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_replication_role='master')
# 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_replication_role(),
# no, so check if there is a master
call.dcs_lock('master'),
call.dcs_get_lock_owner('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
async def test_master_lock_changes_owner(app):
# if the lock changes owner to someone else, shutdown postgresql and exist
plugins = setup_plugins(app,
pg_replication_role='master')
assert app.initialize() == None
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 plugins.mock_calls == [
call.pg_replication_role(),
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect(),
call.master_lock_changed('someone else')
]
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 plugins.mock_calls == [
call.pg_replication_role(),
call.master_lock_changed('42')
]
assert app._master_lock_owner == app.my_id
async def test_master_lock_changes_owner(app):
# if the lock changes owner to someone else, shutdown postgresql and exist
plugins = setup_plugins(app, pg_replication_role='master')
assert app.initialize() == None
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 plugins.mock_calls == [
call.pg_replication_role(),
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect(),
call.master_lock_changed('someone else')
]
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 plugins.mock_calls == [
call.pg_replication_role(),
call.master_lock_changed('42')
]
assert app._master_lock_owner == app.my_id
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_replication_role='master')
# 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_replication_role(),
# no, so check if there is a master
call.dcs_lock('master'),
call.dcs_get_lock_owner('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_restart_master(app, event_loop):
plugins = setup_plugins(app, pg_replication_role='master')
app.initialize()
plugins.reset_mock()
with patch('time.sleep') as sleep:
app.restart(10)
assert sleep.called_once_with(10)
event_loop.run_forever() # must be stopped by restart()
assert plugins.mock_calls == [
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect()
]
def test_restart_master(app, event_loop):
plugins = setup_plugins(app,
pg_replication_role='master')
app.initialize()
plugins.reset_mock()
with patch('time.sleep') as sleep:
app.restart(10)
assert sleep.called_once_with(10)
event_loop.run_forever() # must be stopped by restart()
assert plugins.mock_calls == [
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect()
]
def test_restart_master(app):
plugins = setup_plugins(app,
pg_replication_role='master')
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_replication_role(),
call.pg_stop(),
call.dcs_disconnect()
]
async def test_master_lock_broken(app):
plugins = setup_plugins(app, pg_replication_role='master')
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 plugins.mock_calls == [
call.pg_replication_role(),
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect(),
call.master_lock_changed(None)
]
assert app._master_lock_owner == None
async def test_master_unhealthy(app):
plugins = setup_plugins(app, pg_replication_role='master')
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',
'replication_role': 'master',
'health_problems': {
'boom': {
'reason': 'It went Boom',
'can_be_replica': True
}
}
}),
call.pg_replication_role(),
call.dcs_delete_conn_info(),
]
plugins.reset_mock()
# now we should have _handle_unhealthy_master running
with patch('asyncio.sleep') as sleep, patch(
'zgres.deadman.App._stop') as exit, patch(
'time.sleep') as blocking_sleep:
sleeper = FakeSleeper()
sleep.side_effect = sleeper
exit.side_effect = lambda: sleeper.finish()
# there is no replica, so we just sleep and ping the
# DCS to find a willing replica
states = [iter([])]
plugins.dcs_list_state.side_effect = states
await sleeper.next()
assert plugins.mock_calls == [call.dcs_list_state()]
# we add a willing replica
states = [iter([('other', {'willing': 1})])]
plugins.dcs_list_state.side_effect = states
plugins.reset_mock()
await sleeper.next()
assert plugins.mock_calls == [
call.dcs_list_state(),
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect()
]
async def test_master_lock_broken(app):
plugins = setup_plugins(app,
pg_replication_role='master')
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 plugins.mock_calls == [
call.pg_replication_role(),
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect(),
call.master_lock_changed(None)
]
assert app._master_lock_owner == None
async def test_master_unhealthy(app):
plugins = setup_plugins(app,
pg_replication_role='master')
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',
'replication_role': 'master',
'health_problems': {'boom': {'reason': 'It went Boom', 'can_be_replica': True}}}),
call.pg_replication_role(),
call.dcs_delete_conn_info(),
]
plugins.reset_mock()
# now we should have _handle_unhealthy_master running
with patch('asyncio.sleep') as sleep, patch('sys.exit') as exit, patch('time.sleep') as blocking_sleep:
sleeper = FakeSleeper()
sleep.side_effect = sleeper
exit.side_effect = lambda x: sleeper.finish()
# there is no replica, so we just sleep and ping the
# DCS to find a willing replica
states = [iter([])]
plugins.dcs_get_all_state.side_effect = states
plugins.willing_replicas.side_effect = [iter([])]
await sleeper.next()
assert plugins.mock_calls == [
call.dcs_get_all_state(),
call.willing_replicas(states[0])]
# we add a willing replica
states = [iter([])]
plugins.dcs_get_all_state.side_effect = states
plugins.willing_replicas.side_effect = [iter([('other', {})])]
plugins.reset_mock()
await sleeper.next()
assert plugins.mock_calls == [
call.dcs_get_all_state(),
call.willing_replicas(states[0]),
call.pg_replication_role(),
call.pg_stop(),
call.dcs_disconnect()
]
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
def test_replica_bootstrap(app):
plugins = setup_plugins(app, dcs_get_database_identifier='1234')
plugins.pg_get_database_identifier.side_effect = ['42', '1234']
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(),
# make sure postgresql is stopped
call.pg_stop(),
# postgresql restore
call.pg_initdb(),
call.pg_restore(),
call.pg_setup_replication(None),
call.pg_get_database_identifier(),
call.pg_replication_role()
]
# shut down cleanly and immediately
assert timeout == 0
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
def test_replica_bootstrap_fails_sanity_test(app):
plugins = setup_plugins(app,
pg_replication_role='master',
dcs_get_database_identifier='1234',
pg_get_database_identifier='42')
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(),
# make sure postgresql is stopped
call.pg_stop(),
# postgresql restore
call.pg_initdb(),
call.pg_restore(),
call.pg_setup_replication(None),
call.pg_get_database_identifier(),
call.pg_replication_role(),
call.pg_reset(),
]
# shut down after 5 seconds to try again
assert timeout == 5
