def get_discover(http_context, app, sessions):
logger.info('Starting discovery.')
try:
sysinfo = SysInfo()
with app.postgres.connect() as conn:
pginfo = PgInfo(conn)
ret = dict(hostname=sysinfo.hostname(
app.config.temboard['hostname']),
cpu=sysinfo.n_cpu(),
memory_size=sysinfo.memory_size(),
pg_port=pginfo.setting('port'),
pg_version=pginfo.version()['full'],
pg_version_summary=pginfo.version()['summary'],
pg_data=pginfo.setting('data_directory'),
plugins=[
plugin_name
for plugin_name in app.config.temboard['plugins']
])
logger.info('Discovery done.')
return ret
except (error, Exception, HTTPError) as e:
logger.exception(e)
logger.info('Discovery failed.')
if isinstance(e, HTTPError):
raise e
else:
raise HTTPError(500, "Internal error.")
def api_run_probe(probe_instance, config):
"""
Run a probe instance.
"""
config.plugins['monitoring']['conninfo'] = [{
'host':
config.postgresql['host'],
'port':
config.postgresql['port'],
'user':
config.postgresql['user'],
'database':
config.postgresql['dbname'],
'password':
config.postgresql['password'],
'dbnames':
config.plugins['monitoring']['dbnames'],
'instance':
config.postgresql['instance']
}]
# Validate connection information from the config, and ensure
# the instance is available
instances = []
sysinfo = SysInfo()
hostname = sysinfo.hostname(config.temboard['hostname'])
for conninfo in config.plugins['monitoring']['conninfo']:
logging.debug("Validate connection information on instance \"%s\"",
conninfo['instance'])
instances.append(instance_info(conninfo, hostname))
# Set home path
probe_instance.set_home(config.temboard['home'])
# Gather the data from probes
data = run_probes([probe_instance], instances, delta=False)
return data
def get_discover(http_context, app, sessions):
logger.info('Starting discovery.')
# Optionnal validation of key. For compatibility, we accept unauthenticated
# /discover. But for better reliability, we validate a key sent by HTTP
# header. temboard-agent-register sends key to prevent configuration
# mismatch.
request_key = http_context['headers'].get('X-Temboard-Agent-Key')
if request_key and request_key != app.config.temboard['key']:
raise HTTPError(401, "Invalid key")
discover = dict(
hostname=None,
cpu=None,
memory_size=None,
pg_port=app.config.postgresql['port'],
pg_version=None,
pg_version_summary=None,
pg_data=None,
plugins=[plugin for plugin in app.config.temboard['plugins']],
)
try:
# Gather system informations
sysinfo = SysInfo()
hostname = sysinfo.hostname(app.config.temboard['hostname'])
cpu = sysinfo.n_cpu()
memory_size = sysinfo.memory_size()
except (Exception, HTTPError) as e:
logger.exception(str(e))
logger.error('System discovery failed.')
# We stop here if system information has not been collected
if isinstance(e, HTTPError):
raise e
else:
raise HTTPError(500, "Internal error.")
discover.update(hostname=hostname, cpu=cpu, memory_size=memory_size)
try:
with app.postgres.connect() as conn:
pginfo = PgInfo(conn)
discover.update(pg_block_size=int(pginfo.setting('block_size')),
pg_version=pginfo.version()['full'],
pg_version_summary=pginfo.version()['summary'],
pg_data=pginfo.setting('data_directory'))
except Exception as e:
logger.exception(str(e))
logger.error('Postgres discovery failed.')
# Do not raise HTTPError, just keeping null values for Postgres
# informations.
logger.info('Discovery done.')
logger.debug(discover)
return discover
def run(self):
# Everything is already gathered in the inventory, just add
# the time
out = []
datetime = now()
sysinfo = SysInfo()
for fs in sysinfo.file_systems():
fs['datetime'] = datetime
out.append(fs)
return out
def get_discover(http_context, app, sessions):
logger.info('Starting discovery.')
discover = dict(
hostname=None,
cpu=None,
memory_size=None,
pg_port=app.config.postgresql['port'],
pg_version=None,
pg_version_summary=None,
pg_data=None,
plugins=[plugin for plugin in app.config.temboard['plugins']],
)
try:
# Gather system informations
sysinfo = SysInfo()
hostname = sysinfo.hostname(app.config.temboard['hostname'])
cpu = sysinfo.n_cpu()
memory_size = sysinfo.memory_size()
except (Exception, HTTPError) as e:
logger.exception(str(e))
logger.error('System discovery failed.')
# We stop here if system information has not been collected
if isinstance(e, HTTPError):
raise e
else:
raise HTTPError(500, "Internal error.")
discover.update(
hostname=hostname, cpu=cpu, memory_size=memory_size
)
try:
with app.postgres.connect() as conn:
pginfo = PgInfo(conn)
discover.update(
pg_block_size=int(pginfo.setting('block_size')),
pg_version=pginfo.version()['full'],
pg_version_summary=pginfo.version()['summary'],
pg_data=pginfo.setting('data_directory')
)
except Exception as e:
logger.exception(str(e))
logger.error('Postgres discovery failed.')
# Do not raise HTTPError, just keeping null values for Postgres
# informations.
logger.info('Discovery done.')
logger.debug(discover)
return discover
def get_info(conn, config, _):
dm = DashboardMetrics(conn)
sysinfo = SysInfo()
pginfo = PgInfo(conn)
return {
'hostname': sysinfo.hostname(config.temboard['hostname']),
'os_version': "%s %s" % (sysinfo.os, sysinfo.os_release),
'pg_uptime': dm.get_pg_uptime(),
'pg_version': pginfo.version()['full'],
'pg_data': pginfo.setting('data_directory'),
'pg_port': pginfo.setting('port'),
}
def get_info(conn, config):
dm = DashboardMetrics(conn)
sysinfo = SysInfo()
pginfo = PgInfo(conn)
return dict(
hostname=sysinfo.hostname(config.temboard.hostname),
os_version=' '.join([sysinfo.os, sysinfo.os_release]),
pg_uptime=dm.get_pg_uptime(),
pg_version=pginfo.version()['full'],
pg_data=pginfo.setting('data_directory'),
pg_port=pginfo.setting('port'),
)
def run(self):
sysinfo = SysInfo()
meminfo = sysinfo.mem_info()
return [{
'datetime': now(),
'mem_total': meminfo['MemTotal'],
'mem_used': meminfo['MemTotal'] - meminfo['MemFree'],
'mem_free': meminfo['MemFree'],
'mem_buffers': meminfo['Buffers'],
'mem_cached': meminfo['Cached'],
'swap_total': meminfo['SwapTotal'],
'swap_used': meminfo['SwapTotal'] - meminfo['SwapFree']
}]
def get_metrics(conn, config):
dm = DashboardMetrics(conn)
sysinfo = SysInfo()
pginfo = PgInfo(conn)
cpu_models = [cpu['model_name'] for cpu in sysinfo.cpu_info()['cpus']]
cpu_models_counter = {}
for elem in cpu_models:
cpu_models_counter[elem] = cpu_models_counter.get(elem, 0) + 1
return dict(
buffers=dm.get_buffers(),
hitratio=dm.get_hitratio(),
active_backends=dm.get_active_backends(),
max_connections=dm.get_max_connections(),
cpu=dm.get_cpu_usage(),
loadaverage=dm.get_load_average(),
memory=dm.get_memory_usage(),
hostname=sysinfo.hostname(config.temboard.hostname),
os_version=sysinfo.os_release,
linux_distribution=sysinfo.linux_distribution(),
cpu_models=cpu_models_counter,
databases=dm.get_stat_db(),
pg_uptime=dm.get_pg_uptime(),
n_cpu=sysinfo.n_cpu(),
pg_version=pginfo.version()['full'],
pg_data=pginfo.setting('data_directory'),
pg_port=pginfo.setting('port'),
notifications=dm.get_notifications(config),
)
def get_metrics(conn, config, _=None):
dm = DashboardMetrics(conn)
sysinfo = SysInfo()
pginfo = PgInfo(conn)
cpu_models = [cpu['model_name'] for cpu in sysinfo.cpu_info()['cpus']]
cpu_models_counter = {}
for elem in cpu_models:
cpu_models_counter[elem] = cpu_models_counter.get(elem, 0) + 1
return {
'buffers': dm.get_buffers(),
'hitratio': dm.get_hitratio(),
'active_backends': dm.get_active_backends(),
'max_connections': dm.get_max_connections(),
'cpu': dm.get_cpu_usage(),
'loadaverage': dm.get_load_average(),
'memory': dm.get_memory_usage(),
'hostname': sysinfo.hostname(config.temboard['hostname']),
'os_version': sysinfo.os_release,
'linux_distribution': sysinfo.linux_distribution(),
'cpu_models': cpu_models_counter,
'databases': dm.get_stat_db(),
'pg_uptime': dm.get_pg_uptime(),
'n_cpu': sysinfo.n_cpu(),
'pg_version': pginfo.version()['full'],
'pg_data': pginfo.setting('data_directory'),
'pg_port': pginfo.setting('port'),
'notifications': dm.get_notifications(config)
}
def host_info(hostname_cfg):
"""Gather system information."""
sinfo = SysInfo()
_, _, _, _, arch = sinfo.uname()
hostname = sinfo.hostname(hostname_cfg)
hostinfo = {
"hostname": hostname,
"os": sinfo.os,
"os_version": sinfo.os_release,
"cpu_arch": arch
}
hostinfo.update(sinfo.cpu_info())
hostinfo['memory_size'] = sinfo.mem_info()['MemTotal']
hostinfo['ip_addresses'] = sinfo.ip_addresses()
hostinfo['filesystems'] = sinfo.file_systems()
hostinfo['os_flavor'] = sinfo.os_flavor()
return hostinfo
def api_run_probe(probe_instance, config):
"""
Run a probe instance.
"""
# Validate connection information from the config, and ensure
# the instance is available
conninfo = dict(
host=config.postgresql.host,
port=config.postgresql.port,
user=config.postgresql.user,
database=config.postgresql.dbname,
password=config.postgresql.password,
dbnames=config.monitoring.dbnames,
instance=config.postgresql.instance,
)
sysinfo = SysInfo()
hostname = sysinfo.hostname(config.temboard.hostname)
instance = instance_info(conninfo, hostname)
# Set home path
probe_instance.set_home(config.temboard.home)
# Gather the data from probes
return run_probes([probe_instance], [instance], delta=False)
def api_run_probe(probe_instance, config):
"""
Run a probe instance.
"""
set_logger_name("supervision")
logger = get_logger(config)
# TODO: logging methods in supervision_agent code and supervision_agent should be aligned.
logging.root = logger
config.plugins['supervision']['conninfo'] = [{
'host':
config.postgresql['host'],
'port':
config.postgresql['port'],
'user':
config.postgresql['user'],
'database':
config.postgresql['dbname'],
'password':
config.postgresql['password'],
'dbnames':
config.plugins['supervision']['dbnames'],
'instance':
config.postgresql['instance']
}]
# Validate connection information from the config, and ensure
# the instance is available
instances = []
sysinfo = SysInfo()
hostname = sysinfo.hostname(config.temboard['hostname'])
for conninfo in config.plugins['supervision']['conninfo']:
logging.debug("Validate connection information on instance \"%s\"",
conninfo['instance'])
instances.append(instance_info(conninfo, hostname))
# Gather the data from probes
data = run_probes([probe_instance], instances, delta=False)
return data
def get_discover(http_context, config=None, sessions=None):
conn = connector(host=config.postgresql['host'],
port=config.postgresql['port'],
user=config.postgresql['user'],
password=config.postgresql['password'],
database=config.postgresql['dbname'])
logger.info('Starting discovery.')
try:
conn.connect()
sysinfo = SysInfo()
pginfo = PgInfo(conn)
ret = {
'hostname': sysinfo.hostname(config.temboard['hostname']),
'cpu': sysinfo.n_cpu(),
'memory_size': sysinfo.memory_size(),
'pg_port': pginfo.setting('port'),
'pg_version': pginfo.version()['full'],
'pg_data': pginfo.setting('data_directory'),
'plugins':
[plugin_name for plugin_name in config.temboard['plugins']]
}
conn.close()
logger.info('Discovery done.')
return ret
except (error, Exception, HTTPError) as e:
logger.exception(str(e))
logger.info('Discovery failed.')
try:
conn.close()
except Exception:
pass
if isinstance(e, HTTPError):
raise e
else:
raise HTTPError(500, "Internal error.")
def get_metrics(app):
res = dict()
try:
with app.postgres.connect() as conn:
dm = DashboardMetrics(conn)
pginfo = PgInfo(conn)
res.update(
dict(
buffers=dm.get_buffers(),
hitratio=dm.get_hitratio(),
active_backends=dm.get_active_backends(),
max_connections=dm.get_max_connections(),
databases=dm.get_stat_db(),
pg_uptime=dm.get_pg_uptime(),
pg_version=pginfo.version()['full'],
pg_data=pginfo.setting('data_directory'),
pg_port=pginfo.setting('port'),
))
except UserError:
pass
dm = DashboardMetrics()
res.update(
dict(
cpu=dm.get_cpu_usage(),
loadaverage=dm.get_load_average(),
memory=dm.get_memory_usage(),
notifications=dm.get_notifications(app.config),
))
sysinfo = SysInfo()
cpu_models = [cpu['model_name'] for cpu in sysinfo.cpu_info()['cpus']]
cpu_models_counter = {}
for elem in cpu_models:
cpu_models_counter[elem] = cpu_models_counter.get(elem, 0) + 1
res.update(
dict(hostname=sysinfo.hostname(app.config.temboard.hostname),
os_version=sysinfo.os_release,
linux_distribution=sysinfo.linux_distribution(),
cpu_models=cpu_models_counter,
n_cpu=sysinfo.n_cpu(),
timestamp=time.time()))
return res
def get_os_version(config, _):
sysinfo = SysInfo()
return {'os_version': "%s %s" % (sysinfo.os, sysinfo.os_release)}
def get_n_cpu(config, _):
sysinfo = SysInfo()
return {'n_cpu': sysinfo.n_cpu()}
def get_memory_usage(self, ):
sysinfo = SysInfo()
if sysinfo.os == 'Linux':
return self._get_memory_usage_linux()
def get_n_cpu():
sysinfo = SysInfo()
return dict(n_cpu=sysinfo.n_cpu())
def get_os_version():
sysinfo = SysInfo()
return dict(os_version=' '.join((sysinfo.os, sysinfo.os_release)))
def get_hostname(config):
sysinfo = SysInfo()
return dict(hostname=sysinfo.hostname(config.temboard.hostname))
def get_hostname(config, _):
sysinfo = SysInfo()
return {'hostname': sysinfo.hostname(config.temboard['hostname'])}