[patch 4/8] openmosix/add-kcom_node-error-count.patch

Subject: [patch @num@/@total@] @name@ 

This patch add node error counts, in case something get wrong:

Two error counts :
- a consecutive error count
- a total error count (only used for stats)

The increment/clear process is as follow:

- an error count is maintained on socket reads (and soon on socket write)
- when an operation fail, both error counts are incremented
- when an operation succeed, only the consecutive error count is cleared

- if the consecutive error count reach a critical level, something should
  be done about the node ... (this is not yet done, and will be done on the
  same time than the FIXME in kcomd_exit)

Index: linux/include/hpc/kcom.h
===================================================================
--- linux.orig/include/hpc/kcom.h	2006-11-21 18:50:31.000000000 +0100
+++ linux/include/hpc/kcom.h	2006-11-23 18:42:43.000000000 +0100
@@ -121,7 +121,9 @@
 	struct list_head tasks; 	/* list of task */
 	struct list_head list; 		/* list of nodes */
 	struct list_head process_list;  /* list used internally by kcomd */
-
+	/* Socket read/write error counts */
+	int error_count;		/* Consecutive error count on this node */
+	int error_total;		/* Total error count from the struct creation (stats only)*/
 };
 
 struct kcom_task
@@ -144,4 +146,8 @@
 extern void kcom_node_sock_release(struct kcom_node *node);
 extern struct kcom_node *__create_connection(struct sockaddr *saddr
 				            ,struct kcom_node *node);
+
+
+extern int kcom_node_increment_error(struct kcom_node *node);
+extern void kcom_node_clear_error(struct kcom_node *node);
 #endif /* _HPC_KCOM_H */
Index: linux/hpc/kcom.c
===================================================================
--- linux.orig/hpc/kcom.c	2006-11-21 18:51:05.000000000 +0100
+++ linux/hpc/kcom.c	2006-11-23 18:42:43.000000000 +0100
@@ -1586,3 +1586,49 @@
 
 	return i;
 }
+
+/**
+ * kcom_node_increment_error
+ *
+ * Description:
+ *    Increment both permanent and consecutive error count,
+ *    if the error count reach a critical level, delete all
+ *    tasks and delete the node ...
+ *
+ *    Returns the new consecutive error count
+ *
+ *    FIXME: the node deletion is not yet done, but should be!!
+ **/
+
+int kcom_node_increment_error(struct kcom_node *node)
+{
+	if (!node)
+		return -ENODEV;
+
+	node->error_count++;
+	node->error_total++;
+
+	if (60 > node->error_count) {
+		/*FIXME This should expel all tasks, and delete the node ... should be ...*/
+		OMBUG( "FIXME: A node had reached a critical error count level,"
+		       "but we don't delete it yet :(\n");
+	}
+	return node->error_count;
+}
+
+/**
+ * kcom_node_clear_error
+ *
+ * Description:
+ *    Clear the consecutive error count, but not the total error count
+ *
+ *    Consecutive error count should be cleared every time a socket write success
+ **/
+
+void kcom_node_clear_error(struct kcom_node *node)
+{
+	if (!node)
+		return;
+
+	node->error_count = 0;
+}
Index: linux/hpc/kcomd.c
===================================================================
--- linux.orig/hpc/kcomd.c	2006-11-21 19:01:42.000000000 +0100
+++ linux/hpc/kcomd.c	2006-11-23 18:42:43.000000000 +0100
@@ -657,8 +657,12 @@
 		INIT_LIST_HEAD(&node->process_list);
 
 		/* If we got an error that far ... we must kill the offending connection */
-		if (err < 0)
+		if (err < 0) {
+			kcom_node_increment_error(node);
 			kcom_node_sock_release(node);
+		}
+
+		kcom_node_clear_error(node);
 
 	}
 	if (!list_empty(&process_list)) {