int
wd_notice_server_down(void)
{
int rtn;
wd_IP_down();
/* send notice server down packet */
rtn = wd_send_packet_no(WD_SERVER_DOWN);
return rtn;
}
/*
* Check if pgpool is living
*/
int
wd_lifecheck(void)
{
struct timeval tv;
/* I'm in down.... */
if (WD_MYSELF->status == WD_DOWN)
{
ereport(NOTICE,
(errmsg("watchdog lifecheck, watchdog status is DOWN. You need to restart pgpool")));
return WD_NG;
}
/* set startup time */
gettimeofday(&tv, NULL);
/* check upper connection */
if (strlen(pool_config->trusted_servers))
{
if(wd_is_upper_ok(pool_config->trusted_servers) != WD_OK)
{
ereport(WARNING,
(errmsg("watchdog lifecheck, failed to connect to any trusted servers")));
if (WD_MYSELF->status == WD_MASTER &&
strlen(pool_config->delegate_IP) != 0)
{
wd_IP_down();
}
wd_set_myself(&tv, WD_DOWN);
wd_notice_server_down();
return WD_NG;
}
}
/* skip lifecheck during recovery execution */
if (*InRecovery != RECOVERY_INIT)
{
return WD_OK;
}
/* check and update pgpool status */
check_pgpool_status();
return WD_OK;
}
static int
wd_send_response(int sock, WdPacket * recv_pack)
{
int rtn = WD_NG;
WdInfo * p, *q;
WdNodeInfo * node;
WdLockInfo * lock;
WdPacket send_packet;
struct timeval tv;
char pack_str[WD_MAX_PACKET_STRING];
int pack_str_len;
char hash[(MD5_PASSWD_LEN+1)*2];
bool is_node_packet = false;
if (recv_pack == NULL)
{
return rtn;
}
memset(&send_packet, 0, sizeof(WdPacket));
p = &(recv_pack->wd_body.wd_info);
/* auhtentication */
if (strlen(pool_config->wd_authkey))
{
/* calculate hash from packet */
pack_str_len = wd_packet_to_string(*recv_pack, pack_str, sizeof(pack_str));
wd_calc_hash(pack_str, pack_str_len, hash);
if (strcmp(recv_pack->hash, hash))
{
pool_log("wd_send_response: watchdog authentication failed");
rtn = wd_authentication_failed(sock);
return rtn;
}
}
/* set response packet no */
switch (recv_pack->packet_no)
{
/* add request into the watchdog list */
case WD_ADD_REQ:
p = &(recv_pack->wd_body.wd_info);
if (wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status) > 0)
{
send_packet.packet_no = WD_ADD_ACCEPT;
}
else
{
send_packet.packet_no = WD_ADD_REJECT;
}
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
break;
/* announce candidacy to be the new master */
case WD_STAND_FOR_MASTER:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status);
/* check exist master */
if ((q = wd_is_alive_master()) != NULL)
{
/* vote against the candidate */
send_packet.packet_no = WD_MASTER_EXIST;
memcpy(&(send_packet.wd_body.wd_info), q, sizeof(WdInfo));
}
else
{
if (WD_MYSELF->tv.tv_sec <= p->tv.tv_sec )
{
memcpy(&tv,&(p->tv),sizeof(struct timeval));
tv.tv_sec += 1;
wd_set_myself(&tv, WD_NORMAL);
}
/* vote for the candidate */
send_packet.packet_no = WD_VOTE_YOU;
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
}
break;
/* announce assumption to be the new master */
case WD_DECLARE_NEW_MASTER:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status);
if (WD_MYSELF->status == WD_MASTER)
{
/* resign master server */
pool_log("wd_declare_new_master: ifconfig down to resign master server");
wd_IP_down();
wd_set_myself(NULL, WD_NORMAL);
}
send_packet.packet_no = WD_READY;
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
break;
/* annouce to assume lock holder */
case WD_STAND_FOR_LOCK_HOLDER:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status);
/* only master handles lock holder privilege */
if (WD_MYSELF->status == WD_MASTER)
{
/* if theare are no lock holder yet */
if (wd_get_lock_holder() != NULL)
{
send_packet.packet_no = WD_LOCK_HOLDER_EXIST;
}
}
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
break;
case WD_DECLARE_LOCK_HOLDER:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status);
wd_set_lock_holder(p, true);
send_packet.packet_no = WD_READY;
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
break;
/* annouce to resigne lock holder */
case WD_RESIGN_LOCK_HOLDER:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status);
wd_set_lock_holder(p, false);
send_packet.packet_no = WD_READY;
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
break;
case WD_START_INTERLOCK:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status);
wd_set_interlocking(p, true);
break;
case WD_END_INTERLOCK:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), p->status);
wd_set_interlocking(p, false);
break;
/* announce that server is down */
case WD_SERVER_DOWN:
p = &(recv_pack->wd_body.wd_info);
wd_set_wd_list(p->hostname,p->pgpool_port, p->wd_port, p->delegate_ip, &(p->tv), WD_DOWN);
send_packet.packet_no = WD_READY;
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
if (wd_am_I_oldest() == WD_OK && WD_MYSELF->status != WD_MASTER)
{
wd_escalation();
}
break;
/* announce start online recovery */
case WD_START_RECOVERY:
if (*InRecovery != RECOVERY_INIT)
{
send_packet.packet_no = WD_NODE_FAILED;
}
else
{
send_packet.packet_no = WD_NODE_READY;
*InRecovery = RECOVERY_ONLINE;
if (wait_connection_closed() != 0)
{
send_packet.packet_no = WD_NODE_FAILED;
}
}
break;
case WD_END_RECOVERY:
send_packet.packet_no = WD_NODE_READY;
*InRecovery = RECOVERY_INIT;
kill(wd_ppid, SIGUSR2);
break;
case WD_FAILBACK_REQUEST:
node = &(recv_pack->wd_body.wd_node_info);
wd_set_node_mask(WD_FAILBACK_REQUEST,node->node_id_set,node->node_num);
is_node_packet = true;
send_packet.packet_no = WD_NODE_READY;
break;
case WD_DEGENERATE_BACKEND:
node = &(recv_pack->wd_body.wd_node_info);
wd_set_node_mask(WD_DEGENERATE_BACKEND,node->node_id_set, node->node_num);
is_node_packet = true;
send_packet.packet_no = WD_NODE_READY;
break;
case WD_PROMOTE_BACKEND:
node = &(recv_pack->wd_body.wd_node_info);
wd_set_node_mask(WD_PROMOTE_BACKEND,node->node_id_set, node->node_num);
is_node_packet = true;
send_packet.packet_no = WD_NODE_READY;
break;
case WD_UNLOCK_REQUEST:
lock = &(recv_pack->wd_body.wd_lock_info);
wd_set_lock(lock->lock_id, false);
send_packet.packet_no = WD_LOCK_READY;
break;
default:
send_packet.packet_no = WD_INVALID;
memcpy(&(send_packet.wd_body.wd_info), WD_MYSELF, sizeof(WdInfo));
break;
}
/* send response packet */
rtn = wd_send_packet(sock, &send_packet);
/* send node request signal.
* wd_node_request_singnal() uses a semaphore lock internally, so should be
* called after sending a response pakcet to prevent dead lock.
*/
if (is_node_packet)
wd_node_request_signal(recv_pack->packet_no, node);
return rtn;
}
/*
* fork de-escalation process
*/
pid_t
fork_plunging_process(void)
{
pid_t pid;
pid = fork();
if (pid != 0)
{
if (pid == -1)
ereport(NOTICE,
(errmsg("failed to fork a de-escalation process")));
return pid;
}
on_exit_reset();
processType = PT_WATCHDOG_UTILITY;
POOL_SETMASK(&UnBlockSig);
init_ps_display("", "", "", "");
pool_signal(SIGTERM, wd_exit);
pool_signal(SIGINT, wd_exit);
pool_signal(SIGQUIT, wd_exit);
pool_signal(SIGCHLD, SIG_DFL);
pool_signal(SIGHUP, SIG_IGN);
pool_signal(SIGPIPE, SIG_IGN);
MemoryContextSwitchTo(TopMemoryContext);
set_ps_display("watchdog de-escalation", false);
ereport(LOG,
(errmsg("watchdog: de-escalation started")));
/*
* STEP 1 execute de-escalation command provided by user in pgpool conf
* file
*/
if (strlen(pool_config->wd_de_escalation_command))
{
int r = system(pool_config->wd_de_escalation_command);
if (WIFEXITED(r))
{
if (WEXITSTATUS(r) == EXIT_SUCCESS)
ereport(LOG,
(errmsg("watchdog de-escalation successful")));
else
{
ereport(WARNING,
(errmsg("watchdog de-escalation command failed with exit status: %d", WEXITSTATUS(r))));
}
}
else
{
ereport(WARNING,
(errmsg("watchdog de-escalation command exit abnormally")));
}
}
/*
* STEP 2 bring down the delegate IP
*/
if (strlen(pool_config->delegate_IP) != 0)
{
if (wd_IP_down() != WD_OK)
ereport(WARNING,
(errmsg("watchdog de-escalation failed to bring down delegate IP")));
}
exit(0);
}
