Markus Makela 7ef8b187b5 Do hangups only after server states have been updated
The hangup code was refactored into a common function which should only be
used after the server states have been updated. This will remove erroneus
connections to already failed servers.
2016-11-09 23:21:54 +02:00

826 lines
25 KiB
C

/*
* Copyright (c) 2016 MariaDB Corporation Ab
*
* Use of this software is governed by the Business Source License included
* in the LICENSE.TXT file and at www.mariadb.com/bsl.
*
* Change Date: 2019-01-01
*
* On the date above, in accordance with the Business Source License, use
* of this software will be governed by version 2 or later of the General
* Public License.
*/
/**
* @file galera_mon.c - A MySQL Galera cluster monitor
*
* @verbatim
* Revision History
*
* Date Who Description
* 22/07/13 Mark Riddoch Initial implementation
* 21/05/14 Massimiliano Pinto Monitor sets a master server that has the lowest value of wsrep_local_index
* 23/05/14 Massimiliano Pinto Added 1 configuration option (setInterval). Interval is printed in diagnostics.
* 03/06/14 Mark Riddoch Add support for maintenance mode
* 24/06/14 Massimiliano Pinto Added depth level 0 for each node
* 30/10/14 Massimiliano Pinto Added disableMasterFailback feature
* 10/11/14 Massimiliano Pinto Added setNetworkTimeout for connect,read,write
* 20/04/15 Guillaume Lefranc Added availableWhenDonor feature
* 22/04/15 Martin Brampton Addition of disableMasterRoleSetting
* 08/05/15 Markus Makela Addition of launchable scripts
* 17/10/15 Martin Brampton Change DCB callback to hangup
*
* @endverbatim
*/
#include <galeramon.h>
#include <dcb.h>
static void monitorMain(void *);
static char *version_str = "V2.0.0";
/** Log a warning when a bad 'wsrep_local_index' is found */
static bool warn_erange_on_local_index = true;
/* @see function load_module in load_utils.c for explanation of the following
* lint directives.
*/
/*lint -e14 */
MODULE_INFO info =
{
MODULE_API_MONITOR,
MODULE_GA,
MONITOR_VERSION,
"A Galera cluster monitor"
};
/*lint +e14 */
static void *startMonitor(void *, void*);
static void stopMonitor(void *);
static void diagnostics(DCB *, void *);
static MONITOR_SERVERS *get_candidate_master(MONITOR*);
static MONITOR_SERVERS *set_cluster_master(MONITOR_SERVERS *, MONITOR_SERVERS *, int);
static void disableMasterFailback(void *, int);
bool isGaleraEvent(monitor_event_t event);
static MONITOR_OBJECT MyObject =
{
startMonitor,
stopMonitor,
diagnostics
};
/**
* Implementation of the mandatory version entry point
*
* @return version string of the module
*
* @see function load_module in load_utils.c for explanation of the following
* lint directives.
*/
/*lint -e14 */
char *
version()
{
return version_str;
}
/**
* The module initialisation routine, called when the module
* is first loaded.
* @see function load_module in load_utils.c for explanation of lint
*/
/*lint -e14 */
void
ModuleInit()
{
MXS_NOTICE("Initialise the MySQL Galera Monitor module %s.", version_str);
}
/*lint +e14 */
/**
* The module entry point routine. It is this routine that
* must populate the structure that is referred to as the
* "module object", this is a structure with the set of
* external entry points for this module.
*
* @return The module object
*/
MONITOR_OBJECT *
GetModuleObject()
{
return &MyObject;
}
/*lint +e14 */
/**
* Start the instance of the monitor, returning a handle on the monitor.
*
* This function creates a thread to execute the actual monitoring.
*
* @return A handle to use when interacting with the monitor
*/
static void *
startMonitor(void *arg, void* opt)
{
MONITOR* mon = arg;
GALERA_MONITOR *handle = mon->handle;
CONFIG_PARAMETER* params = (CONFIG_PARAMETER*) opt;
bool have_events = false, script_error = false;
if (handle != NULL)
{
handle->shutdown = 0;
}
else
{
if ((handle = (GALERA_MONITOR *) malloc(sizeof(GALERA_MONITOR))) == NULL)
{
return NULL;
}
handle->shutdown = 0;
handle->id = MONITOR_DEFAULT_ID;
handle->disableMasterFailback = 0;
handle->availableWhenDonor = 0;
handle->disableMasterRoleSetting = 0;
handle->master = NULL;
handle->script = NULL;
handle->use_priority = false;
memset(handle->events, false, sizeof(handle->events));
spinlock_init(&handle->lock);
}
while (params)
{
if (!strcmp(params->name, "disable_master_failback"))
{
handle->disableMasterFailback = config_truth_value(params->value);
}
else if (!strcmp(params->name, "available_when_donor"))
{
handle->availableWhenDonor = config_truth_value(params->value);
}
else if (!strcmp(params->name, "disable_master_role_setting"))
{
handle->disableMasterRoleSetting = config_truth_value(params->value);
}
else if (!strcmp(params->name, "use_priority"))
{
handle->use_priority = config_truth_value(params->value);
}
else if (!strcmp(params->name, "script"))
{
if (externcmd_can_execute(params->value))
{
free(handle->script);
handle->script = strdup(params->value);
}
else
{
script_error = true;
}
}
else if (!strcmp(params->name, "events"))
{
if (mon_parse_event_string((bool *) handle->events,
sizeof(handle->events), params->value) != 0)
{
script_error = true;
}
else
{
have_events = true;
}
}
params = params->next;
}
/** SHOW STATUS doesn't require any special permissions */
if (!check_monitor_permissions(mon, "SHOW STATUS LIKE 'wsrep_local_state'"))
{
MXS_ERROR("Failed to start monitor. See earlier errors for more information.");
free(handle->script);
free(handle);
return NULL;
}
if (script_error)
{
MXS_ERROR("Errors were found in the script configuration parameters "
"for the monitor '%s'. The script will not be used.", mon->name);
free(handle->script);
handle->script = NULL;
}
/** If no specific events are given, enable them all */
if (!have_events)
{
memset(handle->events, true, sizeof(handle->events));
}
if (thread_start(&handle->thread, monitorMain, mon) == NULL)
{
MXS_ERROR("Failed to start monitor thread for monitor '%s'.", mon->name);
}
return handle;
}
/**
* Stop a running monitor
*
* @param arg Handle on thr running monior
*/
static void
stopMonitor(void *arg)
{
MONITOR* mon = (MONITOR*) arg;
GALERA_MONITOR *handle = (GALERA_MONITOR *) mon->handle;
handle->shutdown = 1;
thread_wait(handle->thread);
}
/**
* Diagnostic interface
*
* @param dcb DCB to send output
* @param arg The monitor handle
*/
static void
diagnostics(DCB *dcb, void *arg)
{
MONITOR* mon = (MONITOR*) arg;
GALERA_MONITOR *handle = (GALERA_MONITOR *) mon->handle;
MONITOR_SERVERS *db;
char *sep;
switch (handle->status)
{
case MONITOR_RUNNING:
dcb_printf(dcb, "\tMonitor running\n");
break;
case MONITOR_STOPPING:
dcb_printf(dcb, "\tMonitor stopping\n");
break;
case MONITOR_STOPPED:
dcb_printf(dcb, "\tMonitor stopped\n");
break;
}
dcb_printf(dcb, "\tSampling interval:\t%lu milliseconds\n", mon->interval);
dcb_printf(dcb, "\tMaster Failback:\t%s\n", (handle->disableMasterFailback == 1) ? "off" : "on");
dcb_printf(dcb, "\tAvailable when Donor:\t%s\n", (handle->availableWhenDonor == 1) ? "on" : "off");
dcb_printf(dcb, "\tMaster Role Setting Disabled:\t%s\n",
(handle->disableMasterRoleSetting == 1) ? "on" : "off");
dcb_printf(dcb, "\tConnect Timeout:\t%i seconds\n", mon->connect_timeout);
dcb_printf(dcb, "\tRead Timeout:\t\t%i seconds\n", mon->read_timeout);
dcb_printf(dcb, "\tWrite Timeout:\t\t%i seconds\n", mon->write_timeout);
dcb_printf(dcb, "\tMonitored servers: ");
db = mon->databases;
sep = "";
while (db)
{
dcb_printf(dcb, "%s%s:%d", sep, db->server->name, db->server->port);
sep = ", ";
db = db->next;
}
dcb_printf(dcb, "\n");
}
/**
* Monitor an individual server. Does not deal with the setting of master or
* slave bits, except for clearing them when a server is not joined to the
* cluster.
*
* @param handle The MySQL Monitor object
* @param database The database to probe
*/
static void
monitorDatabase(MONITOR *mon, MONITOR_SERVERS *database)
{
GALERA_MONITOR* handle = (GALERA_MONITOR*) mon->handle;
MYSQL_ROW row;
MYSQL_RES *result, *result2;
int isjoined = 0;
char *server_string;
SERVER temp_server;
/* Don't even probe server flagged as in maintenance */
if (SERVER_IN_MAINT(database->server))
{
return;
}
/** Store previous status */
database->mon_prev_status = database->server->status;
server_transfer_status(&temp_server, database->server);
server_clear_status(&temp_server, SERVER_RUNNING);
/* Also clear Joined */
server_clear_status(&temp_server, SERVER_JOINED);
connect_result_t rval = mon_connect_to_db(mon, database);
if (rval != MONITOR_CONN_OK)
{
if (mysql_errno(database->con) == ER_ACCESS_DENIED_ERROR)
{
server_set_status(&temp_server, SERVER_AUTH_ERROR);
}
else
{
server_clear_status(&temp_server, SERVER_AUTH_ERROR);
}
database->server->node_id = -1;
if (mon_status_changed(database) && mon_print_fail_status(database))
{
mon_log_connect_error(database, rval);
}
server_transfer_status(database->server, &temp_server);
return;
}
/* If we get this far then we have a working connection */
server_set_status(&temp_server, SERVER_RUNNING);
/* get server version string */
server_string = (char *) mysql_get_server_info(database->con);
if (server_string)
{
server_set_version_string(database->server, server_string);
}
/* Check if the the Galera FSM shows this node is joined to the cluster */
if (mysql_query(database->con, "SHOW STATUS LIKE 'wsrep_local_state'") == 0
&& (result = mysql_store_result(database->con)) != NULL)
{
if (mysql_field_count(database->con) < 2)
{
mysql_free_result(result);
MXS_ERROR("Unexpected result for \"SHOW STATUS LIKE 'wsrep_local_state'\". "
"Expected 2 columns. MySQL Version: %s", version_str);
return;
}
while ((row = mysql_fetch_row(result)))
{
if (strcmp(row[1], "4") == 0)
{
isjoined = 1;
}
/* Check if the node is a donor and is using xtrabackup, in this case it can stay alive */
else if (strcmp(row[1], "2") == 0 && handle->availableWhenDonor == 1)
{
if (mysql_query(database->con, "SHOW VARIABLES LIKE 'wsrep_sst_method'") == 0
&& (result2 = mysql_store_result(database->con)) != NULL)
{
if (mysql_field_count(database->con) < 2)
{
mysql_free_result(result);
mysql_free_result(result2);
MXS_ERROR("Unexpected result for \"SHOW VARIABLES LIKE "
"'wsrep_sst_method'\". Expected 2 columns."
" MySQL Version: %s", version_str);
return;
}
while ((row = mysql_fetch_row(result2)))
{
if (strncmp(row[1], "xtrabackup", 10) == 0)
{
isjoined = 1;
}
}
mysql_free_result(result2);
}
}
}
mysql_free_result(result);
}
if (isjoined)
{
/* Check the the Galera node index in the cluster */
if (mysql_query(database->con, "SHOW STATUS LIKE 'wsrep_local_index'") == 0
&& (result = mysql_store_result(database->con)) != NULL)
{
if (mysql_field_count(database->con) < 2)
{
mysql_free_result(result);
MXS_ERROR("Unexpected result for \"SHOW STATUS LIKE 'wsrep_local_index'\". "
"Expected 2 columns. MySQL Version: %s", version_str);
return;
}
while ((row = mysql_fetch_row(result)))
{
char* endchar;
long local_index = strtol(row[1], &endchar, 10);
if (*endchar != '\0' ||
(errno == ERANGE && (local_index == LONG_MAX || local_index == LONG_MIN)))
{
/** TODO: Create a mechanism to log warnings on a per server basis */
if (warn_erange_on_local_index)
{
MXS_WARNING("Invalid 'wsrep_local_index' on server '%s': %s",
database->server->unique_name, row[1]);
warn_erange_on_local_index = false;
}
local_index = -1;
}
database->server->node_id = local_index;
}
mysql_free_result(result);
}
server_set_status(&temp_server, SERVER_JOINED);
}
else
{
server_clear_status(&temp_server, SERVER_JOINED);
}
/* clear bits for non member nodes */
if (!SERVER_IN_MAINT(database->server) && (!SERVER_IS_JOINED(&temp_server)))
{
database->server->depth = -1;
/* clear M/S status */
server_clear_status(&temp_server, SERVER_SLAVE);
server_clear_status(&temp_server, SERVER_MASTER);
/* clear master sticky status */
server_clear_status(&temp_server, SERVER_MASTER_STICKINESS);
}
server_transfer_status(database->server, &temp_server);
}
/**
* The entry point for the monitoring module thread
*
* @param arg The handle of the monitor
*/
static void
monitorMain(void *arg)
{
MONITOR* mon = (MONITOR*) arg;
GALERA_MONITOR *handle;
MONITOR_SERVERS *ptr;
size_t nrounds = 0;
MONITOR_SERVERS *candidate_master = NULL;
int master_stickiness;
int is_cluster = 0;
int log_no_members = 1;
monitor_event_t evtype;
spinlock_acquire(&mon->lock);
handle = (GALERA_MONITOR *) mon->handle;
spinlock_release(&mon->lock);
master_stickiness = handle->disableMasterFailback;
if (mysql_thread_init())
{
MXS_ERROR("mysql_thread_init failed in monitor module. Exiting.");
return;
}
handle->status = MONITOR_RUNNING;
while (1)
{
if (handle->shutdown)
{
handle->status = MONITOR_STOPPING;
mysql_thread_end();
handle->status = MONITOR_STOPPED;
return;
}
/** Wait base interval */
thread_millisleep(MON_BASE_INTERVAL_MS);
/**
* Calculate how far away the monitor interval is from its full
* cycle and if monitor interval time further than the base
* interval, then skip monitoring checks. Excluding the first
* round.
*/
if (nrounds != 0 && ((nrounds * MON_BASE_INTERVAL_MS) % mon->interval) >= MON_BASE_INTERVAL_MS)
{
nrounds += 1;
continue;
}
nrounds += 1;
/* reset cluster members counter */
is_cluster = 0;
ptr = mon->databases;
while (ptr)
{
ptr->mon_prev_status = ptr->server->status;
monitorDatabase(mon, ptr);
/* Log server status change */
if (mon_status_changed(ptr))
{
MXS_DEBUG("Backend server %s:%d state : %s",
ptr->server->name,
ptr->server->port,
STRSRVSTATUS(ptr->server));
}
if (SERVER_IS_DOWN(ptr->server))
{
/** Increase this server'e error count */
ptr->mon_err_count += 1;
}
else
{
/** Reset this server's error count */
ptr->mon_err_count = 0;
}
ptr = ptr->next;
}
/*
* Let's select a master server:
* it could be the candidate master following MIN(node_id) rule or
* the server that was master in the previous monitor polling cycle
* Decision depends on master_stickiness value set in configuration
*/
/* get the candidate master, following MIN(node_id) rule */
candidate_master = get_candidate_master(mon);
/* Select the master, based on master_stickiness */
if (1 == handle->disableMasterRoleSetting)
{
handle->master = NULL;
}
else
{
handle->master = set_cluster_master(handle->master, candidate_master, master_stickiness);
}
ptr = mon->databases;
while (ptr)
{
const int repl_bits = (SERVER_SLAVE | SERVER_MASTER | SERVER_MASTER_STICKINESS);
if (SERVER_IS_JOINED(ptr->server))
{
if (handle->master)
{
if (ptr != handle->master)
{
/* set the Slave role and clear master stickiness */
server_clear_set_status(ptr->server, repl_bits, SERVER_SLAVE);
}
else
{
if (candidate_master &&
handle->master->server->node_id != candidate_master->server->node_id)
{
/* set master role and master stickiness */
server_clear_set_status(ptr->server, repl_bits,
(SERVER_MASTER | SERVER_MASTER_STICKINESS));
}
else
{
/* set master role and clear master stickiness */
server_clear_set_status(ptr->server, repl_bits, SERVER_MASTER);
}
}
}
is_cluster++;
}
else
{
server_clear_set_status(ptr->server, repl_bits, 0);
}
ptr = ptr->next;
}
if (is_cluster == 0 && log_no_members)
{
MXS_ERROR("There are no cluster members");
log_no_members = 0;
}
else
{
if (is_cluster > 0 && log_no_members == 0)
{
MXS_NOTICE("Found cluster members");
log_no_members = 1;
}
}
ptr = mon->databases;
while (ptr)
{
/** Execute monitor script if a server state has changed */
if (mon_status_changed(ptr))
{
evtype = mon_get_event_type(ptr);
if (isGaleraEvent(evtype))
{
mon_log_state_change(ptr);
if (handle->script && handle->events[evtype])
{
monitor_launch_script(mon, ptr, handle->script);
}
}
}
ptr = ptr->next;
}
mon_hangup_failed_servers(mon);
}
}
/**
* get candidate master from all nodes
*
* The current available rule: get the server with MIN(node_id)
* node_id comes from 'wsrep_local_index' variable
*
* @param servers The monitored servers list
* @return The candidate master on success, NULL on failure
*/
static MONITOR_SERVERS *get_candidate_master(MONITOR* mon)
{
MONITOR_SERVERS *moitor_servers = mon->databases;
MONITOR_SERVERS *candidate_master = NULL;
GALERA_MONITOR* handle = mon->handle;
long min_id = -1;
int minval = INT_MAX;
int currval;
char* value;
/* set min_id to the lowest value of moitor_servers->server->node_id */
while (moitor_servers)
{
if (!SERVER_IN_MAINT(moitor_servers->server) && SERVER_IS_JOINED(moitor_servers->server))
{
moitor_servers->server->depth = 0;
if (handle->use_priority && (value = serverGetParameter(moitor_servers->server, "priority")) != NULL)
{
currval = atoi(value);
if (currval < minval && currval > 0)
{
minval = currval;
candidate_master = moitor_servers;
}
}
else if (moitor_servers->server->node_id >= 0 &&
(!handle->use_priority || /** Server priority disabled*/
candidate_master == NULL || /** No candidate chosen */
serverGetParameter(candidate_master->server, "priority") == NULL)) /** Candidate has no priority */
{
if (min_id < 0 || moitor_servers->server->node_id < min_id)
{
min_id = moitor_servers->server->node_id;
candidate_master = moitor_servers;
}
}
}
moitor_servers = moitor_servers->next;
}
return candidate_master;
}
/**
* set the master server in the cluster
*
* master could be the last one from previous monitor cycle Iis running) or
* the candidate master.
* The selection is based on the configuration option mapped to master_stickiness
* The candidate master may change over time due to
* 'wsrep_local_index' value change in the Galera Cluster
* Enabling master_stickiness will avoid master change unless a failure is spotted
*
* @param current_master Previous master server
* @param candidate_master The candidate master server accordingly to the selection rule
* @return The master node pointer (could be NULL)
*/
static MONITOR_SERVERS *set_cluster_master(MONITOR_SERVERS *current_master, MONITOR_SERVERS *candidate_master,
int master_stickiness)
{
/*
* if current master is not set or master_stickiness is not enable
* just return candidate_master.
*/
if (current_master == NULL || master_stickiness == 0)
{
return candidate_master;
}
else
{
/*
* if current_master is still a cluster member use it
*
*/
if (SERVER_IS_JOINED(current_master->server) && (!SERVER_IN_MAINT(current_master->server)))
{
return current_master;
}
else
{
return candidate_master;
}
}
}
/**
* Disable/Enable the Master failback in a Galera Cluster.
*
* A restarted / rejoined node may get back the previous 'wsrep_local_index'
* from Cluster: if the value is the lowest in the cluster it will be selected as Master
* This will cause a Master change even if there is no failure.
* The option if set to 1 will avoid this situation, keeping the current Master (if running) available
*
* @param arg The handle allocated by startMonitor
* @param disable To disable it use 1, 0 keeps failback
*
* NOT USED
static void
disableMasterFailback(void *arg, int disable)
{
GALERA_MONITOR *handle = (GALERA_MONITOR *) arg;
memcpy(&handle->disableMasterFailback, &disable, sizeof(int));
}
*/
/**
* Allow a Galera node to be in sync when Donor.
*
* When enabled, the monitor will check if the node is using xtrabackup or xtrabackup-v2
* as SST method. In that case, node will stay as synced.
*
* @param arg The handle allocated by startMonitor
* @param disable To allow sync status use 1, 0 for traditional behavior
* NOT USED
static void
availableWhenDonor(void *arg, int disable)
{
GALERA_MONITOR *handle = (GALERA_MONITOR *) arg;
memcpy(&handle->availableWhenDonor, &disable, sizeof(int));
}
*/
static monitor_event_t galera_events[] =
{
MASTER_DOWN_EVENT,
MASTER_UP_EVENT,
SLAVE_DOWN_EVENT,
SLAVE_UP_EVENT,
SERVER_DOWN_EVENT,
SERVER_UP_EVENT,
SYNCED_DOWN_EVENT,
SYNCED_UP_EVENT,
DONOR_DOWN_EVENT,
DONOR_UP_EVENT,
LOST_MASTER_EVENT,
LOST_SLAVE_EVENT,
LOST_SYNCED_EVENT,
LOST_DONOR_EVENT,
NEW_MASTER_EVENT,
NEW_SLAVE_EVENT,
NEW_SYNCED_EVENT,
NEW_DONOR_EVENT,
MAX_MONITOR_EVENT
};
/**
* Check if the Galera monitor is monitoring this event type.
* @param event Event to check
* @return True if the event is monitored, false if it is not
* */
bool isGaleraEvent(monitor_event_t event)
{
int i;
for (i = 0; galera_events[i] != MAX_MONITOR_EVENT; i++)
{
if (event == galera_events[i])
{
return true;
}
}
return false;
}