ok

test/test.mk

@@ -1,8 +1,7 @@
 # this is a makefile used to show testcases of rabit
-.PHONY:
+.PHONY:all
 
-test:
-	../tracker/rabit_demo.py -v 1 -n 10 test_model_recover 1 mock=0,0,1,0 mock=1,1,1,0 mock=1,1,1,1 mock=0,1,1,0 mock=4,1,1,0 mock=8,1,2,0 
+all:
 
 # this experiment test recovery with actually process exit, use keepalive to keep program alive
 model_recover_10_10k:
@@ -14,3 +13,6 @@ model_recover_10_10k_die_same:
 model_recover_10_10k_die_hard:
 	../tracker/rabit_demo.py -n 10 test_model_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=1,1,1,1 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=8,1,2,0 mock=4,1,3,0
 
+
+local_recover_10_10k:
+	../tracker/rabit_demo.py -n 10 test_local_recover 10000 mock=0,0,1,0 mock=1,1,1,0 mock=0,1,1,0 mock=4,1,1,0 mock=9,1,1,0 mock=1,1,1,1

tracker/rabit_demo.py

@@ -20,20 +20,40 @@ parser.add_argument('command', nargs='+',
                     help = 'command for rabit program')
 args = parser.parse_args()
 
+# bash script for keepalive
+# use it so that python do not need to communicate with subprocess
+echo="echo %s rabit_num_trial=$nrep;"
+keepalive = """
+nrep=0
+rc=254
+while [ $rc -eq 254 ]; 
+do
+    %s
+    %s %s rabit_num_trial=$nrep
+    rc=$?;
+    nrep=$((nrep+1));
+done
+"""
+
 def exec_cmd(cmd, taskid):
     if cmd[0].find('/') == -1 and os.path.exists(cmd[0]):
         cmd[0] = './' + cmd[0]
     cmd = ' '.join(cmd)
+    arg = ' rabit_task_id=%d' % (taskid) 
+    cmd = cmd + arg
     ntrial = 0
     while True:
         prep = 'PYTHONPATH=\"%s\" ' % WRAPPER_PATH
-        arg = ' rabit_task_id=%d rabit_num_trial=%d' % (taskid, ntrial)        
-        ret = subprocess.call(prep + cmd + arg, shell = True)
-        if ret == 254 or ret == -2:
-            ntrial += 1
-            continue
+        if args.verbose != 0:            
+            bash = keepalive % (echo % cmd, prep, cmd)
+        else:
+            bash = keepalive % ('', prep, cmd)            
+        ret = subprocess.call(bash, shell=True, executable='bash')
         if ret == 0:
+            if args.verbose != 0:        
+                print 'Thread %d exit with 0' % taskid
             return
+        else:
             raise Exception('Get nonzero return code=%d' % ret)
 #
 #  Note: this submit script is only used for demo purpose