Changeset 10659

Timestamp:

Dec 12, 2006, 10:26:22 AM (20 years ago)

Author:

eugene

Message:

HUNG client should not be killed

Location:

trunk/Ohana/src/opihi/pcontrol

Files:

• CheckBusyJob.c (modified) (2 diffs)
• CheckDoneHost.c (modified) (1 diff)
• KillJob.c (modified) (2 diffs)
• ResetJob.c (modified) (1 diff)
• StartJob.c (modified) (1 diff)

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

@@ -36 +36 @@
 
     case PCLIENT_HUNG:
+      // don't do anything drastic, just keep trying
+      if (VerboseMode()) gprint (GP_ERR, "client is busy, not responding");
+      FreeIOBuffer (&buffer);
+      return (TRUE);
+
+# if (0)      
       HarvestHost (host[0].pid);
       // unlink host & job
@@ -44 +50 @@
       FreeIOBuffer (&buffer);
       return (TRUE);
+# endif
 
     case PCLIENT_GOOD:

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

@@ -26 +26 @@
 
     case PCLIENT_HUNG:
-      // XXX should this be DONE or DOWN?/
-      PutHost (host, PCONTROL_HOST_DONE, STACK_BOTTOM);
+      // don't do anything drastic, just try again later
+      PutHost (host, PCONTROL_HOST_BUSY, STACK_BOTTOM);
       if (DEBUG || VerboseMode()) gprint (GP_ERR, "host %s is not responding\n", host[0].hostname);
       FreeIOBuffer (&buffer);

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

@@ -30 +30 @@
 
     case PCLIENT_HUNG:
+      // don't do anything drastic, just keep trying
+      if (VerboseMode()) gprint (GP_ERR, "client is busy, not responding");
+      FreeIOBuffer (&buffer);
+      return (TRUE);
+
+# if (0)
       HarvestHost (host[0].pid);
       // unlink host & job
@@ -38 +44 @@
       FreeIOBuffer (&buffer);
       return (TRUE);
+# endif
 
     case PCLIENT_GOOD:

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

@@ -32 +32 @@
     case PCLIENT_HUNG:
       /*** should we consider a HUNG host DOWN? ***/
-      gprint (GP_ERR, "host %s is not responding\n", host[0].hostname);
+      gprint (GP_ERR, "host %s is not responding (ResetJob)\n", host[0].hostname);
       FreeIOBuffer (&buffer);
       return (FALSE);

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

@@ -38 +38 @@
 
     case PCLIENT_HUNG:
+      // we need the job start to return a valid Job ID, 
+      // give up on jobs which don't get started.
+      gprint (GP_ERR, "host %s is not responding (StartJob)\n", host[0].hostname);
       if (VerboseMode()) gprint (GP_ERR, "host %s is not responding\n", host[0].hostname);
       goto failure;