async def test_master_start(app):
plugins = setup_plugins(app,
dcs_get_database_identifier='1234',
dcs_lock=True,
pg_replication_role='master',
pg_get_database_identifier='1234')
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_replication_role(),
# no, so check if there is a master
call.dcs_lock('master'),
# no master, so sure the DB is running
call.pg_start(),
# start monitoring
call.start_monitoring(),
call.dcs_watch(
app.master_lock_changed,
app._notify_state,
app._notify_conn_info,
),
call.get_conn_info(),
# set our first state
call.dcs_set_state({
'host': '127.0.0.1',
'replication_role': 'master',
'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({
'host': '127.0.0.1',
'replication_role': 'master',
'health_problems': {}}),
call.pg_replication_role(),
call.dcs_lock('master'),
call.dcs_set_conn_info({'host': '127.0.0.1'}),
]
def test_master_bootstrap(app):
plugins = setup_plugins(app,
dcs_get_database_identifier=None,
dcs_lock=True,
pg_get_database_identifier='42')
timeout = app.initialize()
assert plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# check if we have a db identifier set
call.dcs_get_database_identifier(),
# no, ok, init our db
call.pg_initdb(),
# make sure it starts
call.pg_start(),
call.pg_get_database_identifier(),
# lock the database identifier so no-one else gets here
call.dcs_lock('database_identifier'),
# while locked make sure there is no id set in the DCS before we got the lock
call.dcs_get_database_identifier(),
# Make the first backup while locked with no DCS
call.pg_backup(),
# set the database identifier AFTER
call.dcs_set_database_identifier('42')
]
# shut down cleanly and immediately
assert timeout == 0
async def test_replica_tries_to_take_over(app):
plugins = setup_plugins(app,
pg_replication_role='replica')
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 plugins.mock_calls == [call.pg_replication_role(), call.master_lock_changed(None)]
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]
assert plugins.mock_calls == []
# takeover attempted
states = [(app.my_id, {'willing': 99.0}), (app.my_id, {'willing': 100.0})]
plugins.dcs_list_state.return_value = states
await sleeper.next()
assert sleeper.log == [3, 3]
assert plugins.mock_calls == [
call.dcs_list_state(),
call.best_replicas([('42', {'willing': 99.0}), ('42', {'willing': 100.0})]),
call.dcs_lock('master')]
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_master_bootstrap(app):
plugins = setup_plugins(app,
dcs_get_database_identifier=None,
dcs_lock=True,
pg_get_database_identifier='42')
timeout = app.initialize()
assert plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# check if we have a db identifier set
call.dcs_get_database_identifier(),
# no, ok, init our db
call.pg_initdb(),
# make sure it starts
call.pg_start(),
call.pg_get_database_identifier(),
# lock the database identifier so no-one else gets here
call.dcs_lock('database_identifier'),
# while locked make sure there is no id set in the DCS before we got the lock
call.dcs_get_database_identifier(),
# Make the first backup while locked with no DCS
call.pg_backup(),
# set the database identifier AFTER
call.dcs_set_database_identifier('42')
]
# shut down cleanly and immediately
assert timeout == 0
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_master_boostrap_fails_to_lock_db_id(app):
plugins = setup_plugins(app,
dcs_get_database_identifier=None,
dcs_lock=False,
pg_get_database_identifier='42')
timeout = app.initialize()
assert plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# check if we have a db identifier set
call.dcs_get_database_identifier(),
# no, ok, init our db
call.pg_initdb(),
# make sure it starts
call.pg_start(),
call.pg_get_database_identifier(),
# lock the database identifier so no-one else gets here
call.dcs_lock('database_identifier')
]
# shut down cleanly
assert timeout == 5
def test_master_boostrap_fails_to_lock_db_id(app):
plugins = setup_plugins(app,
dcs_get_database_identifier=None,
dcs_lock=False,
pg_get_database_identifier='42')
timeout = app.initialize()
assert plugins.mock_calls == [
call.initialize(),
call.get_my_id(),
# check if we have a db identifier set
call.dcs_get_database_identifier(),
# no, ok, init our db
call.pg_initdb(),
# make sure it starts
call.pg_start(),
call.pg_get_database_identifier(),
# lock the database identifier so no-one else gets here
call.dcs_lock('database_identifier')
]
# shut down cleanly
assert timeout == 5
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')]
async def test_replica_tries_to_take_over(app):
plugins = setup_plugins(app, pg_replication_role='replica')
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 plugins.mock_calls == [
call.pg_replication_role(),
call.master_lock_changed(None)
]
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]
assert plugins.mock_calls == []
# takeover attempted
states = [(app.my_id, {
'willing': 99.0
}), (app.my_id, {
'willing': 100.0
})]
plugins.dcs_list_state.return_value = states
await sleeper.next()
assert sleeper.log == [3, 3]
assert plugins.mock_calls == [
call.dcs_list_state(),
call.best_replicas([('42', {
'willing': 99.0
}), ('42', {
'willing': 100.0
})]),
call.dcs_lock('master')
]
async def test_master_start(app):
plugins = setup_plugins(app,
dcs_get_database_identifier='1234',
dcs_lock=True,
pg_replication_role='master',
pg_get_database_identifier='1234')
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_replication_role(),
# no, so check if there is a master
call.dcs_lock('master'),
# no master, so sure the DB is running
call.pg_start(),
# start monitoring
call.start_monitoring(),
call.dcs_watch(
app.master_lock_changed,
app._notify_state,
app._notify_conn_info,
),
call.get_conn_info(),
# set our first state
call.dcs_set_state({
'host': '127.0.0.1',
'replication_role': 'master',
'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({
'host': '127.0.0.1',
'replication_role': 'master',
'health_problems': {}
}),
call.pg_replication_role(),
call.dcs_lock('master'),
call.dcs_set_conn_info({'host': '127.0.0.1'}),
]