logging tweak

[l2tpns.git] / cluster.c

diff --git a/cluster.c b/cluster.c
index 6f72394..645403c 100644 (file)
--- a/cluster.c
+++ b/cluster.c
@@ -1,6 +1,6 @@
 // L2TPNS Clustering Stuff
 
 // L2TPNS Clustering Stuff
 

-char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.6 2005/05/21 13:05:36 bodea Exp $";
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.26.2.11 2005/05/30 02:55:40 bodea Exp $";

 
 #include <stdio.h>
 #include <stdlib.h>
 
 #include <stdio.h>
 #include <stdlib.h>


@@ -192,7 +192,7 @@ static void add_type(char **p, int type, int more, char *data, int size)
 }
 
 // advertise our presence via BGP or gratuitous ARP
 }
 
 // advertise our presence via BGP or gratuitous ARP

-static void advertise(void)
+static void advertise_routes(void)

 {
 #ifdef BGP
        if (bgp_configured)
 {
 #ifdef BGP
        if (bgp_configured)


@@ -203,6 +203,15 @@ static void advertise(void)
                        send_garp(config->bind_address);        // Start taking traffic.
 }
 
                        send_garp(config->bind_address);        // Start taking traffic.
 }
 

+// withdraw our routes (BGP only)
+static void withdraw_routes(void)
+{
+#ifdef BGP
+       if (bgp_configured)
+               bgp_enable_routing(0);
+#endif /* BGP */
+}
+

 static void cluster_uptodate(void)
 {
        if (config->cluster_iam_uptodate)
 static void cluster_uptodate(void)
 {
        if (config->cluster_iam_uptodate)


@@ -214,7 +223,7 @@ static void cluster_uptodate(void)
        config->cluster_iam_uptodate = 1;
 
        LOG(0, 0, 0, "Now uptodate with master.\n");
        config->cluster_iam_uptodate = 1;
 
        LOG(0, 0, 0, "Now uptodate with master.\n");

-       advertise();
+       advertise_routes();

 }
 
 //
 }
 
 //


@@ -456,17 +465,22 @@ void cluster_check_slaves(void)
                        continue;       // Shutdown peer! Skip them.
 
                if (peers[i].uptodate)
                        continue;       // Shutdown peer! Skip them.
 
                if (peers[i].uptodate)

-                       have_peers = 1;
-
-               if (!peers[i].uptodate)
+                       have_peers++;
+               else

                        config->cluster_iam_uptodate = 0; // Start fast heartbeats
        }
 
                        config->cluster_iam_uptodate = 0; // Start fast heartbeats
        }
 

-#ifdef BGP
-       // in a cluster, withdraw/add routes when we get a peer/lose all peers
-       if (bgp_configured && have_peers != had_peers)
-               bgp_enable_routing(!have_peers);
-#endif /* BGP */
+       // in a cluster, withdraw/add routes when we get a peer/lose peers
+       if (have_peers != had_peers)
+       {
+               if (had_peers < config->cluster_master_min_adv &&
+                   have_peers >= config->cluster_master_min_adv)
+                       withdraw_routes();
+
+               else if (had_peers >= config->cluster_master_min_adv &&
+                   have_peers < config->cluster_master_min_adv)
+                       advertise_routes();
+       }

 }
 
 //
 }
 
 //


@@ -479,6 +493,7 @@ void cluster_check_master(void)
        int last_free = 0;
        clockt t = TIME;
        static int probed = 0;
        int last_free = 0;
        clockt t = TIME;
        static int probed = 0;

+       int have_peers;

 
        if (config->cluster_iam_master)
                return;         // Only runs on the slaves...
 
        if (config->cluster_iam_master)
                return;         // Only runs on the slaves...


@@ -511,7 +526,11 @@ void cluster_check_master(void)
 
        LOG(0, 0, 0, "Master timed out! Holding election...\n");
 
 
        LOG(0, 0, 0, "Master timed out! Holding election...\n");
 

-       for (i = 0; i < num_peers; i++)
+       // In the process of shutting down, can't be master
+       if (main_quit)
+               return;
+
+       for (i = have_peers = 0; i < num_peers; i++)

        {
                if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
                        continue;       // Stale peer! Skip them.
        {
                if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
                        continue;       // Stale peer! Skip them.


@@ -529,6 +548,9 @@ void cluster_check_master(void)
                        LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
                        return;         // They'll win the election. Wait for them to come up.
                }
                        LOG(1, 0, 0, "Expecting %s to become master\n", fmtaddr(peers[i].peer, 0));
                        return;         // They'll win the election. Wait for them to come up.
                }

+
+               if (peers[i].uptodate)
+                       have_peers++;

        }
 
                // Wow. it's been ages since I last heard a heartbeat
        }
 
                // Wow. it's been ages since I last heard a heartbeat


@@ -540,6 +562,11 @@ void cluster_check_master(void)
 
        LOG(0, 0, 0, "I am declaring myself the master!\n");
 
 
        LOG(0, 0, 0, "I am declaring myself the master!\n");
 

+       if (have_peers < config->cluster_master_min_adv)
+               advertise_routes();
+       else
+               withdraw_routes();
+

        if (config->cluster_seq_number == -1)
                config->cluster_seq_number = 0;
 
        if (config->cluster_seq_number == -1)
                config->cluster_seq_number = 0;
 


@@ -620,14 +647,6 @@ void cluster_check_master(void)
        config->cluster_undefined_tunnels = 0;
        config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
 
        config->cluster_undefined_tunnels = 0;
        config->cluster_iam_uptodate = 1; // assume all peers are up-to-date
 

-       if (!num_peers) // lone master
-               advertise();
-#ifdef BGP
-       else if (bgp_configured)
-               bgp_enable_routing(0);
-#endif /* BGP */
-
-

        // FIXME. We need to fix up the tunnel control message
        // queue here! There's a number of other variables we
        // should also update.
        // FIXME. We need to fix up the tunnel control message
        // queue here! There's a number of other variables we
        // should also update.


@@ -903,11 +922,10 @@ static int cluster_catchup_slave(int seq, in_addr_t slave)
 
        LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
        if (!config->cluster_iam_master) {
 
        LOG(1, 0, 0, "Slave %s sent LASTSEEN with seq %d\n", fmtaddr(slave, 0), seq);
        if (!config->cluster_iam_master) {

-               LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Sending a null PING.\n");
-                       // Send a ping to the stray slave saying that we're a master that's
-                       // just shutdown. This should force the slave to listen for the real
-                       // master.
-               peer_send_message(slave, C_PING, 0, NULL, 0);
+               LOG(1, 0, 0, "Got LASTSEEN but I'm not a master! Redirecting it to %s.\n",
+                       fmtaddr(config->cluster_master_address, 0));
+
+               peer_send_message(slave, C_MASTER, config->cluster_master_address, NULL, 0);

                return 0;
        }
 
                return 0;
        }
 


@@ -978,8 +996,10 @@ static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size
        }
 
        // Is this the master shutting down??
        }
 
        // Is this the master shutting down??

-       if (peer == config->cluster_master_address && !basetime) {
-               LOG(3, 0, 0, "Master %s shutting down...\n", fmtaddr(config->cluster_master_address, 0));
+       if (peer == config->cluster_master_address) {
+               LOG(3, 0, 0, "Master %s %s\n", fmtaddr(config->cluster_master_address, 0),
+                       basetime ? "has restarted!" : "shutting down...");
+

                config->cluster_master_address = 0;
                config->cluster_last_hb = 0; // Force an election.
                cluster_check_master();
                config->cluster_master_address = 0;
                config->cluster_last_hb = 0; // Force an election.
                cluster_check_master();


@@ -1020,6 +1040,20 @@ static int cluster_add_peer(in_addr_t peer, time_t basetime, pingt *pp, int size
        return 1;
 }
 
        return 1;
 }
 

+// A slave responds with C_MASTER when it gets a message which should have gone to a master.
+static int cluster_set_master(in_addr_t peer, in_addr_t master)
+{
+       if (config->cluster_iam_master) // Sanity...
+               return 0;
+
+       LOG(3, 0, 0, "Peer %s set the master to %s...\n", fmtaddr(peer, 0),
+               fmtaddr(master, 1));
+
+       config->cluster_master_address = master;
+       cluster_check_master();
+       return 0;
+}
+

 /* Handle the slave updating the byte counters for the master. */
 //
 // Note that we don't mark the session as dirty; We rely on
 /* Handle the slave updating the byte counters for the master. */
 //
 // Note that we don't mark the session as dirty; We rely on


@@ -1394,10 +1428,13 @@ int processcluster(char *data, int size, in_addr_t addr)
        s -= sizeof(uint32_t);
 
        switch (type) {
        s -= sizeof(uint32_t);
 
        switch (type) {

-       case C_PING:    // Update the peers table.
+       case C_PING: // Update the peers table.

                return cluster_add_peer(addr, more, (pingt *) p, s);
 
                return cluster_add_peer(addr, more, (pingt *) p, s);
 

-       case C_LASTSEEN:        // Catch up a slave (slave missed a packet).
+       case C_MASTER: // Our master is wrong
+               return cluster_set_master(addr, more);
+
+       case C_LASTSEEN: // Catch up a slave (slave missed a packet).

                return cluster_catchup_slave(more, addr);
 
        case C_FORWARD: { // Forwarded control packet. pass off to processudp.
                return cluster_catchup_slave(more, addr);
 
        case C_FORWARD: { // Forwarded control packet. pass off to processudp.


@@ -1438,6 +1475,11 @@ int processcluster(char *data, int size, in_addr_t addr)
                return 0;
 
        case C_BYTES:
                return 0;
 
        case C_BYTES:

+               if (!config->cluster_iam_master) {
+                       LOG(0, 0, 0, "I'm not the master, but I got a C_BYTES from %s?\n", fmtaddr(addr, 0));
+                       return -1;
+               }
+

                return cluster_handle_bytes(p, s);
 
        case C_KILL:    // The master asked us to die!? (usually because we're too out of date).
                return cluster_handle_bytes(p, s);
 
        case C_KILL:    // The master asked us to die!? (usually because we're too out of date).