timeouts

plugins/Full_CI_ZMQ/fci_zmq.irp.f

@@ -70,7 +70,7 @@ program fci_zmq
       if (do_pt2) then
         pt2 = 0.d0
         threshold_selectors = 1.d0
-        threshold_generators = 1d0 
+        threshold_generators = 1.d0 
         SOFT_TOUCH threshold_selectors threshold_generators
         call ZMQ_pt2(CI_energy, pt2,relative_error,absolute_error,error) ! Stochastic PT2
         threshold_selectors = threshold_selectors_save

plugins/Full_CI_ZMQ/run_pt2_slave.irp.f

@@ -87,9 +87,11 @@ subroutine run_pt2_slave(thread,iproc,energy)
   end do
 
   integer, external :: disconnect_from_taskserver
-  if (disconnect_from_taskserver(zmq_to_qp_run_socket,worker_id) == -1) then 
-    continue 
-  endif 
+  do i=1,300
+    if (disconnect_from_taskserver(zmq_to_qp_run_socket,worker_id) /= -2) exit
+    call sleep(1)
+    print *,  'Retry disconnect...'
+  end do
 
   call end_zmq_push_socket(zmq_socket_push,thread)
   call end_zmq_to_qp_run_socket(zmq_to_qp_run_socket)

plugins/Generators_full/generators.irp.f

@@ -13,7 +13,7 @@ BEGIN_PROVIDER [ integer, N_det_generators ]
  N_det_generators = N_det
  do i=1,N_det
    norm = norm + psi_average_norm_contrib_sorted(i)
-   if (norm >= threshold_generators) then
+   if (norm > threshold_generators) then
      N_det_generators = i
      exit
    endif

src/ZMQ/utils.irp.f

@@ -601,7 +601,7 @@ subroutine end_parallel_job(zmq_to_qp_run_socket,zmq_socket_pull,name_in)
     stop 'Wrong end of job'
   endif
 
-  do i=300,1,-1
+  do i=3600,1,-1
     rc = f77_zmq_send(zmq_to_qp_run_socket, 'end_job '//trim(zmq_state),8+len(trim(zmq_state)),0)
     rc = f77_zmq_recv(zmq_to_qp_run_socket, message, 512, 0)
     if (trim(message(1:13)) == 'error waiting') then
@@ -612,6 +612,7 @@ subroutine end_parallel_job(zmq_to_qp_run_socket,zmq_socket_pull,name_in)
     endif
   end do
   if (i==0) then
+    print *,  '.. Forcing kill ..'
     rc = f77_zmq_send(zmq_to_qp_run_socket, 'end_job force',13,0)
     rc = f77_zmq_recv(zmq_to_qp_run_socket, message, 512, 0)
   endif
@@ -718,7 +719,7 @@ integer function disconnect_from_taskserver_state(zmq_to_qp_run_socket, worker_i
   rc = f77_zmq_send(zmq_to_qp_run_socket, trim(message), sze, 0)
 
   if (rc /= sze) then
-    disconnect_from_taskserver_state = -1
+    disconnect_from_taskserver_state = -2
     return
   endif
   
@@ -727,7 +728,6 @@ integer function disconnect_from_taskserver_state(zmq_to_qp_run_socket, worker_i
   
   read(message,*, end=10, err=10) reply, state
   if ((trim(reply) == 'disconnect_reply').and.(trim(state) == trim(zmq_state))) then
-    disconnect_from_taskserver_state = -1
     return
   endif
   if (trim(message) == 'error Wrong state') then