Changeset 20120

Timestamp:

02/11/16 14:55:01 (9 years ago)

Author:

Eric.Larour

Message:

CHG: added batch flag for the LaunchQueueJob and reordered the sequence in solve because
in generic clusters, in batch mode, we were not uploading anything!

Location:

issm/trunk-jpl/src/m

Files:

• classes/clusters/generic.m (modified) (3 diffs)
• classes/clusters/generic.py (modified) (2 diffs)
• classes/clusters/pfe.m (modified) (1 diff)
• classes/clusters/pfe.py (modified) (1 diff)
• solve/solve.m (modified) (3 diffs)
• solve/solve.py (modified) (2 diffs)

issm/trunk-jpl/src/m/classes/clusters/generic.m

@@ -155 +155 @@
                 end
                 %}}}
+                function BuildQueueScriptMultipleModels(cluster,dirname,modelname,solution,dirnames,modelnames,nps) % {{{
+                
+                        %some checks: 
+                        if isempty(modelname), error('BuildQueueScriptMultipleModels error message: need a non empty model name!');end
+
+                        %what is the executable being called? 
+                        executable='issm_slr.exe';
+
+                        if ispc(), error('BuildQueueScriptMultipleModels not support yet on windows machines');end;
+                        
+                        %write queuing script 
+                        fid=fopen([modelname '.queue'],'w');
+                        
+                        fprintf(fid,'#!%s\n',cluster.shell);
+
+                        %number of cpus: 
+                        mpistring=sprintf('mpiexec -np %i ',cluster.np);
+
+                        %executable: 
+                        mpistring=[mpistring sprintf('%s/%s ',cluster.codepath,executable)];
+                        
+                        %solution name: 
+                        mpistring=[mpistring sprintf('%s ',EnumToString(solution))];
+
+                        %execution directory and model name: 
+                        mpistring=[mpistring sprintf('%s/%s %s',cluster.executionpath,dirname,modelname)];
+
+                        %inform main executable of how many icecaps, glaciers and earth models are being run: 
+                        mpistring=[mpistring sprintf(' %i ',length(dirnames))];
+                        
+                        %icecaps, glaciers and earth location, names and number of processors associated:
+                        for i=1:length(dirnames),
+                        mpistring=[mpistring sprintf(' %s/%s %s %i ',cluster.executionpath,dirnames{i},modelnames{i},nps{i})];
+                        end
+
+                        %log files: 
+                        if ~cluster.interactive,
+                                mpistring=[mpistring sprintf('2> %s.errlog> %s.outlog',modelname,modelname)];
+                        end
+
+                        %write this long string to disk: 
+                        fprintf(fid,mpistring);
+                        fclose(fid);
+
+                        %in interactive mode, create a run file, and errlog and outlog file
+                        if cluster.interactive,
+                                fid=fopen([modelname '.errlog'],'w'); fclose(fid);
+                                fid=fopen([modelname '.outlog'],'w'); fclose(fid);
+                        end
+                end
+                %}}}
                 function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{
 
@@ -219 +270 @@
                         end
                 end %}}}
-                function LaunchQueueJob(cluster,modelname,dirname,filelist,restart)% {{{
+                function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
 
                         if ~ispc,
@@ -235 +286 @@
                                         launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && cd ' dirname ' && source ' modelname '.queue '];
                                 else
+                                        if ~batch,
                                         launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
                                                 ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz  && source  ' modelname '.queue '];
+                                        else
+                                        launchcommand=['source ' cluster.etcpath '/environment.' shellext ' && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
+                                                ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz '];
+                                        end
                                 end
                                 issmssh(cluster.name,cluster.login,cluster.port,launchcommand);

issm/trunk-jpl/src/m/classes/clusters/generic.py

@@ -183 +183 @@
 
         # }}}
-        def LaunchQueueJob(self,modelname,dirname,filelist,restart):    # {{{
+        def LaunchQueueJob(self,modelname,dirname,filelist,restart,batch):    # {{{
 
                 print 'launching solution sequence on remote cluster'
@@ -189 +189 @@
                         launchcommand='cd %s && cd %s chmod 777 %s.queue && ./%s.queue' % (self.executionpath,dirname,modelname,modelname)
                 else:
-                        launchcommand='cd %s && rm -rf ./%s && mkdir %s && cd %s && mv ../%s.tar.gz ./ && tar -zxf %s.tar.gz  && chmod 777 %s.queue && ./%s.queue' % \
+                        if batch:
+                                launchcommand='cd %s && rm -rf ./%s && mkdir %s && cd %s && mv ../%s.tar.gz ./ && tar -zxf %s.tar.gz  && chmod 777 %s.queue && ./%s.queue' % \
                                 (self.executionpath,dirname,dirname,dirname,dirname,dirname,modelname,modelname)
+                        else:
+                                launchcommand='cd %s && rm -rf ./%s && mkdir %s && cd %s && mv ../%s.tar.gz ./ && tar -zxf %s.tar.gz' % \
+                                (self.executionpath,dirname,dirname,dirname,dirname,dirname)
+
                 issmssh(self.name,self.login,self.port,launchcommand)
         # }}}

issm/trunk-jpl/src/m/classes/clusters/pfe.m

@@ -308 +308 @@
                  end
                  %}}}
-                 function LaunchQueueJob(cluster,modelname,dirname,filelist,restart)% {{{
+                 function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
 
                          %lauch command, to be executed via ssh

issm/trunk-jpl/src/m/classes/clusters/pfe.py

@@ -180 +180 @@
 
                 # }}}
-        def LaunchQueueJob(self,modelname,dirname,filelist,restart):
+        def LaunchQueueJob(self,modelname,dirname,filelist,restart,batch):
                         # {{{
 

issm/trunk-jpl/src/m/solve/solve.m

@@ -36 +36 @@
 md.private.solution=solutionenum;
 cluster=md.cluster;
+if strcmpi(getfieldvalue(options,'batch','no'),'yes') batch=1; else batch=0; end
 
 %check model consistency
@@ -99 +100 @@
 BuildQueueScript(cluster,md.private.runtimename,md.miscellaneous.name,md.private.solution,md.settings.io_gather,md.debug.valgrind,md.debug.gprof,md.qmu.isdakota); % queue file
 
-%Stop here if batch mode
-if strcmpi(getfieldvalue(options,'batch','no'),'yes')
-        if md.verbose.solution,
-                disp('batch mode requested: not launching job interactively');
-                disp('launch solution sequence on remote cluster by hand');
-        end
-        return;
-end
-
 %Upload all required files
 modelname = md.miscellaneous.name;
@@ -126 +118 @@
 
 %launch queue job: 
-LaunchQueueJob(cluster,md.miscellaneous.name,md.private.runtimename,filelist,restart);
+LaunchQueueJob(cluster,md.miscellaneous.name,md.private.runtimename,filelist,restart,batch);
 
+%return if batch: 
+if batch,
+        if md.verbose.solution,
+                disp('batch mode requested: not launching job interactively');
+                disp('launch solution sequence on remote cluster by hand');
+        end
+        return;
+end
 %wait on lock
 if isnan(md.settings.waitonlock),

issm/trunk-jpl/src/m/solve/solve.py

@@ -47 +47 @@
         #recover some fields
         md.private.solution=solutionenum
-        cluster=md.cluster
+        cluster=md.cluster 
+        if m.strcmpi(options.getfieldvalue('batch','no'),'yes'):
+                batch=1
+        else:
+                batch=0;
 
         #check model consistency
@@ -113 +117 @@
         
         #Launch job
-        cluster.LaunchQueueJob(md.miscellaneous.name,md.private.runtimename,filelist,restart)
+        cluster.LaunchQueueJob(md.miscellaneous.name,md.private.runtimename,filelist,restart,batch)
 
         #wait on lock