Changeset 28158 for trunk/Ohana/src/opihi/pcontrol/StopHosts.c

Timestamp:

May 28, 2010, 11:51:46 AM (16 years ago)

Author:

eugene

Message:

redirect pcontrol stderr to log file; add bits to harvest zombies that may accumulate

File:

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

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

@@ -1 +1 @@
 # include "pcontrol.h"
+
+// we attempt to harvest the 'down' hosts in HarvestHost.  However, sometimes the
+// child is busy and does not exit in the timeout period.  we need to keep a list and
+// try again occasionally to free up the needed resources
+static int NUNHARVESTED = 0;
+static int Nunharvested = 0;
+static int *unharvested = NULL;
 
 void DownHost (Host *host) {
@@ -140 +147 @@
       switch (errno) {
         case ECHILD:
-          gprint (GP_ERR, "unknown PID, not a child proc\n");
+          gprint (GP_ERR, "HarvestHost: unknown PID (%d), not a child proc\n", pid);
           gprint (GP_ERR, "did process already exit?  programming error?\n");
           break;
@@ -152 +159 @@
       
     case 0:
-      gprint (GP_ERR, "HarvestHost: child with connection to remote host failed to exit: may be hung");
+      gprint (GP_ERR, "HarvestHost: child with connection to remote host failed to exit: may be hung\n");
+      AddZombie(pid);
       break;
 
@@ -174 +182 @@
   return (TRUE);
 }
+
+int AddZombie(int pid) {
+
+  if (unharvested == NULL) {
+    NUNHARVESTED = 128;
+    ALLOCATE (unharvested, int, NUNHARVESTED);
+    memset (unharvested, 0, NUNHARVESTED*sizeof(int));
+  }
+  unharvested[Nunharvested] = pid;
+
+  Nunharvested ++;
+  if (Nunharvested >= NUNHARVESTED) {
+    NUNHARVESTED += 128;
+    REALLOCATE (unharvested, int, NUNHARVESTED);
+    memset (&unharvested[Nunharvested], 0, (NUNHARVESTED - Nunharvested)*sizeof(int));
+  }
+  return TRUE;
+}
+
+int DelZombies() {
+
+  int i, j;
+
+  if (!unharvested) return FALSE;
+  if (!Nunharvested) return FALSE;
+  if (!NUNHARVESTED) return FALSE;
+
+  int *newlist = NULL;
+
+  ALLOCATE (newlist, int, NUNHARVESTED);
+  memset (newlist, 0, NUNHARVESTED*sizeof(int));
+
+  j = 0;
+  for (i = 0; i < NUNHARVESTED; i++) {
+    if (!unharvested[i]) continue;
+    newlist[j] = unharvested[i];
+    j++;
+  }
+  free (unharvested);
+  unharvested = newlist;
+  Nunharvested = j;
+  return TRUE;
+}
+
+int CheckZombies() {
+
+  int pid, i, result, waitstatus;
+
+  if (!unharvested) return FALSE;
+  if (!Nunharvested) return FALSE;
+  if (!NUNHARVESTED) return FALSE;
+
+  for (i = 0; i < Nunharvested; i++) {
+    if (!unharvested[i]) continue;
+    pid = unharvested[i];
+    result = waitpid (pid, &waitstatus, WNOHANG);
+    switch (result) {
+      case -1:  /* error with waitpid */
+        switch (errno) {
+          case ECHILD:
+            gprint (GP_ERR, "CheckZombies: unknown PID (%d), not a child proc\n", pid);
+            gprint (GP_ERR, "did process already exit?  programming error?\n");
+            break;
+          case EINTR:
+          case EINVAL:
+          default:
+            perror ("unexpected error");
+            ABORT ("CheckZombies impossible condition");
+        }
+        break;
+      
+      case 0:
+        if (VerboseMode()) gprint (GP_ERR, "CheckZombies: still waiting on %d\n", pid);
+        break;
+
+      default:
+        if (result != pid) {
+          gprint (GP_ERR, "waitpid error: mis-matched PID (%d vs %d).  programming error\n", result, pid);
+          ABORT ("CheckZombies impossible condition");
+        }
+        
+        if (WIFEXITED(waitstatus)) {
+          if (VerboseMode()) gprint (GP_ERR, "child exited with status %d\n", WEXITSTATUS(waitstatus));
+        }
+        if (WIFSIGNALED(waitstatus)) {
+          if (VerboseMode()) gprint (GP_ERR, "child crashed with status %d\n", WTERMSIG(waitstatus));
+        }
+        if (WIFSTOPPED(waitstatus)) {
+          ABORT ("waitpid returns 'stopped': programming error\n");
+        }
+        unharvested[i] = 0;
+        break;
+    }
+  }
+  DelZombies();
+  return (TRUE);
+}