def test_replica_start(app):
plugins = setup_plugins(app,
dcs_get_database_identifier='1234',
dcs_lock=True,
pg_am_i_replica=True,
pg_get_database_identifier='1234')
app._conn_info['a'] = 'b'
def start_monitoring():
app.unhealthy('test_monitor', 'Waiting for first check')
plugins.start_monitoring.side_effect = start_monitoring
# sync startup
timeout = app.initialize()
assert plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# compare our id with the id in the DCS
call.dcs_get_database_identifier(),
call.pg_get_database_identifier(),
# check if I am a replica
call.pg_am_i_replica(),
# not master, so sure the DB is running
call.pg_start(),
# start monitoring
call.start_monitoring(),
call.dcs_watch(conn_info=None, state=None),
# setup our connection info
call.get_conn_info(),
# set our first state
call.dcs_set_state({
'a': 'b',
'host': '127.0.0.1',
'health_problems': {'test_monitor':
{'can_be_replica': False, 'reason': 'Waiting for first check'}},
})
]
# Carry on running afterwards
assert timeout == None
assert app.health_problems == {'test_monitor': {'can_be_replica': False, 'reason': 'Waiting for first check'}}
# Our test monitor becomes healthy
plugins.reset_mock()
app.healthy('test_monitor')
assert plugins.mock_calls == [
call.dcs_set_state({'health_problems': {},
'a': 'b',
'host': '127.0.0.1',
}),
call.pg_am_i_replica(),
call.dcs_set_conn_info({'a': 'b', 'host': '127.0.0.1'}),
]
def test_failed_over_master_start(app):
# A master has failed over and restarted, another master has sucessfully advanced
plugins = setup_plugins(app,
dcs_lock=False,
dcs_get_timeline=2,
pg_get_timeline=1,
pg_am_i_replica=False)
# sync startup
timeout = app.initialize()
assert plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# compare our id with the id in the DCS
call.dcs_get_database_identifier(),
call.pg_get_database_identifier(),
# check if I am a replica
call.pg_am_i_replica(),
# no, so check if there is a master
call.dcs_lock('master'),
call.pg_stop(),
# compare our timeline to what's in the DCS
call.pg_get_timeline(),
call.dcs_get_timeline(),
# we're on an older timeline, so reset
call.pg_reset(),
]
# Carry on running afterwards
assert timeout == 5
def test_replica_slightly_sick(app):
plugins = setup_plugins(app,
pg_am_i_replica=True)
app.initialize()
plugins.reset_mock()
app.unhealthy('boom', 'It went Boom', can_be_replica=True)
assert plugins.mock_calls == [
call.dcs_set_state({'host': '127.0.0.1', 'health_problems': {'boom': {'reason': 'It went Boom', 'can_be_replica': True}}}),
call.pg_am_i_replica(),
]
async def test_master_lock_broken(app):
plugins = setup_plugins(app,
pg_am_i_replica=False)
assert app.initialize() == None
plugins.reset_mock()
# if the lock is broken, shutdown postgresql and exist
with patch('time.sleep') as sleep:
with patch('sys.exit') as exit:
app.master_lock_changed(None)
assert exit.called_once_with(0)
assert sleep.called_once_with(10)
assert app._plugins.mock_calls == [
call.pg_am_i_replica(),
call.pg_am_i_replica(),
call.pg_stop(),
call.dcs_disconnect()
]
assert app._master_lock_owner == None
# if the lock changes owner to someone else, shutdown postgresql and exist
plugins.reset_mock()
with patch('time.sleep') as sleep:
with patch('sys.exit') as exit:
app.master_lock_changed('someone else')
assert exit.called_once_with(0)
assert sleep.called_once_with(10)
assert app._plugins.mock_calls == [
call.pg_am_i_replica(),
call.pg_am_i_replica(),
call.pg_stop(),
call.dcs_disconnect()
]
assert app._master_lock_owner == 'someone else'
# if the lock is owned by us, carry on trucking
plugins.reset_mock()
with patch('time.sleep') as sleep:
with patch('sys.exit') as exit:
app.master_lock_changed(app.my_id)
assert exit.called_once_with(0)
assert sleep.called_once_with(10)
assert app._plugins.mock_calls == [
call.pg_am_i_replica(),
]
assert app._master_lock_owner == app.my_id
async def test_master_unhealthy(app):
plugins = setup_plugins(app,
pg_am_i_replica=False)
app.initialize()
plugins.reset_mock()
app.unhealthy('boom', 'It went Boom', can_be_replica=True)
assert plugins.mock_calls == [
call.dcs_set_state({'host': '127.0.0.1', 'health_problems': {'boom': {'reason': 'It went Boom', 'can_be_replica': True}}}),
call.pg_am_i_replica(),
call.dcs_delete_conn_info(),
]
async def test_plugin_subscribes_to_master_lock_change(app):
plugins = setup_plugins(app,
pg_get_timeline=42,
master_lock_changed=[('pluginA', None)],
pg_am_i_replica=True)
assert app.initialize() == None
plugins.reset_mock()
app.master_lock_changed('someone else')
assert app._plugins.mock_calls == [
call.pg_am_i_replica(),
call.master_lock_changed('someone else'),
]
def test_restart_replica(app):
plugins = setup_plugins(app,
pg_am_i_replica=True)
app.initialize()
plugins.reset_mock()
with patch('time.sleep') as sleep:
with patch('sys.exit') as exit:
app.restart(10)
assert exit.called_once_with(0)
assert sleep.called_once_with(10)
assert app._plugins.mock_calls == [
call.pg_am_i_replica(),
call.dcs_disconnect()
]
async def test_replica_reaction_to_master_lock_change(app):
plugins = setup_plugins(app,
pg_get_timeline=42,
pg_am_i_replica=True)
assert app.initialize() == None
plugins.reset_mock()
# if the lock changes owner to someone else, carry on trucking
plugins.reset_mock()
app.master_lock_changed('someone else')
assert app._plugins.mock_calls == [
call.pg_am_i_replica(),
]
assert app._master_lock_owner == 'someone else'
# if the lock is owned by us, er, we stop replication and become the master
plugins.reset_mock()
app.master_lock_changed(app.my_id)
print(app._plugins.mock_calls)
assert app._plugins.mock_calls == [
call.pg_am_i_replica(),
call.pg_stop_replication(),
call.pg_get_timeline(),
call.dcs_set_timeline(42),
]
assert app._master_lock_owner == app.my_id
def test_replica_bootstrap(app):
plugins = setup_plugins(app,
dcs_get_database_identifier='1234',
pg_get_database_identifier='42')
timeout = app.initialize()
assert app._plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# compare our id with the id in the DCS
call.dcs_get_database_identifier(),
call.pg_get_database_identifier(),
# make sure postgresql is stopped
call.pg_stop(),
# postgresql restore
call.pg_restore(),
call.pg_setup_replication(),
call.pg_am_i_replica()
]
# shut down cleanly and immediately
assert timeout == 0
async def test_replica_tries_to_take_over(app):
plugins = setup_plugins(app,
pg_am_i_replica=True)
assert app.initialize() == None
plugins.reset_mock()
# if there is no lock owner, we start looping trying to become master
app.master_lock_changed(None)
assert app._plugins.mock_calls == [call.pg_am_i_replica()]
plugins.reset_mock()
from asyncio import sleep as real_sleep
with patch('asyncio.sleep') as sleep:
sleeper = FakeSleeper()
sleep.side_effect = sleeper
# the first thing is to sleep a bit
await sleeper.next()
assert sleeper.log == [3]
# takeover attempted
await sleeper.next()
assert sleeper.log == [3, 3]
assert app._plugins.mock_calls == [
call.dcs_get_all_state(),
call.dcs_lock('master')]
def test_replica_bootstrap_fails_sanity_test(app):
plugins = setup_plugins(app,
pg_am_i_replica=False,
dcs_get_database_identifier='1234',
pg_get_database_identifier='42')
timeout = app.initialize()
assert app._plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# compare our id with the id in the DCS
call.dcs_get_database_identifier(),
call.pg_get_database_identifier(),
# make sure postgresql is stopped
call.pg_stop(),
# postgresql restore
call.pg_restore(),
call.pg_setup_replication(),
call.pg_am_i_replica(),
call.pg_reset(),
]
# shut down after 5 seconds to try again
assert timeout == 5
