Changeset 27840 for branches/simtest_nebulous_branches/Ohana/src/opihi/pcontrol/CheckIdleHost.c

Timestamp:

May 3, 2010, 8:50:52 AM (16 years ago)

Author:

eugene

Message:

updates from trunk

Location:

branches/simtest_nebulous_branches

Files:

• . (modified) (1 prop)
• Ohana (modified) (1 prop)
• Ohana/src/opihi/pcontrol/CheckIdleHost.c (modified) (8 diffs)

branches/simtest_nebulous_branches/Ohana/src/opihi/pcontrol/CheckIdleHost.c

@@ -1 +1 @@
 # include "pcontrol.h"
 
+// The connection to the remote host is only allowed to live for MAX_CONNECT_TIME seconds.
+// We disconnect and reconnect if a remote host has been connected for too long.  This is
+// a (temporary?) work-around for the problem that the remote pclient job tends to grow
+// too large over time.
+
+static float MAX_WANTHOST_WAIT = 10.0;
+static float MAX_CONNECT_TIME = 36000.0;
 static FILE *logfile = NULL;
+
+  /* if this host has been connected for too long, disconnect (will automatically reconnect) */
+int CheckResetHost (Host *host) {
+
+  struct timeval now;
+  float dtime;
+
+  /* if this host has been connected for too long, disconnect (will automatically reconnect) */
+  gettimeofday (&now, (void *) NULL);
+  dtime = DTIME (now, host[0].connect_time);
+  if (dtime > MAX_CONNECT_TIME) {
+      if (VerboseMode()) gprint (GP_ERR, "disconnect from %s\n", host[0].hostname);
+      StopHost (host, PCONTROL_HOST_DOWN);
+      return (TRUE);
+  }
+  return FALSE;
+}
 
 /* the supplied host is not on a stack: it cannot be taken by the other thread */
@@ -21 +45 @@
   if (host[0].markoff) {
     host[0].markoff = FALSE;
-    StopHost (host);
+    StopHost (host, PCONTROL_HOST_OFF);
     return (TRUE);
   }
     
+  /* check if host has been connected for too long */
+  if (CheckResetHost (host)) {
+    return (TRUE);
+  }
+
   /* search the JOB_PENDING stack for an appropriate job */
   stack = GetJobStack (PCONTROL_JOB_PENDING);
@@ -41 +70 @@
 
     // if (logfile) fprintf (logfile, "start needhost %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]);
+    AddMachineJob (host, job);
 
     /* take the job off the stack and unlock the stack */
@@ -61 +91 @@
 
     // if (logfile) fprintf (logfile, "start wanthost %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]);
+    AddMachineJob (host, job);
 
     /* take the job off the stack and unlock the stack */
@@ -74 +105 @@
     if (job[0].mode != PCONTROL_JOB_ANYHOST) continue;
 
+    if (!CheckMachineJobs (host, job)) continue;
+
     /* we have found an appropriate job; link it to the host and send to StartJob */
     job[0].host = (struct Host *) host;
@@ -79 +112 @@
 
     // if (logfile) fprintf (logfile, "start  anyhost %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]);
+    AddMachineJob (host, job);
 
     /* take the job off the stack and unlock the stack */
@@ -95 +129 @@
     gettimeofday (&now, (void *) NULL);
     dtime = DTIME (now, job[0].start);
-    if (dtime < 10.0) continue;
+    if (dtime < MAX_WANTHOST_WAIT) continue;
+
+    if (!CheckMachineJobs (host, job)) continue;
 
     if (logfile) fprintf (logfile, "start wanthost(2) %s (job host %s) : %s\n", host[0].hostname, job[0].hostname, job[0].argv[0]);
+    AddMachineJob (host, job);
 
     /* we have found an appropriate job; link it to the host and send to StartJob */
@@ -116 +153 @@
 }
 
+void SetMaxWantHostWait (float value) {
+
+  MAX_WANTHOST_WAIT = value;
+  return;
+}
+
+float GetMaxWantHostWait (void) {
+
+  return MAX_WANTHOST_WAIT;
+}
+
+void SetMaxConnectTime (float value) {
+
+  MAX_CONNECT_TIME = value;
+  return;
+}
+
+float GetMaxConnectTime (void) {
+
+  return MAX_CONNECT_TIME;
+}
+
 /** note : host and job popped off IDLE and PENDING stacks, 
     unless no job is available **/