IPALLOCATION : If the node is held pinned down in "init" state

[sahlberg/ctdb.git] / tcp / tcp_init.c

diff --git a/tcp/tcp_init.c b/tcp/tcp_init.c
index f5d4e4c1d6b4a6d2d36792ab6569704bc4756242..95141d5610d8b283d84412c4208b9413f7d1d250 100644 (file)
--- a/tcp/tcp_init.c
+++ b/tcp/tcp_init.c
@@ -19,29 +19,39 @@
 
 #include "includes.h"
 #include "lib/tdb/include/tdb.h"
-#include "lib/events/events.h"
+#include "lib/tevent/tevent.h"
 #include "system/network.h"
 #include "system/filesys.h"
 #include "../include/ctdb_private.h"
 #include "ctdb_tcp.h"
 
+static int tnode_destructor(struct ctdb_tcp_node *tnode)
+{
+  //   struct ctdb_node *node = talloc_find_parent_bytype(tnode, struct ctdb_node);
+
+       if (tnode->fd != -1) {
+               close(tnode->fd);
+               tnode->fd = -1;
+       }
+
+       return 0;
+}
 
 /*
   initialise tcp portion of a ctdb node 
 */
 static int ctdb_tcp_add_node(struct ctdb_node *node)
 {
-       struct ctdb_tcp *ctcp = talloc_get_type(node->ctdb->private_data,
-                                               struct ctdb_tcp);
        struct ctdb_tcp_node *tnode;
-       tnode = talloc_zero(ctcp, struct ctdb_tcp_node);
+       tnode = talloc_zero(node, struct ctdb_tcp_node);
        CTDB_NO_MEMORY(node->ctdb, tnode);
 
        tnode->fd = -1;
        node->private_data = tnode;
+       talloc_set_destructor(tnode, tnode_destructor);
 
-       tnode->out_queue = ctdb_queue_setup(node->ctdb, ctcp, tnode->fd, CTDB_TCP_ALIGNMENT,
-                                       ctdb_tcp_tnode_cb, node);
+       tnode->out_queue = ctdb_queue_setup(node->ctdb, node, tnode->fd, CTDB_TCP_ALIGNMENT,
+                                           ctdb_tcp_tnode_cb, node, "to-node-%s", node->name);
        
        return 0;
 }
@@ -54,11 +64,17 @@ static int ctdb_tcp_initialise(struct ctdb_context *ctdb)
        int i;
 
        /* listen on our own address */
-       if (ctdb_tcp_listen(ctdb) != 0) return -1;
+       if (ctdb_tcp_listen(ctdb) != 0) {
+               DEBUG(DEBUG_CRIT, (__location__ " Failed to start listening on the CTDB socket\n"));
+               exit(1);
+       }
 
-       for (i=0; i<ctdb->num_nodes; i++) {
+       for (i=0; i < ctdb->num_nodes; i++) {
+               if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
+                       continue;
+               }
                if (ctdb_tcp_add_node(ctdb->nodes[i]) != 0) {
-                       DEBUG(0, ("methods->add_node failed at %d\n", i));
+                       DEBUG(DEBUG_CRIT, ("methods->add_node failed at %d\n", i));
                        return -1;
                }
        }
@@ -69,25 +85,40 @@ static int ctdb_tcp_initialise(struct ctdb_context *ctdb)
 /*
   start the protocol going
 */
-static int ctdb_tcp_start(struct ctdb_context *ctdb)
+static int ctdb_tcp_connect_node(struct ctdb_node *node)
 {
-       int i;
+       struct ctdb_context *ctdb = node->ctdb;
+       struct ctdb_tcp_node *tnode = talloc_get_type(
+               node->private_data, struct ctdb_tcp_node);
 
-       /* startup connections to the other servers - will happen on
+       /* startup connection to the other server - will happen on
           next event loop */
-       for (i=0;i<ctdb->num_nodes;i++) {
-               struct ctdb_node *node = *(ctdb->nodes + i);
-               struct ctdb_tcp_node *tnode = talloc_get_type(
-                       node->private_data, struct ctdb_tcp_node);
-               if (!ctdb_same_address(&ctdb->address, &node->address)) {
-                       event_add_timed(ctdb->ev, tnode, timeval_zero(), 
-                                       ctdb_tcp_node_connect, node);
-               }
+       if (!ctdb_same_address(&ctdb->address, &node->address)) {
+               tnode->connect_te = event_add_timed(ctdb->ev, tnode, 
+                                                   timeval_zero(), 
+                                                   ctdb_tcp_node_connect, node);
        }
 
        return 0;
 }
 
+/*
+  shutdown and try to restart a connection to a node after it has been
+  disconnected
+*/
+static void ctdb_tcp_restart(struct ctdb_node *node)
+{
+       struct ctdb_tcp_node *tnode = talloc_get_type(
+               node->private_data, struct ctdb_tcp_node);
+
+       DEBUG(DEBUG_NOTICE,("Tearing down connection to dead node :%d\n", node->pnn));
+
+       ctdb_tcp_stop_connection(node);
+
+       tnode->connect_te = event_add_timed(node->ctdb->ev, tnode, timeval_zero(), 
+                                           ctdb_tcp_node_connect, node);
+}
+
 
 /*
   shutdown the transport
@@ -100,6 +131,23 @@ static void ctdb_tcp_shutdown(struct ctdb_context *ctdb)
        ctdb->private_data = NULL;
 }
 
+/*
+  start the transport
+*/
+static int ctdb_tcp_start(struct ctdb_context *ctdb)
+{
+       int i;
+
+       for (i=0; i < ctdb->num_nodes; i++) {
+               if (ctdb->nodes[i]->flags & NODE_FLAGS_DELETED) {
+                       continue;
+               }
+               ctdb_tcp_connect_node(ctdb->nodes[i]);
+       }
+
+       return 0;
+}
+
 
 /*
   transport packet allocator - allows transport to control memory for packets
@@ -119,10 +167,21 @@ static const struct ctdb_methods ctdb_tcp_methods = {
        .start        = ctdb_tcp_start,
        .queue_pkt    = ctdb_tcp_queue_pkt,
        .add_node     = ctdb_tcp_add_node,
+       .connect_node = ctdb_tcp_connect_node,
        .allocate_pkt = ctdb_tcp_allocate_pkt,
        .shutdown     = ctdb_tcp_shutdown,
+       .restart      = ctdb_tcp_restart,
 };
 
+static int tcp_ctcp_destructor(struct ctdb_tcp *ctcp)
+{
+       ctcp->ctdb->private_data = NULL;
+       ctcp->ctdb->methods = NULL;
+       
+       return 0;
+}
+
+               
 /*
   initialise tcp portion of ctdb 
 */
@@ -133,8 +192,11 @@ int ctdb_tcp_init(struct ctdb_context *ctdb)
        CTDB_NO_MEMORY(ctdb, ctcp);
 
        ctcp->listen_fd = -1;
+       ctcp->ctdb      = ctdb;
        ctdb->private_data = ctcp;
        ctdb->methods = &ctdb_tcp_methods;
+
+       talloc_set_destructor(ctcp, tcp_ctcp_destructor);
        return 0;
 }