof %s/issm.exe gmon.out > %s.performance',cluster.codepath,modelname);
	111	else
	112	%Add --gen-suppressions=all to get suppression lines
	113	fprintf(fid,'LD_PRELOAD=%s \\\n',cluster.valgrindlib);
	114	if ismac,
[16137]	115	if IssmConfig('_HAVE_MPI_'),
[22758]	116	fprintf(fid,'mpiexec -np %i %s --leak-check=full --error-limit=no --dsymutil=yes --suppressions=%s %s/%s %s %s %s 2> %s.errlog >%s.outlog ',...
	117	cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname], modelname,modelname,modelname);
[13395]	118	else
[22758]	119	fprintf(fid,'%s --leak-check=full --dsymutil=yes --error-limit=no --suppressions=%s %s/%s %s %s %s 2> %s.errlog >%s.outlog ',...
	120	cluster.valgrind,cluster.valgrindsup,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname], modelname,modelname,modelname);
[13395]	121	end
	122	else
[16137]	123	if IssmConfig('_HAVE_MPI_'),
[22758]	124	fprintf(fid,'mpiexec -np %i %s --leak-check=full --error-limit=no --suppressions=%s %s/%s %s %s %s 2> %s.errlog >%s.outlog ',...
	125	cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname,modelname,modelname);
[13395]	126	else
[22758]	127	fprintf(fid,'%s --leak-check=full --error-limit=no --suppressions=%s %s/%s %s %s %s 2> %s.errlog >%s.outlog ',...
	128	cluster.valgrind,cluster.valgrindsup,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname,modelname,modelname);
[13395]	129	end
	130	end
	131	end
	132	if ~io_gather, %concatenate the output files:
	133	fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
	134	end
	135	fclose(fid);
[12706]	136
[13395]	137	else % Windows
[12706]	138
[13395]	139	fid=fopen([modelname '.bat'],'w');
	140	fprintf(fid,'@echo off\n');
[15396]	141
[20500]	142	if IssmConfig('_HAVE_PETSC_MPI_'),
	143	warning('parallel runs not allowed yet in Windows. Defaulting to 1 cpus');
	144	cluster.np=1;
	145	end
[17806]	146
[15396]	147	if cluster.np>1,
[22758]	148	fprintf(fid,'"C:\\Program Files\\MPICH2\\bin\\mpiexec.exe" -n %i "%s/%s" %s ./ %s ',cluster.np,cluster.codepath,executable,solution,modelname);
[13395]	149	else
[22758]	150	fprintf(fid,'"%s/%s" %s ./ %s ',cluster.codepath,executable,solution,modelname);
[13395]	151	end
	152	fclose(fid);
	153	end
[8578]	154
[13395]	155	%in interactive mode, create a run file, and errlog and outlog file
	156	if cluster.interactive,
	157	fid=fopen([modelname '.errlog'],'w'); fclose(fid);
	158	fid=fopen([modelname '.outlog'],'w'); fclose(fid);
	159	end
	160	end
	161	%}}}
[20500]	162	function BuildQueueScriptMultipleModels(cluster,dirname,modelname,solution,dirnames,modelnames,nps) % {{{
	163
	164	%some checks:
	165	if isempty(modelname), error('BuildQueueScriptMultipleModels error message: need a non empty model name!');end
	166
	167	%what is the executable being called?
	168	executable='issm_slr.exe';
	169
	170	if ispc(), error('BuildQueueScriptMultipleModels not support yet on windows machines');end;
	171
	172	%write queuing script
	173	fid=fopen([modelname '.queue'],'w');
	174
	175	fprintf(fid,'#!%s\n',cluster.shell);
	176
	177	%number of cpus:
	178	mpistring=sprintf('mpiexec -np %i ',cluster.np);
	179
	180	%executable:
	181	mpistring=[mpistring sprintf('%s/%s ',cluster.codepath,executable)];
	182
	183	%solution name:
[21341]	184	mpistring=[mpistring sprintf('%s ',solution)];
[20500]	185
	186	%execution directory and model name:
	187	mpistring=[mpistring sprintf('%s/%s %s',cluster.executionpath,dirname,modelname)];
	188
	189	%inform main executable of how many icecaps, glaciers and earth models are being run:
	190	mpistring=[mpistring sprintf(' %i ',length(dirnames))];
	191
	192	%icecaps, glaciers and earth location, names and number of processors associated:
	193	for i=1:length(dirnames),
	194	mpistring=[mpistring sprintf(' %s/%s %s %i ',cluster.executionpath,dirnames{i},modelnames{i},nps{i})];
	195	end
	196
	197	%log files:
	198	if ~cluster.interactive,
	199	mpistring=[mpistring sprintf('2> %s.errlog> %s.outlog',modelname,modelname)];
	200	end
	201
	202	%write this long string to disk:
	203	fprintf(fid,mpistring);
	204	fclose(fid);
	205
	206	%in interactive mode, create a run file, and errlog and outlog file
	207	if cluster.interactive,
	208	fid=fopen([modelname '.errlog'],'w'); fclose(fid);
	209	fid=fopen([modelname '.outlog'],'w'); fclose(fid);
	210	end
	211	end
	212	%}}}
[22758]	213	function BuildQueueScriptIceOcean(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota) % {{{
	214
	215	%write queuing script
	216	%what is the executable being called?
	217	executable='issm_ocean.exe';
	218
	219	fid=fopen([modelname '.queue'],'w');
	220	fprintf(fid,'#!%s\n',cluster.shell);
[22822]	221	if ~isvalgrind,
	222	fprintf(fid,'mpiexec -np %i %s/%s %s %s %s : -np %i ./mitgcmuv\n',cluster.np,cluster.codepath,executable,solution,cluster.executionpath,modelname,cluster.npocean);
	223
	224	else
	225	fprintf(fid,'mpiexec -np %i %s --leak-check=full --error-limit=no --dsymutil=yes --suppressions=%s %s/%s %s %s %s : -np %i ./mitgcmuv\n',...
	226	cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,executable,solution,cluster.executionpath,modelname,cluster.npocean);
	227	end
[22758]	228	fclose(fid);
	229
	230	%in interactive mode, create a run file, and errlog and outlog file
	231	if cluster.interactive,
	232	fid=fopen([modelname '.errlog'],'w'); fclose(fid);
	233	fid=fopen([modelname '.outlog'],'w'); fclose(fid);
	234	end
	235	end
	236	%}}}
[13395]	237	function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{
[12706]	238
[13395]	239	%write queuing script
	240	if ~ispc(),
[12706]	241
[13395]	242	fid=fopen([modelname '.queue'],'w');
	243	fprintf(fid,'#!/bin/sh\n');
	244	if ~isvalgrind,
	245	if cluster.interactive
	246	fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s ',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname);
	247	else
	248	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);
	249	end
	250	elseif isgprof,
	251	fprintf(fid,'\n gprof %s/kriging.exe gmon.out > %s.performance',cluster.codepath,modelname);
	252	else
	253	%Add --gen-suppressions=all to get suppression lines
	254	fprintf(fid,'LD_PRELOAD=%s \\\n',cluster.valgrindlib);
	255	fprintf(fid,'mpiexec -np %i %s --leak-check=full --suppressions=%s %s/kriging.exe %s %s 2> %s.errlog >%s.outlog ',...
	256	cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
	257	end
	258	if ~io_gather, %concatenate the output files:
	259	fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
	260	end
	261	fclose(fid);
[11527]	262
[13395]	263	else % Windows
[8578]	264
[13395]	265	fid=fopen([modelname '.bat'],'w');
	266	fprintf(fid,'@echo off\n');
	267	if cluster.interactive
[21341]	268	fprintf(fid,'"%s/issm.exe" %s "%s" %s ',cluster.codepath,solution,[cluster.executionpath '/' modelname],modelname);
[13395]	269	else
	270	fprintf(fid,'"%s/issm.exe" %s "%s" %s 2> %s.errlog >%s.outlog',...
[21341]	271	cluster.codepath,solution,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
[13395]	272	end
	273	fclose(fid);
	274	end
[8578]	275
[13395]	276	%in interactive mode, create a run file, and errlog and outlog file
	277	if cluster.interactive,
	278	fid=fopen([modelname '.errlog'],'w'); fclose(fid);
	279	fid=fopen([modelname '.outlog'],'w'); fclose(fid);
	280	end
	281	end
	282	%}}}
[20500]	283	function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{
[8578]	284
[15396]	285	if ~ispc,
[18301]	286
[15396]	287	%compress the files into one zip.
	288	compressstring=['tar -zcf ' dirname '.tar.gz '];
	289	for i=1:numel(filelist),
[22758]	290	if ~exist(filelist{i},'file')
	291	error(['File ' filelist{i} ' not found']);
	292	end
[15396]	293	compressstring = [compressstring ' ' filelist{i}];
	294	end
	295	if cluster.interactive,
	296	compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog '];
	297	end
	298	system(compressstring);
[11527]	299
[17806]	300	if cluster.verbose, disp('uploading input file and queueing script'); end
[15396]	301	issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']});
[20500]	302	end
	303	end %}}}
	304	function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
[11527]	305
[20500]	306	if ~ispc,
	307
	308	%figure out what shell extension we will use:
	309	if isempty(strfind(cluster.shell,'csh')),
	310	shellext='sh';
	311	else
	312	shellext='csh';
	313	end
	314
[17806]	315	if cluster.verbose, disp('launching solution sequence on remote cluster'); end
[20500]	316
	317	if ~isempty(restart)
	318	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && cd ' dirname ' && source ' modelname '.queue '];
	319	else
	320	if ~batch,
	321	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
	322	' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && source ' modelname '.queue '];
	323	else
	324	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
	325	' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz '];
	326	end
	327	end
[15396]	328	issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
	329	else
	330	system([modelname '.bat']);
	331	end
[20500]	332
[13395]	333	end %}}}
[22758]	334	function LaunchQueueJobIceOcean(cluster,modelname,dirname,filelist,restart,batch)% {{{
	335
	336	if ~ispc,
	337
	338	%figure out what shell extension we will use:
	339	if isempty(strfind(cluster.shell,'csh')),
	340	shellext='sh';
	341	else
	342	shellext='csh';
	343	end
	344
	345	if cluster.verbose, disp('launching solution sequence on remote cluster'); end
	346
	347	if ~isempty(restart)
	348	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && cd ' dirname ' && source ' modelname '.queue '];
	349	else
	350	if ~batch,
	351	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && tar -zxf ' dirname '.tar.gz && source ' modelname '.queue '];
	352	else
	353	launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
	354	' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz '];
	355	end
	356	end
	357	issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
	358	else
	359	system([modelname '.bat']);
	360	end
	361
	362	end %}}}
[13395]	363	function Download(cluster,dirname,filelist)% {{{
[11527]	364
[13395]	365	if ispc(),
[12706]	366	%do nothing
	367	return;
[8578]	368	end
	369
[12706]	370	%copy files from cluster to current directory
	371	directory=[cluster.executionpath '/' dirname '/'];
	372	issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist);
[11527]	373	end %}}}
[8578]	374	end
	375	end

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

Download in other formats: