Changeset 17477

Timestamp:

Apr 23, 2008, 3:38:50 PM (18 years ago)

Author:

eugene

Message:

handle comm errors where pclient sends an orphaned prompt

Location:

trunk/Ohana/src/opihi/pcontrol

Files:

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

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

@@ -60 +60 @@
   /** host is up, need to parse message **/
   p = memstr (buffer[0].buffer, "STATUS", buffer[0].Nbuffer);
-  ASSERT (p != NULL, "missing STATUS in pclient message");
+  if (p == NULL) {
+      if (DEBUG || VerboseMode()) gprint (GP_ERR, "missing STATUS in response; try again\n");
+      PutHost (host, PCONTROL_HOST_BUSY, STACK_BOTTOM);
+      PutJob (job, PCONTROL_JOB_BUSY, STACK_BOTTOM);
+      return (TRUE);
+  }
 
   sscanf (p, "%*s %s", string);

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

@@ -43 +43 @@
   /** successful command, examine result **/
   p = memstr (buffer[0].buffer, "STATUS", buffer[0].Nbuffer);
-  ASSERT (p != NULL, "missing STATUS in pclient message (CheckDoneHost)");
+  if (p == NULL) {
+      if (DEBUG || VerboseMode()) gprint (GP_ERR, "missing STATUS in response; try again\n");
+      PutHost (host, PCONTROL_HOST_DONE, STACK_BOTTOM);
+      return (FALSE);
+  }
 
   sscanf (p, "%*s %d", &status);

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

@@ -50 +50 @@
   /** check on response to pclient command **/
   p = memstr (buffer[0].buffer, "STATUS", buffer[0].Nbuffer);
-  ASSERT (p != NULL, "missing STATUS in pclient message");
+  if (p == NULL) {
+      if (VerboseMode()) gprint (GP_ERR, "missing STATUS in response; try again\n");
+      PutHost (host, PCONTROL_HOST_BUSY, STACK_BOTTOM);
+      PutJob (job, PCONTROL_JOB_KILL, STACK_BOTTOM);
+      return (FALSE);
+  }
   if (VerboseMode()) gprint (GP_ERR, "client message: %s\n", buffer[0].buffer);
 

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

@@ -72 +72 @@
   /* check on result of pclient command */
   p = memstr (buffer[0].buffer, "STATUS", buffer[0].Nbuffer);
-  ASSERT (p != NULL, "missing STATUS in pclient message");
+  if (p == NULL) {
+      // failed to get a valid response.  kill the job and try again, 
+      // or accept a running process without a PID?
+      if (VerboseMode()) gprint (GP_ERR, "failed to get a valid PID, trying to continue without\n");
+      PutHost (host, PCONTROL_HOST_BUSY, STACK_BOTTOM);
+      PutJob (job, PCONTROL_JOB_BUSY, STACK_BOTTOM);
+      gettimeofday (&job[0].start, NULL);
+      return (TRUE);
+  }
 
   sscanf (p, "%*s %d", &status);