Changeset 26411 for trunk/Ohana/src/opihi/pcontrol/CheckIdleHost.c

Timestamp:

Dec 15, 2009, 3:42:06 PM (17 years ago)

Author:

eugene

Message:

add tracking of uniquely-named machines and feature to limit the number of unwanted-host jobs (and additional pantasks / pcontrol commands

File:

• trunk/Ohana/src/opihi/pcontrol/CheckIdleHost.c (modified) (7 diffs)

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

@@ -1 +1 @@
 # include "pcontrol.h"
 
-// The connection to the remote host is only allow 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.
+// 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.
 
-# define MAX_CONNECT_TIME 36000.0
-
+static float MAX_WANTHOST_WAIT = 10.0;
+static float MAX_CONNECT_TIME = 36000.0;
 static FILE *logfile = NULL;
 
@@ -70 +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 */
@@ -90 +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 */
@@ -103 +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;
@@ -108 +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 */
@@ -124 +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 */
@@ -145 +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 **/