class TestPostgresql(unittest.TestCase):
def __init__(self, method_name='runTest'):
self.setUp = self.set_up
self.tearDown = self.tear_down
super(TestPostgresql, self).__init__(method_name)
def set_up(self):
subprocess.call = subprocess_call
shutil.copy = nop
self.p = Postgresql({'name': 'test0', 'data_dir': 'data/test0', 'listen': '127.0.0.1, *:5432',
'connect_address': '127.0.0.2:5432',
'pg_hba': ['hostssl all all 0.0.0.0/0 md5', 'host all all 0.0.0.0/0 md5'],
'superuser': {'password': ''}, 'admin': {'username': '******', 'password': '******'},
'replication': {'username': '******',
'password': '******',
'network': '127.0.0.1/32'},
'parameters': {'foo': 'bar'}, 'recovery_conf': {'foo': 'bar'}},
on_change_callback=lambda state: True)
psycopg2.connect = psycopg2_connect
if not os.path.exists(self.p.data_dir):
os.makedirs(self.p.data_dir)
self.leader = Member('leader', 'postgres://*****:*****@127.0.0.1:5434/postgres', None, 28)
def tear_down(self):
shutil.rmtree('data')
def test_data_directory_empty(self):
self.assertTrue(self.p.data_directory_empty())
def test_initialize(self):
self.assertTrue(self.p.initialize())
self.assertTrue(os.path.exists(os.path.join(self.p.data_dir, 'pg_hba.conf')))
def test_start(self):
self.assertFalse(self.p.start())
self.p.is_running = is_running
with open(os.path.join(self.p.data_dir, 'postmaster.pid'), 'w'):
pass
self.assertTrue(self.p.start())
def test_sync_from_leader(self):
self.assertTrue(self.p.sync_from_leader(self.leader))
def test_follow_the_leader(self):
self.p.demote(self.leader)
self.p.follow_the_leader(None)
self.p.demote(self.leader)
self.p.follow_the_leader(self.leader)
self.p.follow_the_leader(Member('leader', 'postgres://*****:*****@127.0.0.1:5435/postgres', None, 28))
def test_create_replication_slots(self):
self.p.start()
me = Member('test0', 'postgres://*****:*****@127.0.0.1:5434/postgres', None, 28)
other = Member('test1', 'postgres://*****:*****@127.0.0.1:5433/postgres', None, 28)
cluster = Cluster(True, self.leader, 0, [me, other, self.leader])
self.p.create_replication_slots(cluster)
def test_query(self):
self.p.query('select 1')
self.assertRaises(psycopg2.InterfaceError, self.p.query, 'InterfaceError')
self.assertRaises(psycopg2.OperationalError, self.p.query, 'blabla')
self.p._connection.closed = 2
self.assertRaises(psycopg2.OperationalError, self.p.query, 'blabla')
self.p._connection.closed = 2
self.p.disconnect = false
self.assertRaises(psycopg2.OperationalError, self.p.query, 'blabla')
def test_is_healthiest_node(self):
leader = Member('leader', 'postgres://*****:*****@127.0.0.1:5435/postgres', None, 28)
me = Member('test0', 'postgres://*****:*****@127.0.0.1:5434/postgres', None, 28)
other = Member('test1', 'postgres://*****:*****@127.0.0.1:5433/postgres', None, 28)
cluster = Cluster(True, leader, 0, [me, other, leader])
self.assertTrue(self.p.is_healthiest_node(cluster))
self.p.is_leader = false
self.assertFalse(self.p.is_healthiest_node(cluster))
self.p.xlog_position = xlog_position
self.assertTrue(self.p.is_healthiest_node(cluster))
self.p.config['maximum_lag_on_failover'] = -2
self.assertFalse(self.p.is_healthiest_node(cluster))
def test_is_leader(self):
self.p.is_promoted = True
self.assertTrue(self.p.is_leader())
self.assertFalse(self.p.is_promoted)
def test_reload(self):
self.assertTrue(self.p.reload())
def test_is_healthy(self):
self.assertTrue(self.p.is_healthy())
self.p.is_running = is_running
self.assertFalse(self.p.is_healthy())
def test_promote(self):
self.assertTrue(self.p.promote())
def test_last_operation(self):
self.assertEquals(self.p.last_operation(), 0)
etcd.touch_member(postgresql.name, postgresql.connection_string)
etcd_ready = True
except urllib2.URLError:
logging.info("waiting on etcd: %s" % message)
time.sleep(5)
logging.info("Governor Starting up")
# is data directory empty?
if postgresql.data_directory_empty():
logging.info("Governor Starting up: Empty Data Dir")
# racing to initialize
wait_for_etcd("cannot initialize member without ETCD")
if etcd.race("/initialize", postgresql.name):
logging.info("Governor Starting up: Initialisation Race ... WON!!!")
logging.info("Governor Starting up: Initialise Postgres")
postgresql.initialize()
logging.info("Governor Starting up: Initialise Complete")
etcd.take_leader(postgresql.name)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
else:
logging.info("Governor Starting up: Initialisation Race ... LOST")
logging.info("Governor Starting up: Sync Postgres from Leader")
synced_from_leader = False
while not synced_from_leader:
leader = etcd.current_leader()
if not leader:
time.sleep(5)
continue
if postgresql.sync_from_leader(leader):
logging.info("Governor Starting up: Sync Completed")
def run(config):
etcd = Etcd(config["etcd"])
postgresql = Postgresql(config["postgresql"])
ha = Ha(postgresql, etcd)
atexit.register(stop_postgresql, postgresql)
logging.info("Governor Starting up")
# is data directory empty?
if postgresql.data_directory_empty():
logging.info("Governor Starting up: Empty Data Dir")
# racing to initialize
wait_for_etcd("cannot initialize member without ETCD", etcd, postgresql)
if etcd.race("/initialize", postgresql.name):
logging.info("Governor Starting up: Initialisation Race ... WON!!!")
logging.info("Governor Starting up: Initialise Postgres")
postgresql.initialize()
logging.info("Governor Starting up: Initialise Complete")
etcd.take_leader(postgresql.name)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
else:
logging.info("Governor Starting up: Initialisation Race ... LOST")
logging.info("Governor Starting up: Sync Postgres from Leader")
synced_from_leader = False
while not synced_from_leader:
leader = etcd.current_leader()
if not leader:
time.sleep(5)
continue
if postgresql.sync_from_leader(leader):
logging.info("Governor Starting up: Sync Completed")
postgresql.write_recovery_conf(leader)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
synced_from_leader = True
else:
time.sleep(5)
else:
logging.info("Governor Starting up: Existing Data Dir")
postgresql.follow_no_leader()
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
wait_for_etcd("running in readonly mode; cannot participate in cluster HA without etcd", etcd, postgresql)
logging.info("Governor Running: Starting Running Loop")
while True:
try:
logging.info("Governor Running: %s" % ha.run_cycle())
# create replication slots
if postgresql.is_leader():
logging.info("Governor Running: I am the Leader")
for node in etcd.get_client_path("/members?recursive=true")["node"]["nodes"]:
member = node["key"].split('/')[-1]
if member != postgresql.name:
postgresql.query("DO LANGUAGE plpgsql $$DECLARE somevar VARCHAR; BEGIN SELECT slot_name INTO somevar FROM pg_replication_slots WHERE slot_name = '%(slot)s' LIMIT 1; IF NOT FOUND THEN PERFORM pg_create_physical_replication_slot('%(slot)s'); END IF; END$$;" % {"slot": member})
etcd.touch_member(postgresql.name, postgresql.connection_string)
time.sleep(config["loop_wait"])
except urllib2.URLError:
logging.info("Lost connection to etcd, setting no leader and waiting on etcd")
postgresql.follow_no_leader()
wait_for_etcd("running in readonly mode; cannot participate in cluster HA without etcd", etcd, postgresql)
def run(config):
etcd = Etcd(config["etcd"])
postgresql = Postgresql(config["postgresql"])
ha = Ha(postgresql, etcd)
atexit.register(stop_postgresql, postgresql)
logging.info("Governor Starting up")
# is data directory empty?
if postgresql.data_directory_empty():
logging.info("Governor Starting up: Empty Data Dir")
# racing to initialize
wait_for_etcd("cannot initialize member without ETCD", etcd,
postgresql)
if etcd.race("/initialize", postgresql.name) or not etcd.members():
logging.info(
"Governor Starting up: Initialisation Race ... WON!!!")
logging.info("Governor Starting up: Initialise Postgres")
postgresql.initialize()
logging.info("Governor Starting up: Initialise Complete")
etcd.take_leader(postgresql.name)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
else:
logging.info("Governor Starting up: Initialisation Race ... LOST")
logging.info("Governor Starting up: Sync Postgres from Leader")
synced_from_leader = False
while not synced_from_leader:
leader = etcd.current_leader()
if not leader:
time.sleep(5)
continue
if postgresql.sync_from_leader(leader):
logging.info("Governor Starting up: Sync Completed")
postgresql.write_recovery_conf(leader)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
synced_from_leader = True
else:
time.sleep(5)
else:
logging.info("Governor Starting up: Existing Data Dir")
postgresql.follow_no_leader()
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
wait_for_etcd(
"running in readonly mode; cannot participate in cluster HA without etcd",
etcd, postgresql)
logging.info("Governor Running: Starting Running Loop")
while True:
try:
ha.run_cycle()
# create replication slots
if postgresql.is_leader():
logging.info("Governor Running: I am the Leader")
for node in etcd.get_client_path(
"/members?recursive=true").get("node",
{}).get("nodes", []):
member = node["key"].split('/')[-1]
if member != postgresql.name:
postgresql.query(
"DO LANGUAGE plpgsql $$DECLARE somevar VARCHAR; BEGIN SELECT slot_name INTO somevar FROM pg_replication_slots WHERE slot_name = '%(slot)s' LIMIT 1; IF NOT FOUND THEN PERFORM pg_create_physical_replication_slot('%(slot)s'); END IF; END$$;"
% {"slot": member})
etcd.touch_member(postgresql.name, postgresql.connection_string)
time.sleep(config["loop_wait"])
except urllib2.URLError:
logging.info(
"Lost connection to etcd, setting no leader and waiting on etcd"
)
postgresql.follow_no_leader()
wait_for_etcd(
"running in readonly mode; cannot participate in cluster HA without etcd",
etcd, postgresql)
try:
etcd.touch_member(postgresql.name,
postgresql.advertised_connection_string)
etcd_ready = True
except urllib2.URLError:
logging.info("waiting on etcd")
time.sleep(5)
# is data directory empty?
if postgresql.data_directory_empty():
logging.info("Governor Starting up: Empty Data Dir")
# racing to initialize
if etcd.race("/initialize", postgresql.name):
logging.info("Governor Starting up: Initialisation Race ... WON!!!")
logging.info("Governor Starting up: Initialise Postgres")
postgresql.initialize()
logging.info("Governor Starting up: Initialise Complete")
etcd.take_leader(postgresql.name)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start(master=True)
else:
logging.info("Governor Starting up: Initialisation Race ... LOST")
time.sleep(20)
logging.info("Governor Starting up: Sync Postgres from Leader")
synced_from_leader = False
while not synced_from_leader:
leader = etcd.current_leader()
if not leader:
time.sleep(5)
continue
if postgresql.sync_from_leader(leader):
class Patroni:
def __init__(self, config):
self.nap_time = config['loop_wait']
self.postgresql = Postgresql(config['postgresql'])
self.ha = Ha(self.postgresql, self.get_dcs(self.postgresql.name, config))
host, port = config['restapi']['listen'].split(':')
self.api = RestApiServer(self, config['restapi'])
self.skydns2 = config.get('skydns2')
self.next_run = time.time()
self.shutdown_member_ttl = 300
@staticmethod
def get_dcs(name, config):
if 'etcd' in config:
assert config['etcd']['ttl'] > 2 * config['loop_wait']
return Etcd(name, config['etcd'])
if 'zookeeper' in config:
return ZooKeeper(name, config['zookeeper'])
raise Exception('Can not find sutable configuration of distributed configuration store')
def touch_member(self, ttl=None):
connection_string = self.postgresql.connection_string + '?application_name=' + self.api.connection_string
if self.ha.cluster:
for m in self.ha.cluster.members:
# Do not update member TTL when it is far from being expired
if m.name == self.postgresql.name and m.real_ttl() > self.shutdown_member_ttl:
return True
return self.ha.dcs.touch_member(connection_string, ttl)
def initialize(self):
# wait for etcd to be available
while not self.touch_member():
logger.info('waiting on DCS')
sleep(5)
# is data directory empty?
if self.postgresql.data_directory_empty():
# racing to initialize
if self.ha.dcs.race('/initialize'):
self.postgresql.initialize()
self.ha.dcs.take_leader()
self.postgresql.start()
else:
while True:
leader = self.ha.dcs.current_leader()
if leader and self.postgresql.sync_from_leader(leader):
self.postgresql.write_recovery_conf(leader)
self.postgresql.start()
break
sleep(5)
elif self.postgresql.is_running():
self.postgresql.load_replication_slots()
def schedule_next_run(self):
self.next_run += self.nap_time
current_time = time.time()
nap_time = self.next_run - current_time
if nap_time <= 0:
self.next_run = current_time
else:
self.ha.dcs.sleep(nap_time)
def run(self):
self.api.start()
self.next_run = time.time()
while True:
self.touch_member()
logger.info(self.ha.run_cycle())
try:
if self.ha.state_handler.is_leader():
self.ha.cluster and self.ha.state_handler.create_replication_slots(self.ha.cluster)
# SkyDNS2 support: publish leader
if self.skydns2:
self.ha.dcs.client.set(self.skydns2['publish_leader'],
'{{"host": "{0}", "port": {1}}}'.format(*self.postgresql.connect_address), ttl=self.skydns2['ttl'])
else:
self.ha.state_handler.drop_replication_slots()
except:
logger.exception('Exception when changing replication slots')
reap_children()
self.schedule_next_run()
def run(config):
etcd = Etcd(config["etcd"])
postgresql = Postgresql(config["postgresql"])
ha = Ha(postgresql, etcd)
atexit.register(stop_postgresql, postgresql)
signal.signal(signal.SIGTERM, signalhandler)
logging.info("Governor Starting up")
# is data directory empty?
if postgresql.data_directory_empty():
logging.info("Governor Starting up: Empty Data Dir")
# racing to initialize
wait_for_etcd("cannot initialize member without ETCD", etcd, postgresql)
if etcd.race("/initialize", postgresql.name):
logging.info("Governor Starting up: Initialisation Race ... WON!!!")
logging.info("Governor Starting up: Initialise Postgres")
postgresql.initialize()
logging.info("Governor Starting up: Initialise Complete")
etcd.take_leader(postgresql.name)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
else:
logging.info("Governor Starting up: Initialisation Race ... LOST")
logging.info("Governor Starting up: Sync Postgres from Leader")
synced_from_leader = False
while not synced_from_leader:
leader = etcd.current_leader()
if not leader:
time.sleep(5)
continue
if postgresql.sync_from_leader(leader):
logging.info("Governor Starting up: Sync Completed")
postgresql.write_recovery_conf(leader)
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
synced_from_leader = True
else:
time.sleep(5)
else:
logging.info("Governor Starting up: Existing Data Dir")
postgresql.follow_no_leader()
logging.info("Governor Starting up: Starting Postgres")
postgresql.start()
wait_for_etcd("running in readonly mode; cannot participate in cluster HA without etcd", etcd, postgresql)
logging.info("Governor Running: Starting Running Loop")
while True:
try:
logging.info("Governor Running: %s" % ha.run_cycle())
# create replication slots
if postgresql.is_leader():
logging.info("Governor Running: I am the Leader")
for node in etcd.members():
member = node["hostname"]
if member != postgresql.name:
if postgresql.is_leader():
postgresql.ensure_replication_slot(
postgresql.replication_slot_name(member)
)
else:
postgresql.drop_replication_slot(
postgresql.replication_slot_name(member)
)
etcd.touch_member(postgresql.name, postgresql.connection_string)
time.sleep(config["loop_wait"])
except (urllib2.URLError, socket.timeout):
logging.info("Lost connection to etcd, setting no leader and waiting on etcd")
postgresql.follow_no_leader()
wait_for_etcd("running in readonly mode; cannot participate in cluster HA without etcd", etcd, postgresql)
class TestPostgresql(unittest.TestCase):
def __init__(self, method_name='runTest'):
self.setUp = self.set_up
self.tearDown = self.tear_down
super(TestPostgresql, self).__init__(method_name)
def set_up(self):
subprocess.call = subprocess_call
shutil.copy = nop
self.p = Postgresql({'name': 'test0', 'scope': 'batman', 'data_dir': 'data/test0',
'listen': '127.0.0.1, *:5432', 'connect_address': '127.0.0.2:5432',
'pg_hba': ['hostssl all all 0.0.0.0/0 md5', 'host all all 0.0.0.0/0 md5'],
'superuser': {'password': ''},
'admin': {'username': '******', 'password': '******'},
'replication': {'username': '******',
'password': '******',
'network': '127.0.0.1/32'},
'parameters': {'foo': 'bar'}, 'recovery_conf': {'foo': 'bar'},
'callbacks': {'on_start': 'true', 'on_stop': 'true',
'on_restart': 'true', 'on_role_change': 'true',
'on_reload': 'true'
},
'restore': 'true'})
psycopg2.connect = psycopg2_connect
if not os.path.exists(self.p.data_dir):
os.makedirs(self.p.data_dir)
self.leader = Member(0, 'leader', 'postgres://*****:*****@127.0.0.1:5435/postgres', None, None, 28)
self.other = Member(0, 'test1', 'postgres://*****:*****@127.0.0.1:5433/postgres', None, None, 28)
self.me = Member(0, 'test0', 'postgres://*****:*****@127.0.0.1:5434/postgres', None, None, 28)
def tear_down(self):
shutil.rmtree('data')
def mock_query(self, p):
raise psycopg2.OperationalError("not supported")
def test_data_directory_empty(self):
self.assertTrue(self.p.data_directory_empty())
def test_initialize(self):
self.assertTrue(self.p.initialize())
self.assertTrue(os.path.exists(os.path.join(self.p.data_dir, 'pg_hba.conf')))
def test_start_stop(self):
self.assertFalse(self.p.start())
self.p.is_running = is_running
with open(os.path.join(self.p.data_dir, 'postmaster.pid'), 'w'):
pass
self.assertTrue(self.p.start())
self.assertTrue(self.p.stop())
def test_sync_from_leader(self):
self.assertTrue(self.p.sync_from_leader(self.leader))
def test_follow_the_leader(self):
self.p.demote(self.leader)
self.p.follow_the_leader(None)
self.p.demote(self.leader)
self.p.follow_the_leader(self.leader)
self.p.follow_the_leader(self.other)
def test_create_connection_users(self):
cfg = self.p.config
cfg['superuser']['username'] = '******'
p = Postgresql(cfg)
p.create_connection_users()
def test_create_replication_slots(self):
self.p.start()
cluster = Cluster(True, self.leader, 0, [self.me, self.other, self.leader])
self.p.create_replication_slots(cluster)
def test_query(self):
self.p.query('select 1')
self.assertRaises(psycopg2.InterfaceError, self.p.query, 'InterfaceError')
self.assertRaises(psycopg2.OperationalError, self.p.query, 'blabla')
self.p._connection.closed = 2
self.assertRaises(psycopg2.OperationalError, self.p.query, 'blabla')
self.p._connection.closed = 2
self.p.disconnect = false
self.assertRaises(psycopg2.OperationalError, self.p.query, 'blabla')
def test_is_healthiest_node(self):
cluster = Cluster(True, self.leader, 0, [self.me, self.other, self.leader])
self.assertTrue(self.p.is_healthiest_node(cluster))
self.p.is_leader = false
self.assertFalse(self.p.is_healthiest_node(cluster))
self.p.xlog_position = lambda: 1
self.assertTrue(self.p.is_healthiest_node(cluster))
self.p.xlog_position = lambda: 2
self.assertFalse(self.p.is_healthiest_node(cluster))
self.p.config['maximum_lag_on_failover'] = -2
self.assertFalse(self.p.is_healthiest_node(cluster))
def test_is_leader(self):
self.p.is_promoted = True
self.assertTrue(self.p.is_leader())
self.assertFalse(self.p.is_promoted)
def test_reload(self):
self.assertTrue(self.p.reload())
def test_is_healthy(self):
self.assertTrue(self.p.is_healthy())
self.p.is_running = is_running
self.assertFalse(self.p.is_healthy())
def test_promote(self):
self.assertTrue(self.p.promote())
def test_last_operation(self):
self.assertEquals(self.p.last_operation(), '0')
def test_non_existing_callback(self):
self.assertFalse(self.p.call_nowait('foobar'))
def test_is_leader_exception(self):
self.p.start()
self.p.query = self.mock_query
self.assertTrue(self.p.stop())
class Governor:
def __init__(self, config):
assert config["etcd"]["ttl"] > 2 * config["loop_wait"]
self.nap_time = config['loop_wait']
self.etcd = Etcd(config['etcd'])
self.aws = AWSConnection(config)
self.postgresql = Postgresql(config['postgresql'], self.aws.on_role_change)
self.ha = Ha(self.postgresql, self.etcd)
host, port = config['restapi']['listen'].split(':')
self.api = RestApiServer(self, config['restapi'])
self.next_run = time.time()
def touch_member(self, ttl=None):
connection_string = self.postgresql.connection_string + '?application_name=' + self.api.connection_string
return self.etcd.touch_member(self.postgresql.name, connection_string, ttl)
def initialize(self):
# FIXME: isn't there a better way testing if etcd is writable?
# wait for etcd to be available
while not self.touch_member():
logging.info('waiting on etcd')
sleep(5)
# is data directory empty?
if self.postgresql.data_directory_empty():
# racing to initialize
if self.etcd.race('/initialize', self.postgresql.name):
self.postgresql.initialize()
self.etcd.take_leader(self.postgresql.name)
self.postgresql.start()
else:
# FIXME: touch_member?
while True:
leader = self.etcd.current_leader()
if leader and self.postgresql.sync_from_leader(leader):
self.postgresql.write_recovery_conf(leader)
self.postgresql.start()
break
sleep(5)
elif self.postgresql.is_running():
self.postgresql.load_replication_slots()
def schedule_next_run(self):
self.next_run += self.nap_time
current_time = time.time()
nap_time = self.next_run - current_time
if nap_time <= 0:
self.next_run = current_time
else:
sleep(nap_time)
def run(self):
self.api.start()
self.next_run = time.time()
while True:
self.touch_member()
logging.info(self.ha.run_cycle())
self.schedule_next_run()