Changeset 1468 for palm/trunk/SOURCE/check_for_restart.f90

Timestamp:

Sep 24, 2014 2:06:57 PM (10 years ago)

Author:

maronga

Message:

New flag files allow to force unscheduled termination/restarts of batch jobs, progress output is made for batch runs, small adjustments for lxce6 and lccrayh/lccrayb

File:

• palm/trunk/SOURCE/check_for_restart.f90 (modified) (5 diffs)

palm/trunk/SOURCE/check_for_restart.f90

@@ -20 +20 @@
 ! Current revisions:
 ! -----------------
-! 
+! Added support for unscheduled job termination using the flag files
+! DO_STOP_NOW and DO_RESTART_NOW
 ! 
 ! Former revisions:
@@ -65 +66 @@
 
 
-    LOGICAL :: terminate_run_l  !:
+    LOGICAL :: terminate_run_l           !:
+    LOGICAL :: do_stop_now = .FALSE.     !:
+    LOGICAL :: do_restart_now = .FALSE.  !:
 
     REAL(wp) ::  remaining_time !:
@@ -113 +116 @@
 
        terminate_coupled = 3
+
 #if defined( __parallel )
        IF ( myid == 0 ) THEN
@@ -126 +130 @@
     ENDIF
 
-!
-!-- Set the stop flag also, if restart is forced by user
-    IF ( time_restart /= 9999999.9_wp  .AND.                                      &
+
+!
+!-- Check if a flag file exists that forces a termination of the model
+    terminate_run_l = .FALSE.
+    IF ( myid == 0 )  THEN
+       INQUIRE(FILE="DO_STOP_NOW", EXIST=do_stop_now)
+       INQUIRE(FILE="DO_RESTART_NOW", EXIST=do_restart_now)
+
+       IF ( do_stop_now .OR. do_restart_now )  THEN
+
+          terminate_run_l = .TRUE.
+
+          WRITE( message_string, * ) 'run will be terminated because user ',   &
+                                  'forced a job finialization using a flag',   &
+                                  'file:',                                     &
+                                  '&DO_STOP_NOW: ', do_stop_now,               &
+                                  '&DO_RESTART_NOW: ', do_restart_now 
+          CALL message( 'check_for_restart', 'PA0398', 0, 0, 0, 6, 0 )
+
+       ENDIF
+    ENDIF
+
+
+#if defined( __parallel )
+!
+!-- Make a logical OR for all processes. Stop the model run if at least
+!-- one processor has reached the time limit.
+    IF ( collective_wait )  CALL MPI_BARRIER( comm2d, ierr )
+    CALL MPI_ALLREDUCE( terminate_run_l, terminate_run, 1, MPI_LOGICAL,        &
+                        MPI_LOR, comm2d, ierr )
+#else
+    terminate_run = terminate_run_l
+#endif
+
+!
+!-- In case of coupled runs inform the remote model of the termination 
+!-- and its reason, provided the remote model has not already been 
+!-- informed of another termination reason (terminate_coupled > 0) before,
+!-- or vice versa (terminate_coupled_remote > 0).
+    IF ( terminate_run .AND. coupling_mode /= 'uncoupled' .AND.                &
+         terminate_coupled == 0 .AND.  terminate_coupled_remote == 0 )  THEN
+
+       terminate_coupled = 6
+
+#if defined( __parallel )
+       IF ( myid == 0 ) THEN
+          CALL MPI_SENDRECV( terminate_coupled,        1, MPI_INTEGER,      &
+                             target_id,  0,                                 &
+                             terminate_coupled_remote, 1, MPI_INTEGER,      &
+                             target_id,  0,                                 &
+                             comm_inter, status, ierr )   
+       ENDIF
+       CALL MPI_BCAST( terminate_coupled_remote, 1, MPI_INTEGER, 0,         &
+                       comm2d, ierr )  
+#endif
+
+    ENDIF
+
+!
+!-- Set the stop flag also, if restart is forced by user settings
+    IF ( time_restart /= 9999999.9_wp  .AND.                                   &
          time_restart < time_since_reference_point )  THEN
 
@@ -183 +245 @@
 !
 !-- If the run is stopped, set a flag file which is necessary to initiate
-!-- the start of a continuation run
-    IF ( terminate_run  .AND.  myid == 0 )  THEN
+!-- the start of a continuation run, except if the user forced to stop the
+!-- run without restart
+    IF ( terminate_run  .AND.  myid == 0 .AND. .NOT. do_stop_now)  THEN
 
        OPEN ( 90, FILE='CONTINUE_RUN', FORM='FORMATTED' )