// L2TPNS Clustering Stuff
-// $Id: cluster.c,v 1.3 2004-06-23 03:52:24 fred_nerk Exp $
+
+char const *cvs_id_cluster = "$Id: cluster.c,v 1.6 2004-07-05 06:54:01 bodea Exp $";
#include <stdio.h>
#include <sys/file.h>
return -1;
}
- config->cluster_last_hb = config->current_time;
+ config->cluster_last_hb = TIME;
config->cluster_seq_number = -1;
return cluster_sockfd;
if (size > sizeof(past_hearts[0].data)) {
log(0,0,0,0, "Tried to heartbeat something larger than the maximum packet!\n");
kill(0, SIGTERM);
+ exit(1);
}
i = seq % HB_HISTORY_SIZE;
past_hearts[i].seq = seq;
int i, count, tcount, high_sid = 0;
int last_free = 0;
int had_peers = have_peers;
- clockt t = config->current_time;
+ clockt t = TIME;
+ static int probed = 0;
+
+ // Is the master late? If so, try probing it...
+ if (TIME > (config->cluster_last_hb + config->cluster_hb_timeout/8 + 11)) {
+ if (!probed) {
+ if (config->cluster_master_address) {
+ peer_send_message(config->cluster_master_address,
+ C_LASTSEEN, config->cluster_seq_number, NULL, 0);
+ probed = 1;
+ }
+ }
+ } else { // We got a recent heartbeat; reset the probe flag.
+ probed = 0;
+ }
- if (config->current_time < (config->cluster_last_hb + HB_TIMEOUT) )
+ if (TIME < (config->cluster_last_hb + config->cluster_hb_timeout) )
return; // Everything's ok. return.
if (!config->cluster_iam_master)
log(0,0,0,0, "Master timed out! Holding election...\n");
- config->cluster_last_hb = config->current_time + 1;
+ config->cluster_last_hb = TIME + 1;
for (i = have_peers = 0; i < num_peers ; ++i) {
- if ((peers[i].timestamp + HB_TIMEOUT) < t)
+ if ((peers[i].timestamp + config->cluster_hb_timeout) < t)
continue; // Stale peer! Skip them.
if (!peers[i].basetime)
}
if (p > (buff + sizeof(buff))) { // Did we somehow manage to overun the buffer?
- log(0,0,0,0, "Overrun the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
+ log(0,0,0,0, "FATAL: Overran the heartbeat buffer! This is fatal. Exiting. (size %d)\n", p - buff);
kill(0, SIGTERM);
}
//
// Fill out the packet with tunnels from the tunnel table...
+ // This effectively means we walk the tunnel table more quickly
+ // than the session table. This is good because stuffing up a
+ // tunnel is a much bigger deal than stuffing up a session.
//
while ( (p + sizeof(u32) * 2 + sizeof(tunnelt) ) < (buff + MAX_HEART_SIZE) ) {
kill(0, SIGTERM);
}
- log(3,0,0,0, "Sending heartbeat with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun size %d)\n",
- config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
+ log(3,0,0,0, "Sending heartbeat #%d with %d changes (%d x-sess, %d x-tunnels, %d highsess, %d hightun size %d)\n",
+ h.seq, config->cluster_num_changes, count, tcount, config->cluster_highest_sessionid,
config->cluster_highest_tunnelid, (p-buff));
config->cluster_num_changes = 0;
// This peer already exists. Just update the timestamp.
peers[i].basetime = basetime;
- peers[i].timestamp = config->current_time;
+ peers[i].timestamp = TIME;
break;
}
{
if (peers[i].peer != peer)
continue;
- if ((peers[i].timestamp + HB_TIMEOUT * 10) < config->current_time) // Stale.
+ if ((peers[i].timestamp + config->cluster_hb_timeout * 10) < TIME) // Stale.
break;
}
peers[i].peer = peer;
peers[i].basetime = basetime;
- peers[i].timestamp = config->current_time;
+ peers[i].timestamp = TIME;
if (i == num_peers)
++num_peers;
log(0,0,0,0, "I just got a packet claiming to be from a master but _I_ am the master!\n");
if (!h->basetime) {
- log(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(htonl(addr)) );
+ log(0,0,0,0, "Heartbeat from addr %s with zero basetime!\n", inet_toa(addr) );
return -1; // Skip it.
}
if (basetime > h->basetime) {
- log(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(htonl(addr)));
+ log(0,0,0,0, "They're (%s) an older master than me so I'm gone!\n", inet_toa(addr));
kill(0, SIGTERM);
+ exit(1);
}
if (basetime == h->basetime && my_address < addr) { // Tie breaker.
log(0,0,0,0, "They're a higher IP address than me, so I'm gone!\n");
kill(0, SIGTERM);
+ exit(1);
}
return -1; // Skip it.
}
if (config->cluster_seq_number == -1) // Don't have one. Just align to the master...
config->cluster_seq_number = h->seq;
- config->cluster_last_hb = config->current_time; // Reset to ensure that we don't become master!!
+ config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
if (config->cluster_seq_number != h->seq) { // Out of sequence heartbeat!
log(1,0,0,0, "HB: Got seq# %d but was expecting %d. asking for resend.\n", h->seq, config->cluster_seq_number);
peer_send_message(addr, C_LASTSEEN, config->cluster_seq_number, NULL, 0);
- config->cluster_last_hb = config->current_time; // Reset to ensure that we don't become master!!
+ config->cluster_last_hb = TIME; // Reset to ensure that we don't become master!!
// Just drop the packet. The master will resend it as part of the catchup.
}
config->cluster_master_address = addr;
- config->cluster_last_hb = config->current_time; // Successfully received a heartbeat!
+ config->cluster_last_hb = TIME; // Successfully received a heartbeat!
return 0;
shortpacket:
}
if (addr != config->cluster_master_address) {
- log(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)",
+ log(0,0,0,0, "Received a C_KILL from %s which doesn't match config->cluster_master_address (%x)\n",
inet_toa(addr), config->cluster_master_address);
// We can only warn about it. The master might really have switched!
}
- log(0,0,0,0, "Received a valid C_KILL: I'm going to die now.");
+ log(0,0,0,0, "Received a valid C_KILL: I'm going to die now.\n");
kill(0, SIGTERM);
exit(0); // Lets be paranoid;
return -1; // Just signalling the compiler.
{
int i;
+ if (CLI_HELP_REQUESTED)
+ return CLI_HELP_NO_ARGS;
+
cli_print(cli, "Cluster status : %s", config->cluster_iam_master ? "Master" : "Slave" );
cli_print(cli, "My address : %s", inet_toa(my_address));
cli_print(cli, "VIP address : %s", inet_toa(config->bind_address));
if (!config->cluster_iam_master) {
cli_print(cli, "My master : %s (last heartbeat %.1f seconds old)",
config->cluster_master_address ? inet_toa(config->cluster_master_address) : "Not defined",
- 0.1 * (config->current_time - config->cluster_last_hb));
+ 0.1 * (TIME - config->cluster_last_hb));
cli_print(cli, "Uptodate : %s", config->cluster_iam_uptodate ? "Yes" : "No");
cli_print(cli, "Next sequence number expected: %d", config->cluster_seq_number);
cli_print(cli, "%d sessions undefined of %d", config->cluster_undefined_sessions, config->cluster_highest_sessionid);
cli_print(cli, "%20s %10s %8s", "Address", "Basetime", "Age");
for (i = 0; i < num_peers; ++i) {
cli_print(cli, "%20s %10d %8d", inet_toa(peers[i].peer),
- peers[i].basetime, config->current_time - peers[i].timestamp);
+ peers[i].basetime, TIME - peers[i].timestamp);
}
return CLI_OK;
}
projects / l2tpns.git / blobdiff