Changeset 2555

Timestamp:

10/28/09 08:41:52 (15 years ago)

Author:

Mathieu Morlighem

Message:

Now if waitonlock is activated and the job is not running on the same machine, ask the user to input 'y' one the job is completed and the results are automaticaly loaded

Location:

issm/trunk/src/m/classes/public

Files:

• LaunchQueueJobGeneric.m (modified) (2 diffs)
• LaunchQueueJobcosmos.m (modified) (2 diffs)
• LaunchQueueJobgemini.m (modified) (2 diffs)
• LaunchQueueJobgreenplanet.m (modified) (2 diffs)
• solveparallel.m (modified) (1 diff)
• waitonlock.m (modified) (2 diffs)

issm/trunk/src/m/classes/public/LaunchQueueJobGeneric.m

@@ -1 +1 @@
 function md=LaunchQueueJobGeneric(md,executionpath,options)
-%LAUNCHQUEUEJOBGENERIC - ...
+%LAUNCHQUEUEJOBGENERIC - Generic routine to launch queueing job
 %
 %   Usage:
@@ -28 +28 @@
         issmssh(md.cluster,['cd ' executionpath ' && source ' md.name '.queue ']);
 
-        %waitonlock will work if the lock is on the same machine only: 
-        if ~strcmpi(oshostname(),md.cluster),
-                disp('solution launched on remote cluster. log in to detect job completion, then run md=loadresultsfromcluster(md)');
-                md.waitonlock=0;
-        end
-
 else
         disp('batch mode requested: not launching job interactively');
         disp(['launch solution sequence by copying files ' md.name '.bin' ',' md.name '.queue and ' md.name '.qmu.in, onto remote cluster ' md.cluster]);
         disp(['   then log int ' md.cluster ' and source queue file ' md.name '.queue']);
-        md.waitonlock=0;
 end

issm/trunk/src/m/classes/public/LaunchQueueJobcosmos.m

@@ -18 +18 @@
         disp('uploading input file,  queueing script and variables script');
         if md.qmu_analysis, 
-                eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in ' md.cluster ':' executionpath]);
+                system(['scp ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in ' md.cluster ':' executionpath]);
         else
-                eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.cluster ':' executionpath]);
+                system(['scp ' md.name '.bin' ' ' md.name '.queue '  md.cluster ':' executionpath]);
         end
 
@@ -31 +31 @@
         disp(['launch solution sequence on remote cluster by logging into it and typing qsub -S/bin/csh ' md.name '.queue']);
 end
-md.waitonlock=0; %cannot wait for lock

issm/trunk/src/m/classes/public/LaunchQueueJobgemini.m

@@ -18 +18 @@
         disp('uploading input file,  queueing script and variables script');
         if md.qmu_analysis, 
-                eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in ' md.cluster ':' executionpath]);
+                system(['scp ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in ' md.cluster ':' executionpath]);
         else
-                eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.cluster ':' executionpath]);
+                system(['scp ' md.name '.bin' ' ' md.name '.queue '  md.cluster ':' executionpath]);
         end
         
@@ -31 +31 @@
         disp(['launch solution sequence on remote cluster by logging into it and typing qsub < ' md.name '.queue']);
 end
-md.waitonlock=0; %cannot wait for lock

issm/trunk/src/m/classes/public/LaunchQueueJobgreenplanet.m

@@ -18 +18 @@
         disp('uploading input file,  queueing script and variables script');
         if md.qmu_analysis, 
-                eval(['!scp -P 1025 -o HostKeyAlias=huey.jpl.nasa.gov ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in elarour@localhost:' executionpath]);
+                system(['scp -P 1025 -o HostKeyAlias=huey.jpl.nasa.gov ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in elarour@localhost:' executionpath]);
         else
-                eval(['!scp -P 1025 -o HostKeyAlias=huey.jpl.nasa.gov ' md.name '.bin' ' ' md.name '.queue elarour@localhost:' executionpath ]);
+                system(['scp -P 1025 -o HostKeyAlias=huey.jpl.nasa.gov ' md.name '.bin' ' ' md.name '.queue elarour@localhost:' executionpath ]);
         end
 else
@@ -28 +28 @@
 %new greenplanet cannot launch across cluster using ssh
 disp(['launch solution sequence on remote cluster by logging into it and typing qsub < ' md.name '.queue']);
-md.waitonlock=0;

issm/trunk/src/m/classes/public/solveparallel.m

@@ -25 +25 @@
 md=LaunchQueueJob(md,executionpath,options);
 
-%timing 
-t1=clock;
-
 %Do we return, or just wait for results?
 if md.waitonlock~=0,
         %we wait for the done file
-        islock=waitonlock([executionpath '/' md.name '.lock'],md.waitonlock);
+        islock=waitonlock(md,executionpath);
         if islock==0,
                 %no results to be loaded
-                disp('The results must be loaded manually with loadmodelfromcluster.');
-                error(['solveparallel info message: time limit exceeded'])
+                disp('The results must be loaded manually with md=loadmodelfromcluster(md).');
         else
                 %load results
-                %stop timing
-                t2=clock;
-                displaystring(md.verbose,'\n%s\n',['solution converged in ' num2str(etime(t2,t1)) ' seconds']);
-
                 displaystring(md.verbose,'loading results from cluster');
                 md=loadresultsfromcluster(md);

issm/trunk/src/m/classes/public/waitonlock.m

@@ -1 @@
-function flag=waitonlock(filename,timelimit)
+function flag=waitonlock(md,executionpath)
 %WAITONLOCK - wait for a file
 %
@@ -6 +6 @@
 %
 %   Usage:
-%      flag=waitonlock(filename,timelimit)
+%      flag=waitonlock(md,executionpath)
 
-%initialize time and file presence test flag
-time=0;
-ispresent=0;
-disp(['waiting for ' filename ' hold on... (Ctrl+C to exit)'])
+%Get filename (lock file) and options
+filename=[executionpath '/' md.name '.lock'];
+cluster=md.cluster;
+timelinit=md.waitonlock;
 
-%loop till file .lock exist or time is up
-while (ispresent==0 & time<timelimit)
-        ispresent=exist(filename,'file');
-        pause(1);
-        time=time+1/60;
+%waitonlock will work if the lock is on the same machine only: 
+if ~strcmpi(oshostname(),cluster),
+        disp('solution launched on remote cluster. log in to detect job completion.');
+        choice=input('Is the job successfully completed? (y/n)','s');
+        if ~strcmp(choice,'y'), 
+                disp('Results not loaded... exiting'); 
+                flag=0;
+        else
+                flag=1;
+        end
+
+%job is running on the same machine
+else
+
+        %initialize time and file presence test flag
+        time=0; ispresent=0;
+        disp(['waiting for ' filename ' hold on... (Ctrl+C to exit)'])
+
+        %loop till file .lock exist or time is up
+        while (ispresent==0 & time<timelimit)
+                ispresent=exist(filename,'file');
+                pause(1);
+                time=time+1/60;
+        end
+
+        %build output
+        if (time>timelimit),
+                disp(['waitonlock info message: time limit exceeded']);
+                flag=0;
+        else
+                flag=1;
+        end
 end
-
-%build output
-if (time>timelimit),
-        flag=0;
-else
-        flag=1;
-end