Fork deadlock fix

src/slurmd/Makefile.am

@@ -13,7 +13,9 @@ process_lib = libthreaded.la
 io_lib = libthreaded_io.la
 endif
 
+#sbin_PROGRAMS = slurmd slurmd_nbio slurmd_elan
 sbin_PROGRAMS = slurmd slurmd_nbio
+
 noinst_LTLIBRARIES = libelan_interconnect.la \
 		libno_interconnect.la \
 		libforked_io.la  \
@@ -25,13 +27,12 @@ noinst_LTLIBRARIES = libelan_interconnect.la \
 LDADD = $(top_srcdir)/src/common/libcommon.la \
 	$(top_srcdir)/src/common/libdaemonize.la \
 	$(top_srcdir)/src/common/libcred.la \
-	$(interconnect_lib) \
 	$(SSL_LIBS)
 
-slurmd_LDADD = $(process_lib) $(io_lib) $(LDADD)
-slurmd_nbio_LDADD = $(process_lib) libnb_io.la $(LDADD)
+slurmd_LDADD = $(process_lib) $(io_lib) $(LDADD) $(interconnect_lib)
+slurmd_nbio_LDADD = $(process_lib) libnb_io.la $(LDADD) $(interconnect_lib)
 	
-slurmd_SOURCES = slurmd.c \
+common_sources = slurmd.c \
 		get_mach_stat.c \
 		read_proc.c \
 		task_mgr.c \
@@ -40,14 +41,12 @@ slurmd_SOURCES = slurmd.c \
 		pipes.c \
 		locks.c
 
-slurmd_nbio_SOURCES = slurmd.c \
-		get_mach_stat.c \
-		read_proc.c \
-		task_mgr.c \
-		shmem_struct.c \
-		circular_buffer.c \
-		pipes.c \
-		locks.c
+slurmd_SOURCES = $(common_sources)
+slurmd_nbio_SOURCES = $(common_sources)
+
+#slurmd_elan_LDADD = libforked.la libforked_io.la  $(LDADD) libelan_interconnect.la
+#slurmd_elan_SOURCES = $(common_sources)
+
 
 libforked_la_SOURCES = forked_ctrl.c
 libthreaded_la_SOURCES = threaded_ctrl.c

src/slurmd/task_mgr.c

@@ -83,7 +83,11 @@ int fan_out_task_launch ( launch_tasks_request_msg_t * launch_msg )
 		task_start[i] -> io_streams_dest = launch_msg -> streams ; 
 
 		if ( launch_task ( task_start[i] ) )
+		{
+			info ( "launch_task error " ) ;
 			goto kill_tasks_label ;
+		}
+		info ( "task %i launched" , i ) ;
 	}
 	
 	/* wait for all the launched threads to finish */
@@ -194,16 +198,11 @@ void * task_exec_thread ( void * arg )
 			/* ask me how I know :) */
 			/* think once, twice, 10^100000000 times before changeing the below */
 			/*1*/setup_parent_pipes ( task_start->pipes ) ;
-			info ( "before waitpid ") ;
 			/*2*/waitpid ( cpid , & task_return_code , 0 ) ;
-			info ( "taskid: %i before wait_on_io_threads " , cpid ) ;
 			/*3*/wait_on_io_threads ( task_start ) ;
-			info ( "before cleanup_parent_pipes " ) ;
 			/*4*/cleanup_parent_pipes (  task_start->pipes ) ;
-			info ( "before end_task_exit_msg" ) ;
 			
 			send_task_exit_msg ( task_return_code , task_start ) ;
-			info ( "task_exec_thread done" ) ;
 			break;
 	}
 	return ( void * ) SLURM_SUCCESS ;