Changeset 18424 for branches/eam_branch_20080421/Ohana/src/opihi/pcontrol/CheckSystem.c

Timestamp:

Jul 3, 2008, 10:26:51 PM (18 years ago)

Author:

eugene

Message:

updates from HEAD

File:

• branches/eam_branch_20080421/Ohana/src/opihi/pcontrol/CheckSystem.c (modified) (4 diffs)

branches/eam_branch_20080421/Ohana/src/opihi/pcontrol/CheckSystem.c

@@ -99 +99 @@
 
     if (RunLevel != PCONTROL_RUN_NONE) {
+      Nhostchecks += CheckRespHosts(0.020); /* check for incoming messages */
+      TestCheckPoint ();
       Nhostchecks += CheckDoneHosts(0.020); /* reset the host */
       TestCheckPoint ();
@@ -258 +260 @@
 }
 
+int CheckRespHosts (float MaxDelay) {
+
+  struct timeval start, stop;
+  int i, Nobject;
+  Stack *hoststack;
+  Stack *jobstack;
+  Host *host;
+  Job *job;
+  float dtime;
+
+  /* Loop through objects on the stack, no more than once. see note above */
+  hoststack = GetHostStack (PCONTROL_HOST_RESP);
+  jobstack = GetJobStack (PCONTROL_JOB_RESP);
+  Nobject = hoststack[0].Nobject;
+
+  /* always allow at least one test */
+  gettimeofday (&start, (void *) NULL);
+  dtime = 0.0;
+  for (i = 0; (i < Nobject) && (dtime < MaxDelay); i++) {
+    LockStack (jobstack);
+    host = PullStackByLocation (hoststack, STACK_TOP);
+    if (host == NULL) {
+        UnlockStack (jobstack);
+        break;
+    }
+
+    // if the host has a job, we need to pull the job from its stack
+    job = (Job *) host[0].job;
+    if (job != NULL) {
+        RemoveStackByID (jobstack, job[0].JobID);
+    }
+    UnlockStack (jobstack);
+
+    CheckRespHost (host);
+    gettimeofday (&stop, (void *) NULL);
+    dtime = DTIME (stop, start);
+  }
+  if (DEBUG) gprint (GP_ERR, "checked %d hosts\n", i);
+  return (i);
+}
+
 int CheckDoneHosts (float MaxDelay) {
 
@@ -303 +346 @@
     if (host == NULL) break;
     if (host[0].markoff) {
+      // DOWN -> OFF
       host[0].markoff = FALSE;
       OffHost (host);
@@ -311 +355 @@
       PutHost (host, PCONTROL_HOST_DOWN, STACK_BOTTOM);
     } else {
+      // DOWN -> IDLE (maybe)
+      // this is a race condition with "host retry", but the only 
+      // consequence is that both StartHost and reset set the times to 0.0
       StartHost (host);
     }