1
0
mirror of https://github.com/postgrespro/pg_probackup.git synced 2024-11-25 09:01:48 +02:00
pg_probackup/tests/cfs_restore.py
2019-10-19 09:48:35 +03:00

451 lines
15 KiB
Python

"""
restore
Syntax:
pg_probackup restore -B backupdir --instance instance_name
[-D datadir]
[ -i backup_id | [{--time=time | --xid=xid | --lsn=lsn } [--inclusive=boolean]]][--timeline=timeline] [-T OLDDIR=NEWDIR]
[-j num_threads] [--progress] [-q] [-v]
"""
import os
import unittest
import shutil
from .helpers.cfs_helpers import find_by_name
from .helpers.ptrack_helpers import ProbackupTest, ProbackupException
module_name = 'cfs_restore'
tblspace_name = 'cfs_tblspace'
tblspace_name_new = 'cfs_tblspace_new'
class CfsRestoreBase(ProbackupTest, unittest.TestCase):
def setUp(self):
self.fname = self.id().split('.')[3]
self.backup_dir = os.path.join(self.tmp_path, module_name, self.fname, 'backup')
self.node = self.make_simple_node(
base_dir="{0}/{1}/node".format(module_name, self.fname),
set_replication=True,
initdb_params=['--data-checksums'],
pg_options={
# 'ptrack_enable': 'on',
'cfs_encryption': 'off',
}
)
self.init_pb(self.backup_dir)
self.add_instance(self.backup_dir, 'node', self.node)
self.set_archiving(self.backup_dir, 'node', self.node)
self.node.slow_start()
self.create_tblspace_in_node(self.node, tblspace_name, cfs=True)
self.add_data_in_cluster()
self.backup_id = None
try:
self.backup_id = self.backup_node(self.backup_dir, 'node', self.node, backup_type='full')
except ProbackupException as e:
self.fail(
"ERROR: Full backup failed \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
def add_data_in_cluster(self):
pass
def tearDown(self):
self.node.cleanup()
self.del_test_dir(module_name, self.fname)
class CfsRestoreNoencEmptyTablespaceTest(CfsRestoreBase):
# @unittest.expectedFailure
# @unittest.skip("skip")
def test_restore_empty_tablespace_from_fullbackup(self):
"""
Case: Restore empty tablespace from valid full backup.
"""
self.node.stop(["-m", "immediate"])
self.node.cleanup()
shutil.rmtree(self.get_tblspace_path(self.node, tblspace_name))
try:
self.restore_node(self.backup_dir, 'node', self.node, backup_id=self.backup_id)
except ProbackupException as e:
self.fail(
"ERROR: Restore failed. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertTrue(
find_by_name([self.get_tblspace_path(self.node, tblspace_name)], ["pg_compression"]),
"ERROR: Restored data is not valid. pg_compression not found in tablespace dir."
)
try:
self.node.slow_start()
except ProbackupException as e:
self.fail(
"ERROR: Instance not started after restore. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
tblspace = self.node.safe_psql(
"postgres",
"SELECT * FROM pg_tablespace WHERE spcname='{0}'".format(tblspace_name)
)
self.assertTrue(
tblspace_name in tblspace and "compression=true" in tblspace,
"ERROR: The tablespace not restored or it restored without compressions"
)
class CfsRestoreNoencTest(CfsRestoreBase):
def add_data_in_cluster(self):
self.node.safe_psql(
"postgres",
'CREATE TABLE {0} TABLESPACE {1} \
AS SELECT i AS id, MD5(i::text) AS text, \
MD5(repeat(i::text,10))::tsvector AS tsvector \
FROM generate_series(0,1e5) i'.format('t1', tblspace_name)
)
self.table_t1 = self.node.safe_psql(
"postgres",
"SELECT * FROM t1"
)
# --- Restore from full backup ---#
# @unittest.expectedFailure
# @unittest.skip("skip")
def test_restore_from_fullbackup_to_old_location(self):
"""
Case: Restore instance from valid full backup to old location.
"""
self.node.stop()
self.node.cleanup()
shutil.rmtree(self.get_tblspace_path(self.node, tblspace_name))
try:
self.restore_node(self.backup_dir, 'node', self.node, backup_id=self.backup_id)
except ProbackupException as e:
self.fail(
"ERROR: Restore from full backup failed. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertTrue(
find_by_name([self.get_tblspace_path(self.node, tblspace_name)], ['pg_compression']),
"ERROR: File pg_compression not found in tablespace dir"
)
try:
self.node.slow_start()
except ProbackupException as e:
self.fail(
"ERROR: Instance not started after restore. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertEqual(
repr(self.node.safe_psql("postgres", "SELECT * FROM %s" % 't1')),
repr(self.table_t1)
)
# @unittest.expectedFailure
# @unittest.skip("skip")
def test_restore_from_fullbackup_to_old_location_3_jobs(self):
"""
Case: Restore instance from valid full backup to old location.
"""
self.node.stop()
self.node.cleanup()
shutil.rmtree(self.get_tblspace_path(self.node, tblspace_name))
try:
self.restore_node(self.backup_dir, 'node', self.node, backup_id=self.backup_id, options=['-j', '3'])
except ProbackupException as e:
self.fail(
"ERROR: Restore from full backup failed. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertTrue(
find_by_name([self.get_tblspace_path(self.node, tblspace_name)], ['pg_compression']),
"ERROR: File pg_compression not found in backup dir"
)
try:
self.node.slow_start()
except ProbackupException as e:
self.fail(
"ERROR: Instance not started after restore. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertEqual(
repr(self.node.safe_psql("postgres", "SELECT * FROM %s" % 't1')),
repr(self.table_t1)
)
# @unittest.expectedFailure
# @unittest.skip("skip")
def test_restore_from_fullbackup_to_new_location(self):
"""
Case: Restore instance from valid full backup to new location.
"""
self.node.stop()
self.node.cleanup()
shutil.rmtree(self.get_tblspace_path(self.node, tblspace_name))
node_new = self.make_simple_node(base_dir="{0}/{1}/node_new_location".format(module_name, self.fname))
node_new.cleanup()
try:
self.restore_node(self.backup_dir, 'node', node_new, backup_id=self.backup_id)
self.set_auto_conf(node_new, {'port': node_new.port})
except ProbackupException as e:
self.fail(
"ERROR: Restore from full backup failed. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertTrue(
find_by_name([self.get_tblspace_path(self.node, tblspace_name)], ['pg_compression']),
"ERROR: File pg_compression not found in backup dir"
)
try:
node_new.slow_start()
except ProbackupException as e:
self.fail(
"ERROR: Instance not started after restore. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertEqual(
repr(node_new.safe_psql("postgres", "SELECT * FROM %s" % 't1')),
repr(self.table_t1)
)
node_new.cleanup()
# @unittest.expectedFailure
# @unittest.skip("skip")
def test_restore_from_fullbackup_to_new_location_5_jobs(self):
"""
Case: Restore instance from valid full backup to new location.
"""
self.node.stop()
self.node.cleanup()
shutil.rmtree(self.get_tblspace_path(self.node, tblspace_name))
node_new = self.make_simple_node(base_dir="{0}/{1}/node_new_location".format(module_name, self.fname))
node_new.cleanup()
try:
self.restore_node(self.backup_dir, 'node', node_new, backup_id=self.backup_id, options=['-j', '5'])
self.set_auto_conf(node_new, {'port': node_new.port})
except ProbackupException as e:
self.fail(
"ERROR: Restore from full backup failed. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertTrue(
find_by_name([self.get_tblspace_path(self.node, tblspace_name)], ['pg_compression']),
"ERROR: File pg_compression not found in backup dir"
)
try:
node_new.slow_start()
except ProbackupException as e:
self.fail(
"ERROR: Instance not started after restore. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertEqual(
repr(node_new.safe_psql("postgres", "SELECT * FROM %s" % 't1')),
repr(self.table_t1)
)
node_new.cleanup()
# @unittest.expectedFailure
# @unittest.skip("skip")
def test_restore_from_fullbackup_to_old_location_tablespace_new_location(self):
self.node.stop()
self.node.cleanup()
shutil.rmtree(self.get_tblspace_path(self.node, tblspace_name))
os.mkdir(self.get_tblspace_path(self.node, tblspace_name_new))
try:
self.restore_node(
self.backup_dir,
'node', self.node,
backup_id=self.backup_id,
options=["-T", "{0}={1}".format(
self.get_tblspace_path(self.node, tblspace_name),
self.get_tblspace_path(self.node, tblspace_name_new)
)
]
)
except ProbackupException as e:
self.fail(
"ERROR: Restore from full backup failed. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertTrue(
find_by_name([self.get_tblspace_path(self.node, tblspace_name_new)], ['pg_compression']),
"ERROR: File pg_compression not found in new tablespace location"
)
try:
self.node.slow_start()
except ProbackupException as e:
self.fail(
"ERROR: Instance not started after restore. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertEqual(
repr(self.node.safe_psql("postgres", "SELECT * FROM %s" % 't1')),
repr(self.table_t1)
)
# @unittest.expectedFailure
# @unittest.skip("skip")
def test_restore_from_fullbackup_to_old_location_tablespace_new_location_3_jobs(self):
self.node.stop()
self.node.cleanup()
shutil.rmtree(self.get_tblspace_path(self.node, tblspace_name))
os.mkdir(self.get_tblspace_path(self.node, tblspace_name_new))
try:
self.restore_node(
self.backup_dir,
'node', self.node,
backup_id=self.backup_id,
options=["-j", "3", "-T", "{0}={1}".format(
self.get_tblspace_path(self.node, tblspace_name),
self.get_tblspace_path(self.node, tblspace_name_new)
)
]
)
except ProbackupException as e:
self.fail(
"ERROR: Restore from full backup failed. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertTrue(
find_by_name([self.get_tblspace_path(self.node, tblspace_name_new)], ['pg_compression']),
"ERROR: File pg_compression not found in new tablespace location"
)
try:
self.node.slow_start()
except ProbackupException as e:
self.fail(
"ERROR: Instance not started after restore. \n {0} \n {1}".format(
repr(self.cmd),
repr(e.message)
)
)
self.assertEqual(
repr(self.node.safe_psql("postgres", "SELECT * FROM %s" % 't1')),
repr(self.table_t1)
)
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_fullbackup_to_new_location_tablespace_new_location(self):
pass
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_fullbackup_to_new_location_tablespace_new_location_5_jobs(self):
pass
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_ptrack(self):
"""
Case: Restore from backup to old location
"""
pass
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_ptrack_jobs(self):
"""
Case: Restore from backup to old location, four jobs
"""
pass
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_ptrack_new_jobs(self):
pass
# --------------------------------------------------------- #
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_page(self):
"""
Case: Restore from backup to old location
"""
pass
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_page_jobs(self):
"""
Case: Restore from backup to old location, four jobs
"""
pass
# @unittest.expectedFailure
@unittest.skip("skip")
def test_restore_from_page_new_jobs(self):
"""
Case: Restore from backup to new location, four jobs
"""
pass
#class CfsRestoreEncEmptyTablespaceTest(CfsRestoreNoencEmptyTablespaceTest):
# # --- Begin --- #
# def setUp(self):
# os.environ["PG_CIPHER_KEY"] = "super_secret_cipher_key"
# super(CfsRestoreNoencEmptyTablespaceTest, self).setUp()
#
#
#class CfsRestoreEncTest(CfsRestoreNoencTest):
# # --- Begin --- #
# def setUp(self):
# os.environ["PG_CIPHER_KEY"] = "super_secret_cipher_key"
# super(CfsRestoreNoencTest, self).setUp()