Improve unittests and make minor bugfixes.

In particular, remove restore.py in favor of
wale_restore.py, fix minor bugs in the latter
and add unit tests.
This commit is contained in:
Oleksii Kliukin
2015-11-24 15:21:47 +01:00
parent e625c33bef
commit 35efd36c5c
6 changed files with 147 additions and 345 deletions
+6 -2
View File
@@ -229,8 +229,12 @@ class Postgresql:
@staticmethod
def build_connstring(conn):
"""
>>> Postgresql.build_connstring({'host': '127.0.0.1', 'port': '5432'}) == 'host=127.0.0.1 port=5432 '
True
"""
mconn = ""
for param, val in conn.items():
for param, val in sorted(conn.items()):
mconn = mconn + "{0}={1} ".format(param, val)
return mconn
@@ -282,7 +286,7 @@ class Postgresql:
if ret == 0:
break
except Exception as e:
logger.exception('Error creating replica using method {0}: {1}'.format(replica_method, e.str))
logger.exception('Error creating replica using method {0}: {1}'.format(replica_method, str(e)))
ret = 1
return ret
-216
View File
@@ -1,216 +0,0 @@
#!/usr/bin/env python
# arguments are:
# - cluster scope
# - cluster role
# - master connection string
# for the AWS, the folliowing environment variables should be defined:
# - WALE_ENV_DIR: directory where WAL-E environment is kept
# - WAL_S3_BUCKET: a name of the S3 bucket for WAL-E
# - WALE_BACKUP_THRESHOLD_MEGABYTES if WAL amount is above that - use pg_basebackup
# - WALE_BACKUP_THRESHOLD_PERCENTAGE if WAL size exceeds a certain percentage of the
# latest backup size
from collections import namedtuple
import logging
import os
import psycopg2
import subprocess
import sys
if sys.hexversion >= 0x03000000:
long = int
logger = logging.getLogger(__name__)
class Restore(object):
def __init__(self, scope, role, datadir, connstring, env=None):
self.scope = scope
self.role = role
self.master_connection = Restore.parse_connstring(connstring)
self.data_dir = datadir
self.env = os.environ.copy() if not env else env
@staticmethod
def parse_connstring(connstring):
# the connection string is in the form host= port= user=
# return the dictionary with all components as separare keys
result = {}
if connstring:
for x in connstring.split():
if x and '=' in x:
key, val = x.split('=')
result[key.strip()] = val.strip()
return result
def setup(self):
pass
def replica_method(self):
return self.create_replica_with_pg_basebackup
def replica_fallback_method(self):
return None
def run(self):
""" creates a new replica using either pg_basebackup or WAL-E """
method_fn = self.replica_method()
ret = method_fn() if method_fn else 1
if ret != 0 and self.replica_fallback_method() is not None:
ret = (self.replica_fallback_method())()
return ret
def create_replica_with_pg_basebackup(self):
try:
ret = subprocess.call(['pg_basebackup', '-R', '-D',
self.data_dir, '--host=' + self.master_connection['host'],
'--port=' + str(self.master_connection['port']),
'-U', self.master_connection['user']],
env=self.env)
except Exception as e:
logger.error('Error when fetching backup with pg_basebackup: {0}'.format(e))
return 1
return ret
class WALERestore(Restore):
def __init__(self, scope, role, datadir, connstring, env=None):
super(WALERestore, self).__init__(scope, role, datadir, connstring, env)
# check the environment variables
self.init_error = False
def setup(self):
if (self.env.get('WAL_S3_BUCKET') and
self.env.get('WALE_BACKUP_THRESHOLD_PERCENTAGE') and
self.env.get('WALE_BACKUP_THRESHOLD_MEGABYTES')) is None:
self.init_error = True
else:
self.wal_e = namedtuple('WALE',
'threshold_megabytes threshold_backup_size_percentage s3_bucket cmd dir env_file')
self.wal_e.dir = self.env.get('WALE_ENV_DIR', '/home/postgres/etc/wal-e.d/env')
self.wal_e.env_file = os.path.join(self.wal_e.dir, 'WALE_S3_PREFIX')
self.wal_e.cmd = 'envdir {} wal-e --aws-instance-profile '.\
format(self.wal_e.dir)
self.wal_e.s3_bucket = self.env['WAL_S3_BUCKET']
self.wal_e.threshold_megabytes = self.env['WALE_BACKUP_THRESHOLD_MEGABYTES']
self.wal_e.threshold_backup_size_percentage = self.env['WALE_BACKUP_THRESHOLD_PERCENTAGE']
# check that the env file exists, create it otherwise
try:
if not os.path.exists(self.wal_e.dir):
os.makedirs(self.wal_e.dir)
# if this is a directory - make sure we have full access there
elif not (os.path.isdir(self.wal_e.dir) and os.access(self.wal_e.dir, os.R_OK | os.W_OK | os.X_OK)):
logger.error("Unable to access {} or not a directory".format(self.wal_e.dir))
self.init_error = True
# if WAL_S3_PREFIX is not there - create it and write the full path to bucket
if not self.init_error and not os.path.exists(self.wal_e.env_file):
with open(self.wal_e.env_file, 'w') as f:
f.write("s3://{0}/spilo/{1}/wal/\n".format(self.wal_e.s3_bucket, self.scope))
except (os.error, IOError) as e:
logger.error("{0}: WAL-e archiving is disabled".format(e))
self.init_error = True
def replica_method(self):
if self.should_use_s3_to_create_replica():
return self.create_replica_with_s3
return None
def replica_fallback_method(self):
return self.create_replica_with_pg_basebackup
def should_use_s3_to_create_replica(self):
""" determine whether it makes sense to use S3 and not pg_basebackup """
if self.init_error:
return False
threshold_megabytes = self.wal_e.threshold_megabytes
threshold_backup_size_percentage = self.wal_e.threshold_backup_size_percentage
try:
latest_backup = subprocess.check_output(self.wal_e.cmd.split() + ['backup-list', '--detail', 'LATEST'],
env=self.env)
# name last_modified expanded_size_bytes wal_segment_backup_start wal_segment_offset_backup_start
# wal_segment_backup_stop wal_segment_offset_backup_stop
# base_00000001000000000000007F_00000040 2015-05-18T10:13:25.000Z
# 20310671 00000001000000000000007F 00000040
# 00000001000000000000007F 00000240
backup_strings = latest_backup.splitlines() if latest_backup else ()
if len(backup_strings) != 2:
return False
names = backup_strings[0].split()
vals = backup_strings[1].split()
if (len(names) != len(vals)) or (len(names) != 7):
return False
backup_info = dict(zip(names, vals))
except subprocess.CalledProcessError as e:
logger.error("could not query wal-e latest backup: {}".format(e))
return False
try:
backup_size = backup_info['expanded_size_bytes']
backup_start_segment = backup_info['wal_segment_backup_start']
backup_start_offset = backup_info['wal_segment_offset_backup_start']
except Exception as e:
logger.error("unable to get some of S3 backup parameters: {}".format(e))
return False
# WAL filename is XXXXXXXXYYYYYYYY000000ZZ, where X - timeline, Y - LSN logical log file,
# ZZ - 2 high digits of LSN offset. The rest of the offset is the provided decimal offset,
# that we have to convert to hex and 'prepend' to the high offset digits.
lsn_segment = backup_start_segment[8:16]
# first 2 characters of the result are 0x and the last one is L
lsn_offset = hex((long(backup_start_segment[16:32], 16) << 24) + long(backup_start_offset))[2:-1]
# construct the LSN from the segment and offset
backup_start_lsn = '{}/{}'.format(lsn_segment, lsn_offset)
conn = None
cursor = None
diff_in_bytes = long(backup_size)
try:
# get the difference in bytes between the current WAL location and the backup start offset
conn = psycopg2.connect(**(self.master_connection))
conn.autocommit = True
cursor = conn.cursor()
cursor.execute("SELECT pg_xlog_location_diff(pg_current_xlog_location(), %s)", (backup_start_lsn,))
diff_in_bytes = long(cursor.fetchone()[0])
except psycopg2.Error as e:
logger.error('could not determine difference with the master location: {}'.format(e))
return False
finally:
cursor and cursor.close()
conn and conn.close()
# if the size of the accumulated WAL segments is more than a certan percentage of the backup size
# or exceeds the pre-determined size - pg_basebackup is chosen instead.
return (diff_in_bytes < long(threshold_megabytes) * 1048576) and\
(diff_in_bytes < long(backup_size) * float(threshold_backup_size_percentage) / 100)
def create_replica_with_s3(self):
if self.init_error:
return 1
try:
ret = subprocess.call(self.wal_e.cmd + ' backup-fetch {} LATEST'.format(self.data_dir), env=self.env)
except Exception as e:
logger.error('Error when fetching backup with WAL-E: {0}'.format(e))
return 1
return ret
if __name__ == '__main__':
if len(sys.argv) == 5:
# scope, role, datadir, connstring
restore = WALERestore(*(sys.argv[1:]))
restore.setup()
sys.exit(restore.run())
sys.exit("Usage: {0} scope role datadir connstring".format(sys.argv[0]))
+4 -16
View File
@@ -50,27 +50,18 @@ class WALERestore(object):
self.wal_e.dir = env_dir
self.wal_e.threshold_mb = threshold_mb
self.wal_e.threshold_pct = threshold_pct
if use_iam == 1:
self.wal_e.iam_string = ' --aws-instance-profile '
else:
self.wal_e.iam_string = ''
if not os.path.exists(self.wal_e.dir):
self.init_error = True
else:
self.init_error = False
self.wal_e.cmd = 'envdir {0} wal-e {1} '.\
format(self.wal_e.dir, self.wal_e.iam_string)
self.wal_e.iam_string = ' --aws-instance-profile ' if use_iam == 1 else ''
self.wal_e.cmd = 'envdir {0} wal-e {1} '.format(self.wal_e.dir, self.wal_e.iam_string)
self.init_error = (not os.path.exists(self.wal_e.dir))
def run(self):
""" creates a new replica using WAL-E """
if self.should_use_s3_to_create_replica():
if not self.init_error and self.should_use_s3_to_create_replica():
return self.create_replica_with_s3()
return 2
def should_use_s3_to_create_replica(self):
""" determine whether it makes sense to use S3 and not pg_basebackup """
if self.init_error:
return False
threshold_megabytes = self.wal_e.threshold_mb
threshold_backup_size_percentage = self.wal_e.threshold_pct
@@ -138,8 +129,6 @@ class WALERestore(object):
(diff_in_bytes < long(backup_size) * float(threshold_backup_size_percentage) / 100)
def create_replica_with_s3(self):
if self.init_error:
return 1
# if we're set up, restore the replica using fetch latest
try:
ret = subprocess.call(self.wal_e.cmd.split() + ['backup-fetch', '{}'.format(self.data_dir), 'LATEST'])
@@ -151,7 +140,6 @@ class WALERestore(object):
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Script to image replicas using WAL-E')
parser.add_argument('--scope', required=True)
parser.add_argument('--role', required=False)
+10
View File
@@ -284,6 +284,16 @@ class TestPostgresql(unittest.TestCase):
with patch('subprocess.call', Mock(side_effect=[Exception(), 0])):
self.assertEquals(self.p.create_replica(self.leader, ''), 0)
self.p.config['create_replica_method'] = 'wale, basebackup'
self.p.config['wale'] = {'command': 'foo'}
with patch('subprocess.call', Mock(return_value=0)):
self.assertEquals(self.p.create_replica(self.leader, ''), 0)
del self.p.config['wale']
self.assertEquals(self.p.create_replica(self.leader, ''), 0)
with patch('subprocess.call', Mock(side_effect=Exception("foo"))):
self.assertEquals(self.p.create_replica(self.leader, ''), 1)
def test_create_connection_users(self):
cfg = self.p.config
cfg['superuser']['username'] = 'test'
-111
View File
@@ -1,111 +0,0 @@
import unittest
from mock import MagicMock, patch
import os
from patroni.scripts.restore import Restore, WALERestore
def fake_cursor_fetchone(*args, **kwargs):
return ('16777216',)
def fake_call_fail_for_wal_e(*args, **kwargs):
if len(args) > 0 and 'backup-fetch' in args[0]:
return 1
return 0
def fake_call_fail_for_base_backup(*args, **kwargs):
if len(args) > 0 and 'backup-fetch' in args[0]:
return 0
return 1
def fake_backup_data(self, *args, **kwargs):
""" return the fake result of WAL-E backup-list"""
return """name last_modified expanded_size_bytes wal_segment_backup_start wal_segment_offset_backup_start wal_segment_backup_stop wal_segment_offset_backup_stop
base_00000001000000000000007F_00000040 2015-05-18T10:13:25.000Z 167772160 00000001000000000000007F 00000040 00000001000000000000007F 00000240
"""
class TestRestore(unittest.TestCase):
def setUp(self):
self.restore = Restore("batman", "master", "/data", "host=batman port=5432 user=batman")
pass
def tearDown(self):
pass
def test_parse_connstring(self):
self.assertDictEqual(self.restore.master_connection, {'host': 'batman', 'port': '5432', 'user': 'batman'})
@patch('subprocess.call', MagicMock(return_value=0))
def test_run(self):
ret = self.restore.run()
self.assertEqual(ret, 0)
@patch('subprocess.call', MagicMock(return_value=1))
def test_run_fail(self):
ret = self.restore.run()
self.assertEqual(ret, 1)
@patch('os.access', MagicMock(return_value=True))
@patch('os.makedirs', MagicMock(return_value=True))
@patch('os.path.exists', MagicMock(return_value=True))
@patch('os.path.isdir', MagicMock(return_value=True))
@patch('psycopg2.extensions.cursor.fetchone', MagicMock(side_effect=fake_cursor_fetchone))
@patch('psycopg2.extensions.cursor', MagicMock(autospec=True))
@patch('psycopg2.extensions.connection', MagicMock(autospec=True))
@patch('psycopg2.connect', MagicMock(autospec=True))
@patch('subprocess.check_output', MagicMock(side_effect=fake_backup_data))
class TestWALERestore(unittest.TestCase):
def setUp(self):
env = {}
env['WAL_S3_BUCKET'] = 'batman'
env['WALE_BACKUP_THRESHOLD_PERCENTAGE'] = 100
env['WALE_BACKUP_THRESHOLD_MEGABYTES'] = 100
self.wale_restore = WALERestore("batman", "master", "/data", "host=batman port=5432 user=batman", env=env)
def tearDown(self):
pass
def test_setup(self):
self.wale_restore.setup()
self.assertFalse(self.wale_restore.init_error)
# have to redefine the class-level os.access mock inside the function
# since the class-level mock will be applied after the function level one.
@patch('os.access', return_value=False)
def test_setup_fail(self, mock_no_access):
os.access = mock_no_access
self.wale_restore.setup()
self.assertTrue(self.wale_restore.init_error)
# The 3 tests above only differ with the mock function instead of a subprocess call
# in the first one, subprocess call should return success only for wal-e command,
# checking the primary use-case of restoring from WAL-E backup.
# In the second one, we test fallbacks by failing at WAL-E, but succeeding at
# pg_basebackup.
# Finally, the last use case is when all subprocess.call fails. resulting in a
# failure to restore from replica
@patch('subprocess.call',
MagicMock(side_effect=lambda *args, **kwargs: 0 if 'wal-e' in args[0] else 1))
def test_run(self):
self.wale_restore.setup()
ret = self.wale_restore.run()
self.assertEqual(ret, 0)
@patch('subprocess.call',
MagicMock(side_effect=lambda *args, **kwargs: 0 if 'pg_basebackup' in args[0] else 1))
def test_run_fallback(self):
self.wale_restore.setup()
ret = self.wale_restore.run()
self.assertEqual(ret, 0)
@patch('subprocess.call', MagicMock(return_value=1))
def test_run_all_fail(self):
self.wale_restore.setup()
ret = self.wale_restore.run()
self.assertEqual(ret, 1)
+127
View File
@@ -0,0 +1,127 @@
import unittest
from mock import MagicMock, patch, PropertyMock
import os
import psycopg2
import subprocess
from patroni.scripts.wale_restore import WALERestore
def fake_cursor_fetchone(*args, **kwargs):
return ('16777216',)
def fake_call_fail_for_wal_e(*args, **kwargs):
if len(args) > 0 and 'backup-fetch' in args[0]:
return 1
return 0
def fake_call_fail_for_base_backup(*args, **kwargs):
if len(args) > 0 and 'backup-fetch' in args[0]:
return 0
return 1
def fake_backup_data(self, *args, **kwargs):
""" return the fake result of WAL-E backup-list"""
return """name last_modified expanded_size_bytes wal_segment_backup_start wal_segment_offset_backup_start wal_segment_backup_stop wal_segment_offset_backup_stop
base_00000001000000000000007F_00000040 2015-05-18T10:13:25.000Z 167772160 00000001000000000000007F 00000040 00000001000000000000007F 00000240
"""
def fake_backup_data_2(self, *args, **kwargs):
""" return the fake result of WAL-E backup-list"""
return """name last_modified expanded_size_bytes wal_segment_backup_start wal_segment_offset_backup_start wal_segment_backup_stop wal_segment_offset_backup_stop """
def fake_backup_data_3(self, *args, **kwargs):
""" return the fake result of WAL-E backup-list"""
return """name last_modified expanded_size_bytes wal_segment_backup_start wal_segment_offset_backup_start wal_segment_backup_stop
base_00000001000000000000007F_00000040 2015-05-18T10:13:25.000Z 167772160 00000001000000000000007F 00000040 00000001000000000000007F 00000240
"""
def fake_backup_data_4(self, *args, **kwargs):
""" return the fake result of WAL-E backup-list"""
return """name last_modified expanded_size_foo wal_segment_backup_start wal_segment_offset_backup_start wal_segment_backup_stop wal_segment_offset_backup_stop
base_00000001000000000000007F_00000040 2015-05-18T10:13:25.000Z 167772160 00000001000000000000007F 00000040 00000001000000000000007F 00000240
"""
@patch('os.access', MagicMock(return_value=True))
@patch('os.makedirs', MagicMock(return_value=True))
@patch('os.path.exists', MagicMock(return_value=True))
@patch('os.path.isdir', MagicMock(return_value=True))
@patch('psycopg2.extensions.cursor.fetchone', MagicMock(side_effect=fake_cursor_fetchone))
@patch('psycopg2.extensions.cursor', MagicMock(autospec=True))
@patch('psycopg2.extensions.connection', MagicMock(autospec=True))
@patch('psycopg2.connect', MagicMock(autospec=True))
@patch('subprocess.check_output', MagicMock(side_effect=fake_backup_data))
class TestWALERestore(unittest.TestCase):
def setUp(self):
self.wale_restore = WALERestore("batman", "/data",
"host=batman port=5432 user=batman", "/etc", 100, 100, 1)
def tearDown(self):
pass
def test_should_use_s3_to_create_replica(self):
with patch('psycopg2.connect', MagicMock(side_effect=psycopg2.Error("foo"))):
self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
with patch('subprocess.check_output', MagicMock(side_effect=subprocess.CalledProcessError(1, "cmd", "foo"))):
self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
with patch('subprocess.check_output', MagicMock(side_effect=fake_backup_data_2)):
self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
with patch('subprocess.check_output', MagicMock(side_effect=fake_backup_data_3)):
self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
with patch('subprocess.check_output', MagicMock(side_effect=fake_backup_data_4)):
self.assertFalse(self.wale_restore.should_use_s3_to_create_replica())
self.wale_restore.should_use_s3_to_create_replica()
def test_create_replica_with_s3(self):
with patch('subprocess.call', MagicMock(return_value=0)):
self.assertEqual(self.wale_restore.create_replica_with_s3(), 0)
with patch('subprocess.call', MagicMock(side_effect=Exception("foo"))):
self.assertEqual(self.wale_restore.create_replica_with_s3(), 1)
def test_run(self):
with patch.object(self.wale_restore, 'init_error', PropertyMock(return_value=True)):
self.assertEqual(self.wale_restore.run(), 2)
with patch.object(self.wale_restore, 'should_use_s3_to_create_replica', MagicMock(return_value=True)):
with patch.object(self.wale_restore, 'create_replica_with_s3', MagicMock(return_value=0)):
self.assertEqual(self.wale_restore.run(), 0)
# with patch.object(self.wale_restore, 'init_error', PropertyMock(return_value=True)):
# self.assertFalse(self.wale_restore.create_replica_with_s3())
# @patch('subprocess.call', MagicMock(return_value=0))
# def test_run(self):
# ret = self.wale_restore.run()
# self.assertEqual(ret, 0)
# The 3 tests above only differ with the mock function instead of a subprocess call
# in the first one, subprocess call should return success only for wal-e command,
# checking the primary use-case of restoring from WAL-E backup.
# In the second one, we test fallbacks by failing at WAL-E, but succeeding at
# pg_basebackup.
# Finally, the last use case is when all subprocess.call fails. resulting in a
# failure to restore from replica
# @patch('subprocess.call',
# MagicMock(side_effect=lambda *args, **kwargs: 0 if 'wal-e' in args[0] else 1))
# def test_run(self):
# self.wale_restore.setup()
# ret = self.wale_restore.run()
# self.assertEqual(ret, 0)
# @patch('subprocess.call',
# MagicMock(side_effect=lambda *args, **kwargs: 0 if 'pg_basebackup' in args[0] else 1))
# def test_run_fallback(self):
# self.wale_restore.setup()
# ret = self.wale_restore.run()
# self.assertEqual(ret, 0)
# @patch('subprocess.call', MagicMock(return_value=1))
# def test_run_all_fail(self):
# self.wale_restore.setup()
# ret = self.wale_restore.run()
# self.assertEqual(ret, 1)