Changeset 8546 for trunk/Ohana/src/opihi/pcontrol/StopHosts.c

Timestamp:

Aug 23, 2006, 5:31:09 PM (20 years ago)

Author:

eugene

Message:

cleanup of timeouts and thread-safety issues

File:

• trunk/Ohana/src/opihi/pcontrol/StopHosts.c (modified) (4 diffs)

trunk/Ohana/src/opihi/pcontrol/StopHosts.c

@@ -17 +17 @@
 }
 
+/* for use by shutdown: force machines which are up to go down
+   wait for a little while for the client thread to take care 
+   of them
+*/
+   
 int DownHosts () {
 
+  int i, Nobject, Nwait;
   Stack *stack;
   Host  *host;
 
+  SetCheckPoint (); // ensure we can find the specified host
   stack = GetHostStack (PCONTROL_HOST_IDLE);
-  while ((host = PullStackByLocation (stack, STACK_BOTTOM)) != NULL) {
-    StopHost (host);
-    DownHost (host);
+  Nobject = stack[0].Nobject;
+  for (i = 0; i < Nobject; i++) {
+    host = PullStackByLocation (stack, STACK_TOP);
+    if (host == NULL) continue;
+    host[0].markoff = TRUE;
+    PutHost (host, PCONTROL_HOST_IDLE, STACK_BOTTOM);
   }
 
   stack = GetHostStack (PCONTROL_HOST_BUSY);
-  while ((host = PullStackByLocation (stack, STACK_BOTTOM)) != NULL) {
-    StopHost (host);
-    DownHost (host);
+  Nobject = stack[0].Nobject;
+  for (i = 0; i < Nobject; i++) {
+    host = PullStackByLocation (stack, STACK_TOP);
+    if (host == NULL) continue;
+    host[0].markoff = TRUE;
+    PutHost (host, PCONTROL_HOST_IDLE, STACK_BOTTOM);
+  }
+  ClearCheckPoint ();
+
+  Nwait = 0;
+  stack = GetHostStack (PCONTROL_HOST_IDLE);
+  gprint (GP_ERR, "waiting for clients to exit");
+  while ((Nwait < 15) && stack[0].Nobject) {
+    gprint (GP_ERR, ".");
+    usleep (100000); // wait for clients to exit
+    Nwait++;
+  }
+  gprint (GP_ERR, "\n");
+  if (stack[0].Nobject) {
+    gprint (GP_ERR, "trouble shutting down all pclient instances: %d still alive\n", stack[0].Nobject);
+  } else {
+    gprint (GP_ERR, "done\n");
   }
   return (TRUE);
@@ -69 +98 @@
   int i, result, waitstatus;
 
-  gprint (GP_ERR, "harvesting within thread %d\n", pthread_self());
-  gprint (GP_ERR, "child process %d is down, wait for exit status\n", pid);
+  // gprint (GP_ERR, "harvesting within thread %d\n", pthread_self());
+  // gprint (GP_ERR, "child process %d is down, wait for exit status\n", pid);
   
   // Loop a few times waiting for child to exit
@@ -76 +105 @@
     result = waitpid (pid, &waitstatus, WNOHANG);
     if ((result == -1) && (errno == ECHILD)) {
-      usleep (10000);
+      usleep (10000); // wait for child to exit
       continue;
     } else {
@@ -109 +138 @@
       
       if (WIFEXITED(waitstatus)) {
-        gprint (GP_ERR, "child exited with status %d\n", WEXITSTATUS(waitstatus));
+        if (VerboseMode()) gprint (GP_ERR, "child exited with status %d\n", WEXITSTATUS(waitstatus));
       }
       if (WIFSIGNALED(waitstatus)) {
-        gprint (GP_ERR, "child crashed with status %d\n", WTERMSIG(waitstatus));
+        if (VerboseMode()) gprint (GP_ERR, "child crashed with status %d\n", WTERMSIG(waitstatus));
       }
       if (WIFSTOPPED(waitstatus)) {
-        gprint (GP_ERR, "waitpid returns 'stopped': programming error\n");
+        if (VerboseMode()) gprint (GP_ERR, "waitpid returns 'stopped': programming error\n");
         exit (1);
       }