hcq/MaxScale
hcq/MaxScale
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
MaxScale/server/core/monitor.cc
Johan Wikman faaf7f483e 2.4.13 Update Change Date
2020-10-14 09:15:46 +03:00

2247 lines
63 KiB
C++
Raw Blame History

/*
* 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/bsl11.
*
* Change Date: 2024-10-14
*
* 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 monitor.c - The monitor module management routines
*/
#include <maxscale/monitor.hh>
#include <atomic>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <chrono>
#include <string>
#include <sstream>
#include <set>
#include <zlib.h>
#include <sys/stat.h>
#include <vector>
#include <mutex>
#include <maxbase/alloc.h>
#include <maxscale/clock.h>
#include <maxscale/json_api.hh>
#include <maxscale/mariadb.hh>
#include <maxscale/mysql_utils.hh>
#include <maxscale/paths.h>
#include <maxscale/pcre2.h>
#include <maxscale/routingworker.hh>
#include <maxscale/secrets.h>
#include <maxscale/utils.hh>
#include <maxscale/json_api.hh>
#include <mysqld_error.h>
#include <maxbase/format.hh>
#include "internal/config.hh"
#include "internal/externcmd.hh"
#include "internal/maxscale.hh"
#include "internal/monitor.hh"
#include "internal/modules.hh"
#include "internal/server.hh"
#include "internal/service.hh"
/** Schema version, journals must have a matching version */
#define MMB_SCHEMA_VERSION 2
/** Constants for byte lengths of the values */
#define MMB_LEN_BYTES 4
#define MMB_LEN_SCHEMA_VERSION 1
#define MMB_LEN_CRC32 4
#define MMB_LEN_VALUE_TYPE 1
#define MMB_LEN_SERVER_STATUS 8
/** Type of the stored value */
enum stored_value_type
{
SVT_SERVER = 1, // Generic server state information
SVT_MASTER = 2, // The master server name
};
using std::string;
using std::set;
using Guard = std::lock_guard<std::mutex>;
using maxscale::Monitor;
using maxscale::MonitorServer;
using ConnectResult = maxscale::MonitorServer::ConnectResult;
const char CN_BACKEND_CONNECT_ATTEMPTS[] = "backend_connect_attempts";
const char CN_BACKEND_CONNECT_TIMEOUT[] = "backend_connect_timeout";
const char CN_BACKEND_READ_TIMEOUT[] = "backend_read_timeout";
const char CN_BACKEND_WRITE_TIMEOUT[] = "backend_write_timeout";
const char CN_DISK_SPACE_CHECK_INTERVAL[] = "disk_space_check_interval";
const char CN_EVENTS[] = "events";
const char CN_JOURNAL_MAX_AGE[] = "journal_max_age";
const char CN_MONITOR_INTERVAL[] = "monitor_interval";
const char CN_SCRIPT[] = "script";
const char CN_SCRIPT_TIMEOUT[] = "script_timeout";
namespace
{
class ThisUnit
{
public:
/**
* Mark a monitor as the monitor of the server. A server may only be monitored by one monitor.
*
* @param server Server to claim. The name is not checked to be a valid server name.
* @param new_owner Monitor which claims the server
* @param existing_owner If server is already monitored, the owning monitor name is written here
* @return True if success, false if server was claimed by another monitor
*/
bool claim_server(const string& server, const string& new_owner, string* existing_owner)
{
mxb_assert(Monitor::is_admin_thread());
bool claim_success = false;
auto iter = m_server_owners.find(server);
if (iter != m_server_owners.end())
{
// Server is already claimed by a monitor.
* existing_owner = iter->second;
}
else
{
m_server_owners[server] = new_owner;
claim_success = true;
}
return claim_success;
}
/**
* Mark a server as unmonitored.
*
* @param server The server name
*/
void release_server(const string& server)
{
mxb_assert(Monitor::is_admin_thread());
auto iter = m_server_owners.find(server);
mxb_assert(iter != m_server_owners.end());
m_server_owners.erase(iter);
}
string claimed_by(const string& server)
{
mxb_assert(Monitor::is_admin_thread());
string rval;
auto iter = m_server_owners.find(server);
if (iter != m_server_owners.end())
{
rval = iter->second;
}
return rval;
}
private:
// Global map of servername->monitorname. Not mutexed, as this should only be accessed
// from the admin thread.
std::map<string, string> m_server_owners;
};
ThisUnit this_unit;
/** Server type specific bits */
const uint64_t server_type_bits = SERVER_MASTER | SERVER_SLAVE | SERVER_JOINED;
/** All server bits */
const uint64_t all_server_bits = SERVER_RUNNING | SERVER_MAINT | SERVER_MASTER | SERVER_SLAVE
| SERVER_JOINED;
const char journal_name[] = "monitor.dat";
const char journal_template[] = "%s/%s/%s";
/**
* @brief Remove .tmp suffix and rename file
*
* @param src File to rename
* @return True if file was successfully renamed
*/
bool rename_tmp_file(Monitor* monitor, const char* src)
{
bool rval = true;
char dest[PATH_MAX + 1];
snprintf(dest, sizeof(dest), journal_template, get_datadir(), monitor->name(), journal_name);
if (rename(src, dest) == -1)
{
rval = false;
MXS_ERROR("Failed to rename journal file '%s' to '%s': %d, %s",
src,
dest,
errno,
mxs_strerror(errno));
}
return rval;
}
/**
* @brief Open temporary file
*
* @param monitor Monitor
* @param path Output where the path is stored
* @return Opened file or NULL on error
*/
FILE* open_tmp_file(Monitor* monitor, char* path)
{
int nbytes = snprintf(path, PATH_MAX, journal_template, get_datadir(), monitor->name(), "");
int max_bytes = PATH_MAX - (int)sizeof(journal_name);
FILE* rval = NULL;
if (nbytes < max_bytes && mxs_mkdir_all(path, 0744))
{
strcat(path, journal_name);
strcat(path, "XXXXXX");
int fd = mkstemp(path);
if (fd == -1)
{
MXS_ERROR("Failed to open file '%s': %d, %s", path, errno, mxs_strerror(errno));
}
else
{
rval = fdopen(fd, "w");
}
}
else
{
MXS_ERROR("Path is too long: %d characters exceeds the maximum path "
"length of %d bytes",
nbytes,
max_bytes);
}
return rval;
}
/**
* @brief Store server data to in-memory buffer
*
* @param monitor Monitor
* @param data Pointer to in-memory buffer used for storage, should be at least
* PATH_MAX bytes long
* @param size Size of @c data
*/
void store_data(Monitor* monitor, MonitorServer* master, uint8_t* data, uint32_t size)
{
uint8_t* ptr = data;
/** Store the data length */
mxb_assert(sizeof(size) == MMB_LEN_BYTES);
ptr = mxs_set_byte4(ptr, size);
/** Then the schema version */
*ptr++ = MMB_SCHEMA_VERSION;
/** Store the states of all servers */
for (MonitorServer* db : monitor->servers())
{
*ptr++ = (char)SVT_SERVER; // Value type
memcpy(ptr, db->server->name(), strlen(db->server->name()));// Name of the server
ptr += strlen(db->server->name());
*ptr++ = '\0'; // Null-terminate the string
auto status = db->server->status;
static_assert(sizeof(status) == MMB_LEN_SERVER_STATUS,
"Status size should be MMB_LEN_SERVER_STATUS bytes");
ptr = maxscale::set_byteN(ptr, status, MMB_LEN_SERVER_STATUS);
}
/** Store the current root master if we have one */
if (master)
{
*ptr++ = (char)SVT_MASTER;
memcpy(ptr, master->server->name(), strlen(master->server->name()));
ptr += strlen(master->server->name());
*ptr++ = '\0'; // Null-terminate the string
}
/** Calculate the CRC32 for the complete payload minus the CRC32 bytes */
uint32_t crc = crc32(0L, NULL, 0);
crc = crc32(crc, (uint8_t*)data + MMB_LEN_BYTES, size - MMB_LEN_CRC32);
mxb_assert(sizeof(crc) == MMB_LEN_CRC32);
ptr = mxs_set_byte4(ptr, crc);
mxb_assert(ptr - data == size + MMB_LEN_BYTES);
}
/**
* Check that memory area contains a null terminator
*/
static bool has_null_terminator(const char* data, const char* end)
{
while (data < end)
{
if (*data == '\0')
{
return true;
}
data++;
}
return false;
}
/**
* Process a generic server
*/
const char* process_server(Monitor* monitor, const char* data, const char* end)
{
for (MonitorServer* db : monitor->servers())
{
if (strcmp(db->server->name(), data) == 0)
{
const unsigned char* sptr = (unsigned char*)strchr(data, '\0');
mxb_assert(sptr);
sptr++;
uint64_t status = maxscale::get_byteN(sptr, MMB_LEN_SERVER_STATUS);
db->mon_prev_status = status;
db->server->set_status(status);
db->set_pending_status(status);
break;
}
}
data += strlen(data) + 1 + MMB_LEN_SERVER_STATUS;
return data;
}
/**
* Process a master
*/
const char* process_master(Monitor* monitor, MonitorServer** master,
const char* data, const char* end)
{
if (master)
{
for (MonitorServer* db : monitor->servers())
{
if (strcmp(db->server->name(), data) == 0)
{
*master = db;
break;
}
}
}
data += strlen(data) + 1;
return data;
}
/**
* Check that the calculated CRC32 matches the one stored on disk
*/
bool check_crc32(const uint8_t* data, uint32_t size, const uint8_t* crc_ptr)
{
uint32_t crc = mxs_get_byte4(crc_ptr);
uint32_t calculated_crc = crc32(0L, NULL, 0);
calculated_crc = crc32(calculated_crc, data, size);
return calculated_crc == crc;
}
/**
* Process the stored journal data
*/
bool process_data_file(Monitor* monitor, MonitorServer** master,
const char* data, const char* crc_ptr)
{
const char* ptr = data;
MXB_AT_DEBUG(const char* prevptr = ptr);
while (ptr < crc_ptr)
{
/** All values contain a null terminated string */
if (!has_null_terminator(ptr, crc_ptr))
{
MXS_ERROR("Possible corrupted journal file (no null terminator found). Ignoring.");
return false;
}
stored_value_type type = (stored_value_type)ptr[0];
ptr += MMB_LEN_VALUE_TYPE;
switch (type)
{
case SVT_SERVER:
ptr = process_server(monitor, ptr, crc_ptr);
break;
case SVT_MASTER:
ptr = process_master(monitor, master, ptr, crc_ptr);
break;
default:
MXS_ERROR("Possible corrupted journal file (unknown stored value). Ignoring.");
return false;
}
mxb_assert(prevptr != ptr);
MXB_AT_DEBUG(prevptr = ptr);
}
mxb_assert(ptr == crc_ptr);
return true;
}
bool check_disk_space_exhausted(MonitorServer* pMs,
const std::string& path,
const maxscale::disk::SizesAndName& san,
int32_t max_percentage)
{
bool disk_space_exhausted = false;
int32_t used_percentage = ((san.total() - san.available()) / (double)san.total()) * 100;
if (used_percentage >= max_percentage)
{
MXS_ERROR("Disk space on %s at %s is exhausted; %d%% of the the disk "
"mounted on the path %s has been used, and the limit it %d%%.",
pMs->server->name(),
pMs->server->address,
used_percentage,
path.c_str(),
max_percentage);
disk_space_exhausted = true;
}
return disk_space_exhausted;
}
const char ERR_CANNOT_MODIFY[] =
"The server is monitored, so only the maintenance status can be "
"set/cleared manually. Status was not modified.";
const char WRN_REQUEST_OVERWRITTEN[] =
"Previous maintenance request was not yet read by the monitor and was overwritten.";
}
namespace maxscale
{
Monitor::Monitor(const string& name, const string& module)
: m_name(name)
, m_module(module)
{
memset(m_journal_hash, 0, sizeof(m_journal_hash));
}
void Monitor::stop()
{
do_stop();
for (auto db : m_servers)
{
// TODO: Should be db->close().
mysql_close(db->con);
db->con = NULL;
}
}
const char* Monitor::name() const
{
return m_name.c_str();
}
using std::chrono::milliseconds;
using std::chrono::seconds;
bool Monitor::configure(const MXS_CONFIG_PARAMETER* params)
{
m_settings.interval = params->get_duration<milliseconds>(CN_MONITOR_INTERVAL).count();
m_settings.journal_max_age = params->get_duration<seconds>(CN_JOURNAL_MAX_AGE).count();
m_settings.events = params->get_enum(CN_EVENTS, mxs_monitor_event_enum_values);
MonitorServer::ConnectionSettings& conn_settings = m_settings.conn_settings;
conn_settings.read_timeout = params->get_duration<seconds>(CN_BACKEND_READ_TIMEOUT).count();
conn_settings.write_timeout = params->get_duration<seconds>(CN_BACKEND_WRITE_TIMEOUT).count();
conn_settings.connect_timeout = params->get_duration<seconds>(CN_BACKEND_CONNECT_TIMEOUT).count();
conn_settings.connect_attempts = params->get_integer(CN_BACKEND_CONNECT_ATTEMPTS);
conn_settings.username = params->get_string(CN_USER);
conn_settings.password = params->get_string(CN_PASSWORD);
// Disk check interval is given in ms, duration is constructed from seconds.
auto dsc_interval = params->get_duration<milliseconds>(CN_DISK_SPACE_CHECK_INTERVAL).count();
// 0 implies disabling -> save negative value to interval.
m_settings.disk_space_check_interval = (dsc_interval > 0) ?
mxb::Duration(static_cast<double>(dsc_interval) / 1000) : mxb::Duration(-1);
// The monitor serverlist has already been checked to be valid. Empty value is ok too.
// First, remove all servers.
remove_all_servers();
auto servers_temp = params->get_server_list(CN_SERVERS);
bool error = false;
for (auto elem : servers_temp)
{
if (!add_server(elem))
{
error = true;
}
}
/* The previous config values were normal types and were checked by the config manager
* to be correct. The following is a complicated type and needs to be checked separately. */
auto threshold_string = params->get_string(CN_DISK_SPACE_THRESHOLD);
if (!set_disk_space_threshold(threshold_string))
{
MXS_ERROR("Invalid value for '%s' for monitor %s: %s",
CN_DISK_SPACE_THRESHOLD, name(), threshold_string.c_str());
error = true;
}
m_settings.script_timeout = params->get_duration<seconds>(CN_SCRIPT_TIMEOUT).count();
m_settings.script = params->get_string(CN_SCRIPT);
if (m_settings.script.empty())
{
// Reset current external cmd if any.
m_scriptcmd.reset();
}
else
{
m_scriptcmd = ExternalCmd::create(m_settings.script, m_settings.script_timeout);
if (!m_scriptcmd)
{
MXS_ERROR("Failed to initialize script '%s'.", m_settings.script.c_str());
error = true;
}
}
if (!error)
{
// Store module name into parameter storage.
m_parameters.set(CN_MODULE, m_module);
// Add all config settings to text-mode storage. Needed for serialization.
m_parameters.set_multiple(*params);
}
return !error;
}
const MXS_CONFIG_PARAMETER& Monitor::parameters() const
{
return m_parameters;
}
const Monitor::Settings& Monitor::settings() const
{
return m_settings;
}
long Monitor::ticks() const
{
return m_ticks.load(std::memory_order_acquire);
}
const char* Monitor::state_string() const
{
return is_running() ? "Running" : "Stopped";
}
Monitor::~Monitor()
{
for (auto server : m_servers)
{
// TODO: store unique pointers in the array
delete server;
}
m_servers.clear();
}
/**
* Add a server to the monitor. Fails if server is already monitored.
*
* @param server A server
* @return True if server was added
*/
bool Monitor::add_server(SERVER* server)
{
// This should only be called from the admin thread while the monitor is stopped.
mxb_assert(!is_running() && is_admin_thread());
bool success = false;
string existing_owner;
if (this_unit.claim_server(server->name(), m_name, &existing_owner))
{
auto new_server = new MonitorServer(server, m_settings.disk_space_limits);
m_servers.push_back(new_server);
server_added(server);
success = true;
}
else
{
MXS_ERROR("Server '%s' is already monitored by '%s', cannot add it to another monitor.",
server->name(), existing_owner.c_str());
}
return success;
}
void Monitor::server_added(SERVER* server)
{
service_add_server(this, server);
}
void Monitor::server_removed(SERVER* server)
{
service_remove_server(this, server);
}
/**
* Remove all servers from the monitor.
*/
void Monitor::remove_all_servers()
{
// This should only be called from the admin thread while the monitor is stopped.
mxb_assert(!is_running() && is_admin_thread());
for (auto mon_server : m_servers)
{
mxb_assert(this_unit.claimed_by(mon_server->server->name()) == m_name);
this_unit.release_server(mon_server->server->name());
server_removed(mon_server->server);
delete mon_server;
}
m_servers.clear();
}
void Monitor::show(DCB* dcb)
{
dcb_printf(dcb, "Name: %s\n", name());
dcb_printf(dcb, "State: %s\n", state_string());
dcb_printf(dcb, "Times monitored: %li\n", ticks());
dcb_printf(dcb, "Sampling interval: %lu milliseconds\n", m_settings.interval);
dcb_printf(dcb, "Connect Timeout: %i seconds\n", m_settings.conn_settings.connect_timeout);
dcb_printf(dcb, "Read Timeout: %i seconds\n", m_settings.conn_settings.read_timeout);
dcb_printf(dcb, "Write Timeout: %i seconds\n", m_settings.conn_settings.write_timeout);
dcb_printf(dcb, "Connect attempts: %i \n", m_settings.conn_settings.connect_attempts);
dcb_printf(dcb, "Monitored servers: ");
const char* sep = "";
for (const auto& db : m_servers)
{
dcb_printf(dcb, "%s[%s]:%d", sep, db->server->address, db->server->port);
sep = ", ";
}
dcb_printf(dcb, "\n");
if (is_running())
{
diagnostics(dcb);
}
else
{
dcb_printf(dcb, " (no diagnostics)\n");
}
dcb_printf(dcb, "\n");
}
json_t* Monitor::to_json(const char* host) const
{
// This function mostly reads settings-type data, which is only written to by the admin thread,
// The rest is safe to read without mutexes.
mxb_assert(Monitor::is_admin_thread());
json_t* rval = json_object();
json_t* attr = json_object();
json_t* rel = json_object();
auto my_name = name();
json_object_set_new(rval, CN_ID, json_string(my_name));
json_object_set_new(rval, CN_TYPE, json_string(CN_MONITORS));
json_object_set_new(attr, CN_MODULE, json_string(m_module.c_str()));
json_object_set_new(attr, CN_STATE, json_string(state_string()));
json_object_set_new(attr, CN_TICKS, json_integer(ticks()));
/** Monitor parameters */
json_object_set_new(attr, CN_PARAMETERS, parameters_to_json());
if (is_running())
{
json_t* diag = diagnostics_json();
if (diag)
{
json_object_set_new(attr, CN_MONITOR_DIAGNOSTICS, diag);
}
}
if (!m_servers.empty())
{
json_t* mon_rel = mxs_json_relationship(host, MXS_JSON_API_SERVERS);
for (MonitorServer* db : m_servers)
{
mxs_json_add_relation(mon_rel, db->server->name(), CN_SERVERS);
}
json_object_set_new(rel, CN_SERVERS, mon_rel);
}
json_object_set_new(rval, CN_RELATIONSHIPS, rel);
json_object_set_new(rval, CN_ATTRIBUTES, attr);
json_object_set_new(rval, CN_LINKS, mxs_json_self_link(host, CN_MONITORS, my_name));
return rval;
}
json_t* Monitor::parameters_to_json() const
{
json_t* rval = json_object();
const MXS_MODULE* mod = get_module(m_module.c_str(), MODULE_MONITOR);
auto my_config = parameters();
config_add_module_params_json(&my_config,
{CN_TYPE, CN_MODULE, CN_SERVERS},
config_monitor_params,
mod->parameters,
rval);
return rval;
}
bool Monitor::test_permissions(const string& query)
{
if (m_servers.empty() || config_get_global_options()->skip_permission_checks)
{
return true;
}
char* dpasswd = decrypt_password(m_settings.conn_settings.password.c_str());
bool rval = false;
for (MonitorServer* mondb : m_servers)
{
auto result = mondb->ping_or_connect(m_settings.conn_settings);
if (!connection_is_ok(result))
{
MXS_ERROR("[%s] Failed to connect to server '%s' ([%s]:%d) when"
" checking monitor user credentials and permissions.",
name(),
mondb->server->name(),
mondb->server->address,
mondb->server->port);
if (result != ConnectResult::ACCESS_DENIED)
{
rval = true;
}
}
else if (mxs_mysql_query(mondb->con, query.c_str()) != 0)
{
switch (mysql_errno(mondb->con))
{
case ER_TABLEACCESS_DENIED_ERROR:
case ER_COLUMNACCESS_DENIED_ERROR:
case ER_SPECIFIC_ACCESS_DENIED_ERROR:
case ER_PROCACCESS_DENIED_ERROR:
case ER_KILL_DENIED_ERROR:
rval = false;
break;
default:
rval = true;
break;
}
MXS_ERROR("[%s] Failed to execute query '%s' with user '%s'. MySQL error message: %s",
name(), query.c_str(), m_settings.conn_settings.username.c_str(),
mysql_error(mondb->con));
}
else
{
rval = true;
MYSQL_RES* res = mysql_use_result(mondb->con);
if (res == NULL)
{
MXS_ERROR("[%s] Result retrieval failed when checking monitor permissions: %s",
name(),
mysql_error(mondb->con));
}
else
{
mysql_free_result(res);
}
}
}
MXS_FREE(dpasswd);
return rval;
}
void MonitorServer::stash_current_status()
{
mon_prev_status = server->status;
pending_status = server->status;
}
void MonitorServer::set_pending_status(uint64_t bits)
{
pending_status |= bits;
}
void MonitorServer::clear_pending_status(uint64_t bits)
{
pending_status &= ~bits;
}
/*
* Determine a monitor event, defined by the difference between the old
* status of a server and the new status.
*
* @return monitor_event_t A monitor event (enum)
*
* @note This function must only be called from mon_process_state_changes
*/
mxs_monitor_event_t MonitorServer::get_event_type() const
{
typedef enum
{
DOWN_EVENT,
UP_EVENT,
LOSS_EVENT,
NEW_EVENT,
UNSUPPORTED_EVENT
} general_event_type;
general_event_type event_type = UNSUPPORTED_EVENT;
uint64_t prev = mon_prev_status & all_server_bits;
uint64_t present = server->status & all_server_bits;
if (prev == present)
{
/* This should never happen */
mxb_assert(false);
return UNDEFINED_EVENT;
}
if ((prev & SERVER_RUNNING) == 0)
{
/* The server was not running previously */
if ((present & SERVER_RUNNING) != 0)
{
event_type = UP_EVENT;
}
else
{
/* Otherwise, was not running and still is not running. This should never happen. */
mxb_assert(false);
}
}
else
{
/* Previous state must have been running */
if ((present & SERVER_RUNNING) == 0)
{
event_type = DOWN_EVENT;
}
else
{
/** These are used to detect whether we actually lost something or
* just transitioned from one state to another */
uint64_t prev_bits = prev & (SERVER_MASTER | SERVER_SLAVE);
uint64_t present_bits = present & (SERVER_MASTER | SERVER_SLAVE);
/* Was running and still is */
if ((!prev_bits || !present_bits || prev_bits == present_bits)
&& (prev & server_type_bits))
{
/* We used to know what kind of server it was */
event_type = LOSS_EVENT;
}
else
{
/* We didn't know what kind of server it was, now we do*/
event_type = NEW_EVENT;
}
}
}
mxs_monitor_event_t rval = UNDEFINED_EVENT;
switch (event_type)
{
case UP_EVENT:
rval = (present & SERVER_MASTER) ? MASTER_UP_EVENT :
(present & SERVER_SLAVE) ? SLAVE_UP_EVENT :
(present & SERVER_JOINED) ? SYNCED_UP_EVENT :
SERVER_UP_EVENT;
break;
case DOWN_EVENT:
rval = (prev & SERVER_MASTER) ? MASTER_DOWN_EVENT :
(prev & SERVER_SLAVE) ? SLAVE_DOWN_EVENT :
(prev & SERVER_JOINED) ? SYNCED_DOWN_EVENT :
SERVER_DOWN_EVENT;
break;
case LOSS_EVENT:
rval = (prev & SERVER_MASTER) ? LOST_MASTER_EVENT :
(prev & SERVER_SLAVE) ? LOST_SLAVE_EVENT :
(prev & SERVER_JOINED) ? LOST_SYNCED_EVENT :
UNDEFINED_EVENT;
break;
case NEW_EVENT:
rval = (present & SERVER_MASTER) ? NEW_MASTER_EVENT :
(present & SERVER_SLAVE) ? NEW_SLAVE_EVENT :
(present & SERVER_JOINED) ? NEW_SYNCED_EVENT :
UNDEFINED_EVENT;
break;
default:
/* This should never happen */
mxb_assert(false);
break;
}
mxb_assert(rval != UNDEFINED_EVENT);
return rval;
}
const char* Monitor::get_event_name(mxs_monitor_event_t event)
{
for (int i = 0; mxs_monitor_event_enum_values[i].name; i++)
{
if (mxs_monitor_event_enum_values[i].enum_value == event)
{
return mxs_monitor_event_enum_values[i].name;
}
}
mxb_assert(!true);
return "undefined_event";
}
const char* MonitorServer::get_event_name()
{
return Monitor::get_event_name((mxs_monitor_event_t) server->last_event);
}
string Monitor::gen_serverlist(int status, CredentialsApproach approach)
{
string rval;
rval.reserve(100 * m_servers.size());
string separator;
for (auto mon_server : m_servers)
{
auto server = static_cast<Server*>(mon_server->server);
if (status == 0 || server->status & status)
{
if (approach == CredentialsApproach::EXCLUDE)
{
rval += separator + mxb::string_printf("[%s]:%i", server->address, server->port);
}
else
{
string user = m_settings.conn_settings.username;
string password = m_settings.conn_settings.password;
string server_specific_monuser = server->monitor_user();
if (!server_specific_monuser.empty())
{
user = server_specific_monuser;
password = server->monitor_password();
}
rval += separator + mxb::string_printf("%s:%s@[%s]:%d", user.c_str(), password.c_str(),
server->address, server->port);
}
separator = ",";
}
}
return rval;
}
/**
* Check if current monitored server status has changed.
*
* @return true if status has changed
*/
bool MonitorServer::status_changed()
{
bool rval = false;
/* Previous status is -1 if not yet set */
if (mon_prev_status != static_cast<uint64_t>(-1))
{
uint64_t old_status = mon_prev_status & all_server_bits;
uint64_t new_status = server->status & all_server_bits;
/**
* The state has changed if the relevant state bits are not the same,
* the server is either running, stopping or starting and the server is
* not going into maintenance or coming out of it
*/
if (old_status != new_status
&& ((old_status | new_status) & SERVER_MAINT) == 0
&& ((old_status | new_status) & SERVER_RUNNING) == SERVER_RUNNING)
{
rval = true;
}
}
return rval;
}
/**
* Check if current monitored server has a loggable failure status.
*
* @return true if failed status can be logged or false
*/
bool MonitorServer::should_print_fail_status()
{
return server->is_down() && mon_err_count == 0;
}
MonitorServer* Monitor::find_parent_node(MonitorServer* target)
{
MonitorServer* rval = NULL;
if (target->server->master_id > 0)
{
for (MonitorServer* node : m_servers)
{
if (node->server->node_id == target->server->master_id)
{
rval = node;
break;
}
}
}
return rval;
}
std::string Monitor::child_nodes(MonitorServer* parent)
{
std::stringstream ss;
if (parent->server->node_id > 0)
{
bool have_content = false;
for (MonitorServer* node : m_servers)
{
if (node->server->master_id == parent->server->node_id)
{
if (have_content)
{
ss << ",";
}
ss << "[" << node->server->address << "]:" << node->server->port;
have_content = true;
}
}
}
return ss.str();
}
int Monitor::launch_command(MonitorServer* ptr)
{
m_scriptcmd->reset_substituted();
// A generator function is ran only if the matching substitution keyword is found.
auto gen_initiator = [ptr] {
return mxb::string_printf("[%s]:%d", ptr->server->address, ptr->server->port);
};
auto gen_parent = [this, ptr] {
string ss;
MonitorServer* parent = find_parent_node(ptr);
if (parent)
{
ss = mxb::string_printf("[%s]:%d", parent->server->address, parent->server->port);
}
return ss;
};
m_scriptcmd->match_substitute("$INITIATOR", gen_initiator);
m_scriptcmd->match_substitute("$PARENT", gen_parent);
m_scriptcmd->match_substitute("$CHILDREN", [this, ptr] {
return child_nodes(ptr);
});
m_scriptcmd->match_substitute("$EVENT", [ptr] {
return ptr->get_event_name();
});
m_scriptcmd->match_substitute("$CREDENTIALS", [this] {
// Provides credentials for all servers.
return gen_serverlist(0, CredentialsApproach::INCLUDE);
});
m_scriptcmd->match_substitute("$NODELIST", [this] {
return gen_serverlist(SERVER_RUNNING);
});
m_scriptcmd->match_substitute("$LIST", [this] {
return gen_serverlist(0);
});
m_scriptcmd->match_substitute("$MASTERLIST", [this] {
return gen_serverlist(SERVER_MASTER);
});
m_scriptcmd->match_substitute("$SLAVELIST", [this] {
return gen_serverlist(SERVER_SLAVE);
});
m_scriptcmd->match_substitute("$SYNCEDLIST", [this] {
return gen_serverlist(SERVER_JOINED);
});
int rv = m_scriptcmd->externcmd_execute();
if (rv == 0)
{
MXS_NOTICE("Executed monitor script on event '%s'. Script was: '%s'",
ptr->get_event_name(), m_scriptcmd->substituted());
}
else if (rv == -1)
{
// Internal error
MXS_ERROR("Failed to execute script on server state change event '%s'. Script was: '%s'",
ptr->get_event_name(), m_scriptcmd->substituted());
}
else
{
// Script returned a non-zero value
MXS_ERROR("Script returned %d on event '%s'. Script was: '%s'",
rv, ptr->get_event_name(), m_scriptcmd->substituted());
}
return rv;
}
MonitorServer::ConnectResult
Monitor::ping_or_connect_to_db(const MonitorServer::ConnectionSettings& sett, SERVER& server, MYSQL** ppConn)
{
mxb_assert(ppConn);
auto pConn = *ppConn;
if (pConn)
{
/** Return if the connection is OK */
if (mysql_ping(pConn) == 0)
{
return ConnectResult::OLDCONN_OK;
}
/** Otherwise close the handle. */
mysql_close(pConn);
pConn = nullptr;
}
ConnectResult conn_result = ConnectResult::REFUSED;
string uname = sett.username;
string passwd = sett.password;
const Server& srv = static_cast<const Server&>(server); // Clean this up later.
string server_specific_monuser = srv.monitor_user();
if (!server_specific_monuser.empty())
{
uname = server_specific_monuser;
passwd = srv.monitor_password();
}
char* dpwd = decrypt_password(passwd.c_str());
for (int i = 0; i < sett.connect_attempts; i++)
{
pConn = mysql_init(NULL);
mysql_optionsv(pConn, MYSQL_OPT_CONNECT_TIMEOUT, &sett.connect_timeout);
mysql_optionsv(pConn, MYSQL_OPT_READ_TIMEOUT, &sett.read_timeout);
mysql_optionsv(pConn, MYSQL_OPT_WRITE_TIMEOUT, &sett.write_timeout);
mysql_optionsv(pConn, MYSQL_PLUGIN_DIR, get_connector_plugindir());
time_t start = time(NULL);
if (mxs_mysql_real_connect(pConn, &server, uname.c_str(), dpwd))
{
conn_result = ConnectResult::NEWCONN_OK;
break;
}
else if (conn_result == ConnectResult::REFUSED
&& difftime(time(NULL), start) >= sett.connect_timeout)
{
conn_result = ConnectResult::TIMEOUT;
}
auto err = mysql_errno(pConn);
mysql_close(pConn);
pConn = nullptr;
if (err == ER_ACCESS_DENIED_ERROR || err == ER_ACCESS_DENIED_NO_PASSWORD_ERROR)
{
conn_result = ConnectResult::ACCESS_DENIED;
}
}
MXS_FREE(dpwd);
*ppConn = pConn;
return conn_result;
}
ConnectResult MonitorServer::ping_or_connect(const ConnectionSettings& settings)
{
return Monitor::ping_or_connect_to_db(settings, *server, &con);
}
/**
* Is the return value one of the 'OK' values.
*
* @param connect_result Return value of mon_ping_or_connect_to_db
* @return True of connection is ok
*/
bool Monitor::connection_is_ok(ConnectResult connect_result)
{
return connect_result == ConnectResult::OLDCONN_OK || connect_result == ConnectResult::NEWCONN_OK;
}
string Monitor::get_server_monitor(const SERVER* server)
{
return this_unit.claimed_by(server->name());
}
bool Monitor::is_admin_thread()
{
return running_in_admin_thread();
}
/**
* Log an error about the failure to connect to a backend server and why it happened.
*
* @param rval Return value of mon_ping_or_connect_to_db
*/
void MonitorServer::log_connect_error(ConnectResult rval)
{
mxb_assert(!Monitor::connection_is_ok(rval));
const char TIMED_OUT[] = "Monitor timed out when connecting to server %s[%s:%d] : '%s'";
const char REFUSED[] = "Monitor was unable to connect to server %s[%s:%d] : '%s'";
MXS_ERROR(rval == ConnectResult::TIMEOUT ? TIMED_OUT : REFUSED,
server->name(),
server->address,
server->port,
mysql_error(con));
}
void MonitorServer::log_state_change()
{
string prev = SERVER::status_to_string(mon_prev_status);
string next = server->status_string();
MXS_NOTICE("Server changed state: %s[%s:%u]: %s. [%s] -> [%s]",
server->name(), server->address, server->port,
get_event_name(),
prev.c_str(), next.c_str());
}
void Monitor::hangup_failed_servers()
{
for (MonitorServer* ptr : m_servers)
{
if (ptr->status_changed() && (!(ptr->server->is_usable()) || !(ptr->server->is_in_cluster())))
{
dcb_hangup_foreach(ptr->server);
}
}
}
void MonitorServer::mon_report_query_error()
{
MXS_ERROR("Failed to execute query on server '%s' ([%s]:%d): %s",
server->name(),
server->address,
server->port,
mysql_error(con));
}
/**
* Check if admin is requesting setting or clearing maintenance status on the server and act accordingly.
* Should be called at the beginning of a monitor loop.
*/
void Monitor::check_maintenance_requests()
{
/* In theory, the admin may be modifying the server maintenance status during this function. The overall
* maintenance flag should be read-written atomically to prevent missing a value. */
bool was_pending = m_status_change_pending.exchange(false, std::memory_order_acq_rel);
if (was_pending)
{
for (auto ptr : m_servers)
{
// The admin can only modify the [Maintenance] and [Drain] bits.
int admin_msg = atomic_exchange_int(&ptr->status_request, MonitorServer::NO_CHANGE);
switch (admin_msg)
{
case MonitorServer::MAINT_ON:
ptr->server->set_status(SERVER_MAINT);
break;
case MonitorServer::MAINT_OFF:
ptr->server->clear_status(SERVER_MAINT);
break;
case MonitorServer::BEING_DRAINED_ON:
ptr->server->set_status(SERVER_DRAINING);
break;
case MonitorServer::BEING_DRAINED_OFF:
ptr->server->clear_status(SERVER_DRAINING);
break;
case MonitorServer::NO_CHANGE:
break;
default:
mxb_assert(!true);
}
}
}
}
void Monitor::detect_handle_state_changes()
{
bool master_down = false;
bool master_up = false;
for (MonitorServer* ptr : m_servers)
{
if (ptr->status_changed())
{
/**
* The last executed event will be needed if a passive MaxScale is
* promoted to an active one and the last event that occurred on
* a server was a master_down event.
*
* In this case, a failover script should be called if no master_up
* or new_master events are triggered within a pre-defined time limit.
*/
mxs_monitor_event_t event = ptr->get_event_type();
ptr->server->last_event = event;
ptr->server->triggered_at = mxs_clock();
ptr->log_state_change();
if (event == MASTER_DOWN_EVENT)
{
master_down = true;
}
else if (event == MASTER_UP_EVENT || event == NEW_MASTER_EVENT)
{
master_up = true;
}
if (m_scriptcmd && (event & m_settings.events))
{
launch_command(ptr);
}
}
}
if (master_down && master_up)
{
MXS_NOTICE("Master switch detected: lost a master and gained a new one");
}
}
int Monitor::get_data_file_path(char* path) const
{
int rv = snprintf(path, PATH_MAX, journal_template, get_datadir(), name(), journal_name);
return rv;
}
/**
* @brief Open stored journal file
*
* @param monitor Monitor to reload
* @param path Output where path is stored
* @return Opened file or NULL on error
*/
FILE* Monitor::open_data_file(Monitor* monitor, char* path)
{
FILE* rval = NULL;
int nbytes = monitor->get_data_file_path(path);
if (nbytes < PATH_MAX)
{
if ((rval = fopen(path, "rb")) == NULL && errno != ENOENT)
{
MXS_ERROR("Failed to open journal file: %d, %s", errno, mxs_strerror(errno));
}
}
else
{
MXS_ERROR("Path is too long: %d characters exceeds the maximum path "
"length of %d bytes",
nbytes,
PATH_MAX);
}
return rval;
}
void Monitor::store_server_journal(MonitorServer* master)
{
auto monitor = this; // TODO: cleanup later
/** Calculate how much memory we need to allocate */
uint32_t size = MMB_LEN_SCHEMA_VERSION + MMB_LEN_CRC32;
for (MonitorServer* db : m_servers)
{
/** Each server is stored as a type byte and a null-terminated string
* followed by eight byte server status. */
size += MMB_LEN_VALUE_TYPE + strlen(db->server->name()) + 1 + MMB_LEN_SERVER_STATUS;
}
if (master)
{
/** The master server name is stored as a null terminated string */
size += MMB_LEN_VALUE_TYPE + strlen(master->server->name()) + 1;
}
/** 4 bytes for file length, 1 byte for schema version and 4 bytes for CRC32 */
uint32_t buffer_size = size + MMB_LEN_BYTES;
uint8_t* data = (uint8_t*)MXS_MALLOC(buffer_size);
char path[PATH_MAX + 1];
if (data)
{
/** Store the data in memory first and compare the current hash to
* the hash of the last stored journal. This isn't a fool-proof
* method of detecting changes but any failures are mainly of
* theoretical nature. */
store_data(monitor, master, data, size);
uint8_t hash[SHA_DIGEST_LENGTH];
SHA1(data, size, hash);
if (memcmp(monitor->m_journal_hash, hash, sizeof(hash)) != 0)
{
FILE* file = open_tmp_file(monitor, path);
if (file)
{
/** Write the data to a temp file and rename it to the final name */
if (fwrite(data, 1, buffer_size, file) == buffer_size && fflush(file) == 0)
{
if (!rename_tmp_file(monitor, path))
{
unlink(path);
}
else
{
memcpy(monitor->m_journal_hash, hash, sizeof(hash));
}
}
else
{
MXS_ERROR("Failed to write journal data to disk: %d, %s",
errno,
mxs_strerror(errno));
}
fclose(file);
}
}
}
MXS_FREE(data);
}
void Monitor::load_server_journal(MonitorServer** master)
{
auto monitor = this; // TODO: cleanup later
char path[PATH_MAX];
FILE* file = open_data_file(monitor, path);
if (file)
{
uint32_t size = 0;
size_t bytes = fread(&size, 1, MMB_LEN_BYTES, file);
mxb_assert(sizeof(size) == MMB_LEN_BYTES);
if (bytes == MMB_LEN_BYTES)
{
/** Payload contents:
*
* - One byte of schema version
* - `size - 5` bytes of data
* - Trailing 4 bytes of CRC32
*/
char* data = (char*)MXS_MALLOC(size);
if (data && (bytes = fread(data, 1, size, file)) == size)
{
if (*data == MMB_SCHEMA_VERSION)
{
if (check_crc32((uint8_t*)data,
size - MMB_LEN_CRC32,
(uint8_t*)data + size - MMB_LEN_CRC32))
{
if (process_data_file(monitor,
master,
data + MMB_LEN_SCHEMA_VERSION,
data + size - MMB_LEN_CRC32))
{
MXS_NOTICE("Loaded server states from journal file: %s", path);
}
}
else
{
MXS_ERROR("CRC32 mismatch in journal file. Ignoring.");
}
}
else
{
MXS_ERROR("Unknown journal schema version: %d", (int)*data);
}
}
else if (data)
{
if (ferror(file))
{
MXS_ERROR("Failed to read journal file: %d, %s", errno, mxs_strerror(errno));
}
else
{
MXS_ERROR("Failed to read journal file: Expected %u bytes, "
"read %lu bytes.",
size,
bytes);
}
}
MXS_FREE(data);
}
else
{
if (ferror(file))
{
MXS_ERROR("Failed to read journal file length: %d, %s",
errno,
mxs_strerror(errno));
}
else
{
MXS_ERROR("Failed to read journal file length: Expected %d bytes, "
"read %lu bytes.",
MMB_LEN_BYTES,
bytes);
}
}
fclose(file);
}
}
void Monitor::remove_server_journal()
{
char path[PATH_MAX];
if (get_data_file_path(path) < PATH_MAX)
{
unlink(path);
}
else
{
MXS_ERROR("Path to monitor journal directory is too long.");
}
}
bool Monitor::journal_is_stale() const
{
bool is_stale = true;
char path[PATH_MAX];
auto max_age = m_settings.journal_max_age;
if (get_data_file_path(path) < PATH_MAX)
{
struct stat st;
if (stat(path, &st) == 0)
{
time_t tdiff = time(NULL) - st.st_mtim.tv_sec;
if (tdiff >= max_age)
{
MXS_NOTICE("Journal file was created %ld seconds ago. Maximum journal "
"age is %ld seconds.", tdiff, max_age);
}
else
{
is_stale = false;
}
}
else if (errno != ENOENT)
{
MXS_ERROR("Failed to inspect journal file: %d, %s", errno, mxs_strerror(errno));
}
}
else
{
MXS_ERROR("Path to monitor journal directory is too long.");
}
return is_stale;
}
MonitorServer* Monitor::get_monitored_server(SERVER* search_server)
{
mxb_assert(search_server);
for (const auto iter : m_servers)
{
if (iter->server == search_server)
{
return iter;
}
}
return nullptr;
}
std::vector<MonitorServer*> Monitor::get_monitored_serverlist(const string& key, bool* error_out)
{
std::vector<MonitorServer*> monitored_array;
// Check that value exists.
if (!m_parameters.contains(key))
{
return monitored_array;
}
string name_error;
auto servers = m_parameters.get_server_list(key, &name_error);
if (!servers.empty())
{
// All servers in the array must be monitored by the given monitor.
for (auto elem : servers)
{
MonitorServer* mon_serv = get_monitored_server(elem);
if (mon_serv)
{
monitored_array.push_back(mon_serv);
}
else
{
MXS_ERROR("Server '%s' is not monitored by monitor '%s'.", elem->name(), name());
*error_out = true;
}
}
if (monitored_array.size() < servers.size())
{
monitored_array.clear();
}
}
else
{
MXS_ERROR("Serverlist setting '%s' contains invalid server name '%s'.",
key.c_str(), name_error.c_str());
*error_out = true;
}
return monitored_array;
}
bool Monitor::set_disk_space_threshold(const string& dst_setting)
{
mxb_assert(!is_running());
SERVER::DiskSpaceLimits new_dst;
bool rv = config_parse_disk_space_threshold(&new_dst, dst_setting.c_str());
if (rv)
{
m_settings.disk_space_limits = new_dst;
}
return rv;
}
bool Monitor::set_server_status(SERVER* srv, int bit, string* errmsg_out)
{
MonitorServer* msrv = get_monitored_server(srv);
mxb_assert(msrv);
if (!msrv)
{
MXS_ERROR("Monitor %s requested to set status of server %s that it does not monitor.",
name(), srv->address);
return false;
}
bool written = false;
if (is_running())
{
/* This server is monitored, in which case modifying any other status bit than Maintenance is
* disallowed. */
if (bit & ~(SERVER_MAINT | SERVER_DRAINING))
{
MXS_ERROR(ERR_CANNOT_MODIFY);
if (errmsg_out)
{
*errmsg_out = ERR_CANNOT_MODIFY;
}
}
else
{
/* Maintenance and being-drained are set/cleared using a special variable which the
* monitor reads when starting the next update cycle. */
int request;
if (bit & SERVER_MAINT)
{
request = MonitorServer::MAINT_ON;
}
else
{
mxb_assert(bit & SERVER_DRAINING);
request = MonitorServer::BEING_DRAINED_ON;
}
int previous_request = atomic_exchange_int(&msrv->status_request, request);
written = true;
// Warn if the previous request hasn't been read.
if (previous_request != MonitorServer::NO_CHANGE)
{
MXS_WARNING(WRN_REQUEST_OVERWRITTEN);
}
// Also set a flag so the next loop happens sooner.
m_status_change_pending.store(true, std::memory_order_release);
}
}
else
{
/* The monitor is not running, the bit can be set directly */
srv->set_status(bit);
written = true;
}
return written;
}
bool Monitor::clear_server_status(SERVER* srv, int bit, string* errmsg_out)
{
MonitorServer* msrv = get_monitored_server(srv);
mxb_assert(msrv);
if (!msrv)
{
MXS_ERROR("Monitor %s requested to clear status of server %s that it does not monitor.",
name(), srv->address);
return false;
}
bool written = false;
if (is_running())
{
if (bit & ~(SERVER_MAINT | SERVER_DRAINING))
{
MXS_ERROR(ERR_CANNOT_MODIFY);
if (errmsg_out)
{
*errmsg_out = ERR_CANNOT_MODIFY;
}
}
else
{
int request;
if (bit & SERVER_MAINT)
{
request = MonitorServer::MAINT_OFF;
}
else
{
mxb_assert(bit & SERVER_DRAINING);
request = MonitorServer::BEING_DRAINED_OFF;
}
int previous_request = atomic_exchange_int(&msrv->status_request, request);
written = true;
// Warn if the previous request hasn't been read.
if (previous_request != MonitorServer::NO_CHANGE)
{
MXS_WARNING(WRN_REQUEST_OVERWRITTEN);
}
// Also set a flag so the next loop happens sooner.
m_status_change_pending.store(true, std::memory_order_release);
}
}
else
{
/* The monitor is not running, the bit can be cleared directly */
srv->clear_status(bit);
written = true;
}
return written;
}
void Monitor::populate_services()
{
mxb_assert(!is_running());
for (MonitorServer* pMs : m_servers)
{
service_add_server(this, pMs->server);
}
}
void Monitor::deactivate()
{
if (is_running())
{
stop();
}
remove_all_servers();
}
bool Monitor::check_disk_space_this_tick()
{
bool should_update_disk_space = false;
auto check_interval = m_settings.disk_space_check_interval;
if ((check_interval.secs() > 0) && m_disk_space_checked.split() > check_interval)
{
should_update_disk_space = true;
// Whether or not disk space check succeeds, reset the timer. This way, disk space is always
// checked during the same tick for all servers.
m_disk_space_checked.restart();
}
return should_update_disk_space;
}
bool Monitor::server_status_request_waiting() const
{
return m_status_change_pending.load(std::memory_order_acquire);
}
const Monitor::ServerVector& Monitor::servers() const
{
return m_servers;
}
MonitorWorker::MonitorWorker(const string& name, const string& module)
: Monitor(name, module)
, m_thread_running(false)
, m_shutdown(0)
, m_checked(false)
, m_loop_called(get_time_ms())
{
}
MonitorWorker::~MonitorWorker()
{
}
bool MonitorWorker::is_running() const
{
return Worker::state() != Worker::STOPPED && Worker::state() != Worker::FINISHED;
}
void MonitorWorker::do_stop()
{
// This should only be called by monitor_stop().
mxb_assert(Monitor::is_admin_thread());
mxb_assert(is_running());
mxb_assert(m_thread_running.load() == true);
Worker::shutdown();
Worker::join();
m_thread_running.store(false, std::memory_order_release);
}
void MonitorWorker::diagnostics(DCB* pDcb) const
{
}
json_t* MonitorWorker::diagnostics_json() const
{
return json_object();
}
bool MonitorWorker::start()
{
// This should only be called by monitor_start(). NULL worker is allowed since the main worker may
// not exist during program start/stop.
mxb_assert(Monitor::is_admin_thread());
mxb_assert(!is_running());
mxb_assert(m_thread_running.load() == false);
if (journal_is_stale())
{
MXS_NOTICE("Removing stale journal file for monitor '%s'.", name());
remove_server_journal();
}
if (!m_checked)
{
if (!has_sufficient_permissions())
{
MXS_ERROR("Failed to start monitor. See earlier errors for more information.");
}
else
{
m_checked = true;
}
}
bool started = false;
if (m_checked)
{
m_loop_called = get_time_ms() - settings().interval; // Next tick should happen immediately.
if (!Worker::start())
{
MXS_ERROR("Failed to start worker for monitor '%s'.", name());
}
else
{
// Ok, so the thread started. Let's wait until we can be certain the
// state has been updated.
m_semaphore.wait();
started = m_thread_running.load(std::memory_order_acquire);
if (!started)
{
// Ok, so the initialization failed and the thread will exit.
// We need to wait on it so that the thread resources will not leak.
Worker::join();
}
}
}
return started;
}
// static
int64_t MonitorWorker::get_time_ms()
{
timespec t;
MXB_AT_DEBUG(int rv = ) clock_gettime(CLOCK_MONOTONIC_COARSE, &t);
mxb_assert(rv == 0);
return t.tv_sec * 1000 + (t.tv_nsec / 1000000);
}
bool MonitorServer::can_update_disk_space_status() const
{
return ok_to_check_disk_space && (!monitor_limits.empty() || server->have_disk_space_limits());
}
void MonitorServer::update_disk_space_status()
{
auto pMs = this; // TODO: Clean
std::map<std::string, disk::SizesAndName> info;
int rv = disk::get_info_by_path(pMs->con, &info);
if (rv == 0)
{
// Server-specific setting takes precedence.
auto dst = pMs->server->get_disk_space_limits();
if (dst.empty())
{
dst = monitor_limits;
}
bool disk_space_exhausted = false;
int32_t star_max_percentage = -1;
std::set<std::string> checked_paths;
for (const auto& dst_item : dst)
{
string path = dst_item.first;
int32_t max_percentage = dst_item.second;
if (path == "*")
{
star_max_percentage = max_percentage;
}
else
{
auto j = info.find(path);
if (j != info.end())
{
const disk::SizesAndName& san = j->second;
disk_space_exhausted = check_disk_space_exhausted(pMs, path, san, max_percentage);
checked_paths.insert(path);
}
else
{
MXS_WARNING("Disk space threshold specified for %s even though server %s at %s"
"does not have that.",
path.c_str(),
pMs->server->name(),
pMs->server->address);
}
}
}
if (star_max_percentage != -1)
{
for (auto j = info.begin(); j != info.end(); ++j)
{
string path = j->first;
if (checked_paths.find(path) == checked_paths.end())
{
const disk::SizesAndName& san = j->second;
disk_space_exhausted = check_disk_space_exhausted(pMs, path, san, star_max_percentage);
}
}
}
if (disk_space_exhausted)
{
pMs->pending_status |= SERVER_DISK_SPACE_EXHAUSTED;
}
else
{
pMs->pending_status &= ~SERVER_DISK_SPACE_EXHAUSTED;
}
}
else
{
SERVER* pServer = pMs->server;
if (mysql_errno(pMs->con) == ER_UNKNOWN_TABLE)
{
// Disable disk space checking for this server.
pMs->ok_to_check_disk_space = false;
MXS_ERROR("Disk space cannot be checked for %s at %s, because either the "
"version (%s) is too old, or the DISKS information schema plugin "
"has not been installed. Disk space checking has been disabled.",
pServer->name(),
pServer->address,
pServer->version_string().c_str());
}
else
{
MXS_ERROR("Checking the disk space for %s at %s failed due to: (%d) %s",
pServer->name(),
pServer->address,
mysql_errno(pMs->con),
mysql_error(pMs->con));
}
}
}
bool MonitorWorker::configure(const MXS_CONFIG_PARAMETER* pParams)
{
return Monitor::configure(pParams);
}
bool MonitorWorker::has_sufficient_permissions()
{
return true;
}
void MonitorWorker::flush_server_status()
{
for (MonitorServer* pMs : servers())
{
if (!pMs->server->is_in_maint())
{
pMs->server->status = pMs->pending_status;
}
}
}
void MonitorWorkerSimple::pre_loop()
{
m_master = nullptr;
load_server_journal(&m_master);
// Add another overridable function for derived classes (e.g. pre_loop_monsimple) if required.
}
void MonitorWorkerSimple::post_loop()
{
}
void MonitorWorkerSimple::pre_tick()
{
}
void MonitorWorkerSimple::post_tick()
{
}
void MonitorWorkerSimple::tick()
{
check_maintenance_requests();
pre_tick();
const bool should_update_disk_space = check_disk_space_this_tick();
for (MonitorServer* pMs : servers())
{
if (!pMs->server->is_in_maint())
{
pMs->mon_prev_status = pMs->server->status;
pMs->pending_status = pMs->server->status;
ConnectResult rval = pMs->ping_or_connect(settings().conn_settings);
if (connection_is_ok(rval))
{
pMs->clear_pending_status(SERVER_AUTH_ERROR);
pMs->set_pending_status(SERVER_RUNNING);
if (should_update_disk_space && pMs->can_update_disk_space_status())
{
pMs->update_disk_space_status();
}
update_server_status(pMs);
}
else
{
/**
* TODO: Move the bits that do not represent a state out of
* the server state bits. This would allow clearing the state by
* zeroing it out.
*/
pMs->clear_pending_status(MonitorServer::SERVER_DOWN_CLEAR_BITS);
if (rval == ConnectResult::ACCESS_DENIED)
{
pMs->set_pending_status(SERVER_AUTH_ERROR);
}
if (pMs->status_changed() && pMs->should_print_fail_status())
{
pMs->log_connect_error(rval);
}
}
#if defined (SS_DEBUG)
if (pMs->status_changed() || pMs->should_print_fail_status())
{
// The current status is still in pMs->pending_status.
MXS_DEBUG("Backend server [%s]:%d state : %s",
pMs->server->address, pMs->server->port,
SERVER::status_to_string(pMs->pending_status).c_str());
}
#endif
if (pMs->server->is_down())
{
pMs->mon_err_count += 1;
}
else
{
pMs->mon_err_count = 0;
}
}
}
post_tick();
flush_server_status();
process_state_changes();
hangup_failed_servers();
store_server_journal(m_master);
}
void MonitorWorker::pre_loop()
{
}
void MonitorWorker::post_loop()
{
}
void MonitorWorker::process_state_changes()
{
detect_handle_state_changes();
}
bool MonitorWorker::pre_run()
{
bool rv = false;
if (mysql_thread_init() == 0)
{
rv = true;
// Write and post the semaphore to signal the admin thread that the start is succeeding.
m_thread_running.store(true, std::memory_order_release);
m_semaphore.post();
pre_loop();
delayed_call(1, &MonitorWorker::call_run_one_tick, this);
}
else
{
MXS_ERROR("mysql_thread_init() failed for %s. The monitor cannot start.", name());
m_semaphore.post();
}
return rv;
}
void MonitorWorker::post_run()
{
post_loop();
mysql_thread_end();
}
bool MonitorWorker::call_run_one_tick(Worker::Call::action_t action)
{
/** This is both the minimum sleep between two ticks and also the maximum time between early
* wakeup checks. */
const int base_interval_ms = 100;
if (action == Worker::Call::EXECUTE)
{
int64_t now = get_time_ms();
// Enough time has passed,
if ((now - m_loop_called > settings().interval)
// or a server status change request is waiting,
|| server_status_request_waiting()
// or a monitor-specific condition is met.
|| immediate_tick_required())
{
m_loop_called = now;
run_one_tick();
now = get_time_ms();
}
int64_t ms_to_next_call = settings().interval - (now - m_loop_called);
// ms_to_next_call will be negative, if the run_one_tick() call took
// longer than one monitor interval.
int64_t delay = ((ms_to_next_call <= 0) || (ms_to_next_call >= base_interval_ms)) ?
base_interval_ms : ms_to_next_call;
delayed_call(delay, &MonitorWorker::call_run_one_tick, this);
}
return false;
}
void MonitorWorker::run_one_tick()
{
tick();
m_ticks.fetch_add(1, std::memory_order_acq_rel);
}
bool MonitorWorker::immediate_tick_required() const
{
return false;
}
MonitorServer::MonitorServer(SERVER* server, const SERVER::DiskSpaceLimits& monitor_limits)
: server(server)
, monitor_limits(monitor_limits)
{
}
MonitorServer::~MonitorServer()
{
if (con)
{
mysql_close(con);
}
}
}
Reference in New Issue View Git Blame Copy Permalink