Changeset 14356

Timestamp:

02/19/13 08:07:18 (12 years ago)

Author:

Mathieu Morlighem

Message:

CHG: we need to wait for the outlog since the lock file might be written before the outlog and errlog

File:

• issm/trunk-jpl/src/m/solve/waitonlock.m (modified) (5 diffs)

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

@@ -2 +2 @@
 %WAITONLOCK - wait for a file
 %
-%   This routine will return when a file named 'filename' is written to disk.
+%   This routine will return when a file named 'lockfilename' is written to disk.
+%   Also check for outlog file be cause it might bewritten several seconds
+%   after the lock file.
 %   If the time limit given in input is exceeded, return 0
 %
@@ -10 +12 @@
 %Return if waitonlock < 0 (no need to wait)
 
-%Get filename (lock file) and options
+%Get lockfilename (lock file) and options
 executionpath = md.cluster.executionpath;
 timelimit     = md.settings.waitonlock;
-filename      = [executionpath '/' md.private.runtimename '/' md.miscellaneous.name '.lock'];
 cluster       = md.cluster;
+lockfilename  = [executionpath '/' md.private.runtimename '/' md.miscellaneous.name '.lock'];
+logfilename   = [executionpath '/' md.private.runtimename '/' md.miscellaneous.name '.outlog'];
 
 %If we are using the generic cluster in interactive mode, job is already complete
@@ -24 +27 @@
 
 %initialize time and file presence test flag
-time=0; ispresent=0;
-disp(['waiting for ' filename ' hold on... (Ctrl+C to exit)'])
+time=0; ispresent=0; time0=clock;
+disp(['waiting for ' lockfilename ' hold on... (Ctrl+C to exit)'])
 
 %prepare command if the job is not running on the local machine
@@ -35 +38 @@
         end
         if port,
-                command = ['ssh -l ' login ' -p ' num2str(port) ' localhost " [ -f ' filename '  ]" 2>/dev/null'];
+                command = ['ssh -l ' login ' -p ' num2str(port) ' localhost "[ -f ' lockfilename ' ] && [ -f ' logfilename ' ]" 2>/dev/null'];
         else
-                command = ['ssh -l ' login ' ' cluster.name ' " [ -f ' filename '  ]" 2>/dev/null'];
+                command = ['ssh -l ' login ' ' cluster.name ' "[ -f ' lockfilename ' ] && [ -f ' logfilename ' ]" 2>/dev/null'];
         end
 end
@@ -44 +47 @@
 while (ispresent==0 & time<timelimit)
         if strcmpi(oshostname(),cluster),
-                ispresent=exist(filename,'file');
                 pause(1);
-                time=time+1/60;
+                ispresent=(exist(lockfilename,'file') & exist(logfilename,'file'));
+                time=etime(clock,time0)/60;
         else
                 pause(10);
+                time=etime(clock,time0)/60;
                 fprintf('\rchecking for job completion (time = %3.2f min)    ',time);
-                ispresent=~system(command); if ispresent, fprintf('\n'); end
-                time=time+10/60;
+                ispresent=~system(command);
+                if ispresent, fprintf('\n'); end
         end
 end