def backup_check(self, check_value, expected_ts_log, expected_ts_nolog):
newdir = "BACKUP"
copy_wiredtiger_home('.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
cur_ts_log = session.open_cursor(self.table_ts_log)
cur_ts_nolog = session.open_cursor(self.table_ts_nolog)
# Count how many times the given value is present in the
# logged timestamp table.
actual_ts_log = 0
for k, v in cur_ts_log:
if check_value == v:
actual_ts_log += 1
cur_ts_log.close()
# Count how many times the given value is present in the
# not logged timestamp table
actual_ts_nolog = 0
for k, v in cur_ts_nolog:
if check_value == v:
actual_ts_nolog += 1
cur_ts_nolog.close()
conn.close()
self.assertEqual(actual_ts_log, expected_ts_log)
self.assertEqual(actual_ts_nolog, expected_ts_nolog)
def test_bug018(self):
'''Test closing multiple tables'''
self.close_conn()
subdir = 'SUBPROCESS'
[ignore_result, new_home_dir] = self.run_subprocess_function(
subdir, 'test_bug018.test_bug018.subprocess_bug018')
# Make a backup for forensics in case something goes wrong.
backup_dir = 'BACKUP'
copy_wiredtiger_home(self, new_home_dir, backup_dir, True)
# After reopening and running recovery both tables should be in
# sync even though table 1 was successfully written and table 2
# had an error on close.
self.open_conn(new_home_dir)
results1 = list(self.session.open_cursor(self.uri1))
# It's possible the second table can't even be opened.
# That can happen only if the root page was not pushed out.
# We can't depend on the text of a particular error message to be
# emitted, so we'll just ignore the error.
self.captureerr.check(self) # check there is no error output so far
try:
results2 = list(self.session.open_cursor(self.uri2))
except:
# Make sure there's some error, but we don't care what.
self.captureerr.checkAdditionalPattern(self, '.')
results2 = []
self.assertEqual(results1, results2)
def test_compress02(self):
ds = SimpleDataSet(self,
self.uri,
0,
key_format="S",
value_format="S",
config='block_compressor=zstd')
ds.populate()
valuea = "aaaaa" * 100
cursor = self.session.open_cursor(self.uri)
self.large_updates(self.uri, valuea, ds, self.nrows)
self.check(valuea, self.uri, self.nrows)
self.session.checkpoint()
#Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(self, ".", "RESTART")
# Close the connection and reopen it with a different zstd compression level configuration.
restart_config = 'builtin_extension_config={zstd={compression_level=9}},cache_size=10MB'
self.close_conn()
self.reopen_conn("RESTART", restart_config)
# Open the new directory.
self.session = self.setUpSessionOpen(self.conn)
# Check the table contains the last checkpointed value.
self.check(valuea, self.uri, self.nrows)
def test_modify_smoke_recover(self):
# Close the original database.
self.conn.close()
# Open a new database with logging configured.
self.conn_config = \
'log=(enabled=true),transaction_sync=(method=dsync,enabled)'
self.conn = self.setUpConnectionOpen(".")
self.session = self.setUpSessionOpen(self.conn)
# Populate a database, and checkpoint it so it exists after recovery.
ds = SimpleDataSet(self,
self.uri, 100, key_format=self.keyfmt, value_format=self.valuefmt)
ds.populate()
self.session.checkpoint()
self.modify_load(ds, False)
# Crash and recover in a new directory.
newdir = 'RESTART'
copy_wiredtiger_home('.', newdir)
self.conn.close()
self.conn = self.setUpConnectionOpen(newdir)
self.session = self.setUpSessionOpen(self.conn)
self.session.verify(self.uri)
self.modify_confirm(ds, False)
def backup_check(self, check_value, expected_ts_log, expected_ts_nolog):
newdir = "BACKUP"
copy_wiredtiger_home('.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
cur_ts_log = session.open_cursor(self.table_ts_log)
cur_ts_nolog = session.open_cursor(self.table_ts_nolog)
# Count how many times the given value is present in the
# logged timestamp table.
actual_ts_log = 0
for k, v in cur_ts_log:
if check_value == v:
actual_ts_log += 1
cur_ts_log.close()
# Count how many times the given value is present in the
# not logged timestamp table
actual_ts_nolog = 0
for k, v in cur_ts_nolog:
if check_value == v:
actual_ts_nolog += 1
cur_ts_nolog.close()
conn.close()
self.assertEqual(actual_ts_log, expected_ts_log)
self.assertEqual(actual_ts_nolog, expected_ts_nolog)
def check_unclean(self):
backup = "WT_COPYDIR"
copy_wiredtiger_home(self, self.home, backup, True)
msg = '/needs recovery/'
# 2. an unclean shutdown and reopening readonly
self.assertRaisesWithMessage(wiredtiger.WiredTigerError,
lambda: self.wiredtiger_open(backup, self.conn_params_rd), msg)
def test_checkpoint_snapshot(self):
ds = SimpleDataSet(self,
self.uri,
self.nrows,
key_format="S",
value_format='u')
ds.populate()
value = b"aaaaa" * 100
sessions = [0] * self.nsessions
cursors = [0] * self.nsessions
for j in range(0, self.nsessions):
sessions[j] = self.conn.open_session()
cursors[j] = sessions[j].open_cursor(self.uri)
sessions[j].begin_transaction('isolation=snapshot')
start = (j * self.nkeys)
end = start + self.nkeys
for i in range(start, end):
cursors[j].set_key(ds.key(self.nrows + i))
cursors[j].set_value(value)
self.assertEquals(cursors[j].insert(), 0)
session_p2 = self.conn.open_session()
session_p2.checkpoint()
#Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(".", "RESTART")
# Open the new directory.
self.conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(self.conn)
def check_unclean(self):
backup = "WT_COPYDIR"
copy_wiredtiger_home(self.home, backup, True)
msg = '/needs recovery/'
# 2. an unclean shutdown and reopening readonly
self.assertRaisesWithMessage(wiredtiger.WiredTigerError,
lambda: self.wiredtiger_open(backup, self.conn_params_rd), msg)
def test_bug018(self):
'''Test closing multiple tables'''
self.close_conn()
subdir = 'SUBPROCESS'
[ignore_result, new_home_dir] = self.run_subprocess_function(
subdir, 'test_bug018.test_bug018.subprocess_bug018')
# Make a backup for forensics in case something goes wrong.
backup_dir = 'BACKUP'
copy_wiredtiger_home(new_home_dir, backup_dir, True)
# After reopening and running recovery both tables should be in
# sync even though table 1 was successfully written and table 2
# had an error on close.
self.open_conn(new_home_dir)
results1 = list(self.session.open_cursor(self.uri1))
# It's possible the second table can't even be opened.
# That can happen only if the root page was not pushed out.
# So if we get an error, make sure we're getting the right
# error message.
self.captureerr.check(self) # check error messages until now
try:
results2 = list(self.session.open_cursor(self.uri2))
except:
self.captureerr.checkAdditionalPattern(self,
'unable to read root page')
results2 = []
self.assertEqual(results1, results2)
def test_modify_smoke_recover(self):
# Close the original database.
self.conn.close()
# Open a new database with logging configured.
self.conn_config = \
'log=(enabled=true),transaction_sync=(method=dsync,enabled)'
self.conn = self.setUpConnectionOpen(".")
self.session = self.setUpSessionOpen(self.conn)
# Populate a database, and checkpoint it so it exists after recovery.
ds = SimpleDataSet(self,
self.uri, 100, key_format=self.keyfmt, value_format=self.valuefmt)
ds.populate()
self.session.checkpoint()
self.modify_load(ds, False)
# Crash and recover in a new directory.
newdir = 'RESTART'
copy_wiredtiger_home('.', newdir)
self.conn.close()
self.conn = self.setUpConnectionOpen(newdir)
self.session = self.setUpSessionOpen(self.conn)
self.session.verify(self.uri)
self.modify_confirm(ds, False)
def test_checkpoint_snapshot(self):
ds = SimpleDataSet(self, self.uri, 0, key_format="S", value_format="S",config='log=(enabled=false)')
ds.populate()
valuea = "aaaaa" * 100
valueb = "bbbbb" * 100
valuec = "ccccc" * 100
valued = "ddddd" * 100
cursor = self.session.open_cursor(self.uri)
self.large_updates(self.uri, valuea, ds, self.nrows)
self.check(valuea, self.uri, self.nrows)
session1 = self.conn.open_session()
session1.begin_transaction()
cursor1 = session1.open_cursor(self.uri)
for i in range(self.nrows, self.nrows*2):
cursor1.set_key(ds.key(i))
cursor1.set_value(valuea)
self.assertEqual(cursor1.insert(), 0)
# Create a checkpoint thread
done = threading.Event()
ckpt = checkpoint_thread(self.conn, done)
try:
ckpt.start()
# Sleep for sometime so that checkpoint starts before committing last transaction.
time.sleep(2)
session1.commit_transaction()
finally:
done.set()
ckpt.join()
#Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(self, ".", "RESTART")
# Open the new directory.
self.conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(self.conn)
# Check the table contains the last checkpointed value.
self.check(valuea, self.uri, self.nrows)
stat_cursor = self.session.open_cursor('statistics:', None, None)
inconsistent_ckpt = stat_cursor[stat.conn.txn_rts_inconsistent_ckpt][2]
keys_removed = stat_cursor[stat.conn.txn_rts_keys_removed][2]
keys_restored = stat_cursor[stat.conn.txn_rts_keys_restored][2]
pages_visited = stat_cursor[stat.conn.txn_rts_pages_visited][2]
upd_aborted = stat_cursor[stat.conn.txn_rts_upd_aborted][2]
stat_cursor.close()
self.assertGreater(inconsistent_ckpt, 0)
self.assertEqual(upd_aborted, 0)
self.assertGreaterEqual(keys_removed, 0)
self.assertEqual(keys_restored, 0)
self.assertGreaterEqual(pages_visited, 0)
def check_crash_restart(self, olddir, newdir):
''' Simulate a crash from olddir and restart in newdir. '''
# with the connection still open, copy files to new directory
copy_wiredtiger_home(olddir, newdir)
# Open the new directory
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
session.verify(self.uri)
conn.close()
def check_crash_restart(self, olddir, newdir):
''' Simulate a crash from olddir and restart in newdir. '''
# with the connection still open, copy files to new directory
copy_wiredtiger_home(olddir, newdir)
# Open the new directory
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
session.verify(self.uri)
conn.close()
def test_backup25(self):
log2 = "WiredTigerLog.0000000002"
# Create a logged table.
self.session.create(self.uri, self.config_log)
# Insert small amounts of data at a time stopping just after we
# cross into log file 2.
while not os.path.exists(log2):
self.add_data(self.uri, 'key', 'value')
self.session.checkpoint()
# Add more data after the checkpoint.
c = self.session.open_cursor(self.uri)
c["newkey"] = "newvalue"
c.close()
# Open the backup cursor and then add new data to the table.
bkup_c = self.session.open_cursor('backup:', None, None)
# Add new data twice and checkpoint to have later checkpoints after the backup
# cursor is open. Add an uncheckpointed but logged modification too.
c = self.session.open_cursor(self.uri)
c["bkupkey1"] = "bkupvalue1"
c.close()
self.session.checkpoint()
c = self.session.open_cursor(self.uri)
c["bkupkey2"] = "bkupvalue2"
c.close()
self.session.checkpoint()
c = self.session.open_cursor(self.uri)
c["bkupkey3"] = "bkupvalue3"
c.close()
# Make sure any data log records are on disk.
self.session.log_flush('sync=on')
# Make a copy of the database to another directory to restart after the "crash"
# with the backup cursor open.
os.mkdir(self.dir)
copy_wiredtiger_home(self, '.', self.dir)
bkup_c.close()
# Open the new directory and verify we can see the data after the backup cursor was opened.
with self.expectedStdoutPattern(
'Both WiredTiger.turtle and WiredTiger.backup exist.*'):
new_conn = self.wiredtiger_open(self.dir)
new_sess = new_conn.open_session()
c = new_sess.open_cursor(self.uri)
self.assertEqual(c["bkupkey1"], "bkupvalue1")
self.assertEqual(c["bkupkey2"], "bkupvalue2")
self.assertEqual(c["bkupkey3"], "bkupvalue3")
c.close()
new_conn.close()
def backup_check(self, check_value, expected_ts_log, expected_ts_nolog,
expected_nots_log, expected_nots_nolog):
newdir = "BACKUP"
copy_wiredtiger_home(self, '.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
cur_ts_log = session.open_cursor(self.uri + self.table_ts_log, None)
cur_ts_nolog = session.open_cursor(self.uri + self.table_ts_nolog,
None)
cur_nots_log = session.open_cursor(self.uri + self.table_nots_log,
None)
cur_nots_nolog = session.open_cursor(self.uri + self.table_nots_nolog,
None)
# In FLCS the values are bytes, which are numbers, but the tests below are via
# string inclusion rather than just equality of values. Not sure why that is, but
# I'm going to assume there's a reason for it and not change things. Compensate.
if self.value_format == '8t':
check_value = str(check_value)
# Count how many times the check_value is present in the
# logged timestamp table.
actual_ts_log = 0
for k, v in cur_ts_log:
if check_value in str(v):
actual_ts_log += 1
cur_ts_log.close()
# Count how many times the check_value is present in the
# not logged timestamp table
actual_ts_nolog = 0
for k, v in cur_ts_nolog:
if check_value in str(v):
actual_ts_nolog += 1
cur_ts_nolog.close()
# Count how many times the check_value is present in the
# logged non-timestamp table.
actual_nots_log = 0
for k, v in cur_nots_log:
if check_value in str(v):
actual_nots_log += 1
cur_nots_log.close()
# Count how many times the check_value is present in the
# not logged non-timestamp table.
actual_nots_nolog = 0
for k, v in cur_nots_nolog:
if check_value in str(v):
actual_nots_nolog += 1
cur_nots_nolog.close()
conn.close()
self.assertEqual(actual_ts_log, expected_ts_log)
self.assertEqual(actual_ts_nolog, expected_ts_nolog)
self.assertEqual(actual_nots_log, expected_nots_log)
self.assertEqual(actual_nots_nolog, expected_nots_nolog)
def durable_check(self, check_value, uri, ds):
# Simulating recovery.
newdir = "BACKUP"
copy_wiredtiger_home(self, '.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
cursor = session.open_cursor(uri, None)
cursor.next()
self.assertTrue(
check_value == cursor.get_value(),
"for key " + str(1) + ", expected " + str(check_value) + ", got " +
str(cursor.get_value()))
cursor.close()
session.close()
conn.close()
def backup_check(self, check_value, valcnt, valcnt2, valcnt3):
newdir = "BACKUP"
copy_wiredtiger_home(self, '.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
c = session.open_cursor(self.uri + self.tablename, None)
c2 = session.open_cursor(self.uri + self.tablename2, None)
c3 = session.open_cursor(self.uri + self.tablename3, None)
# In FLCS the values are bytes, which are numbers, but the tests below are via
# string inclusion rather than just equality of values. Not sure why that is, but
# I'm going to assume there's a reason for it and not change things. Compensate.
if self.value_format == '8t':
check_value = str(check_value)
# Count how many times the second value is present
count = 0
for k, v in c:
if check_value in str(v):
# print "check_value found in key " + str(k)
count += 1
c.close()
# Count how many times the second value is present in the
# non-timestamp table.
count2 = 0
for k, v in c2:
if check_value in str(v):
# print "check_value found in key " + str(k)
count2 += 1
c2.close()
# Count how many times the second value is present in the
# logged timestamp table.
count3 = 0
for k, v in c3:
if check_value in str(v):
count3 += 1
c3.close()
conn.close()
# print "CHECK BACKUP: Count " + str(count) + " Count2 " + str(count2) + " Count3 " + str(count3)
# print "CHECK BACKUP: Expect value2 count " + str(valcnt)
# print "CHECK BACKUP: 2nd table Expect value2 count " + str(valcnt2)
# print "CHECK BACKUP: 3rd table Expect value2 count " + str(valcnt3)
self.assertEqual(count, valcnt)
self.assertEqual(count2, valcnt2)
self.assertEqual(count3, valcnt3)
def test_bug018(self):
'''Test closing multiple tables'''
basename = 'bug018.'
baseuri = 'file:' + basename
c1 = self.create_table(baseuri + '01.wt')
c2 = self.create_table(baseuri + '02.wt')
self.session.begin_transaction()
c1['key'] = 'value'
c2['key'] = 'value'
self.session.commit_transaction()
# Simulate a write failure by closing the file descriptor for the second
# table out from underneath WiredTiger. We do this right before
# closing the connection so that the write error happens during close
# when writing out the final data. Allow table 1 to succeed and force
# an erorr writing out table 2.
#
# This is Linux-specific code to figure out the file descriptor.
for f in os.listdir('/proc/self/fd'):
try:
if os.readlink('/proc/self/fd/' + f).endswith(basename +
'02.wt'):
os.close(int(f))
except OSError:
pass
# Expect an error and messages, so turn off stderr checking.
with self.expectedStderrPattern(''):
try:
self.close_conn()
except wiredtiger.WiredTigerError:
self.conn = None
# Make a backup for forensics in case something goes wrong.
backup_dir = 'BACKUP'
copy_wiredtiger_home('.', backup_dir, True)
# After reopening and running recovery both tables should be in
# sync even though table 1 was successfully written and table 2
# had an error on close.
self.open_conn()
c1 = self.session.open_cursor(baseuri + '01.wt')
c2 = self.session.open_cursor(baseuri + '02.wt')
self.assertEqual(list(c1), list(c2))
def copy_and_restore(self, backup_cursor, last_doc_in_backup,
last_doc_in_data):
log_files_to_copy = 0
os.mkdir(self.backup_dir)
if self.all_log_files:
helper.copy_wiredtiger_home(self, '.', self.backup_dir)
log_files_copied = [
x for x in os.listdir(self.backup_dir)
if x.startswith('WiredTigerLog.')
]
self.assertEqual(len(log_files_copied), 2)
else:
while True:
ret = backup_cursor.next()
if ret != 0:
break
shutil.copy(backup_cursor.get_key(), self.backup_dir)
if backup_cursor.get_key().startswith('WiredTigerLog.'):
log_files_to_copy += 1
self.assertEqual(ret, wiredtiger.WT_NOTFOUND)
self.assertEqual(log_files_to_copy, 1)
backup_conn = self.wiredtiger_open(self.backup_dir, self.conn_config)
if self.all_log_files:
self.captureout.checkAdditionalPattern(
self, 'Both WiredTiger.turtle and WiredTiger.backup exist.*')
session = backup_conn.open_session()
cursor = session.open_cursor(self.uri)
if self.all_log_files:
doc_cnt = 0
for key, val in cursor:
doc_cnt += 1
self.assertLessEqual(key, last_doc_in_data)
self.assertEqual(doc_cnt, last_doc_in_data)
else:
doc_cnt = 0
for key, val in cursor:
doc_cnt += 1
self.assertLessEqual(key, last_doc_in_backup)
self.assertEqual(doc_cnt, last_doc_in_backup)
def durable_check(self, check_value, uri, ds, nrows):
# Checkpoint and backup so as to simulate recovery
self.session.checkpoint()
newdir = "BACKUP"
copy_wiredtiger_home('.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
cursor = session.open_cursor(uri, None)
# Skip the initial rows, which were not updated
for i in range(0, nrows+1):
self.assertEquals(cursor.next(), 0)
#print "Check value : " + str(check_value)
#print "value : " + str(cursor.get_value())
self.assertTrue(check_value == cursor.get_value())
cursor.close()
session.close()
conn.close()
def check_manual_backup(self, i, olddir, newdir):
''' Simulate a manual backup from olddir and restart in newdir. '''
self.session.checkpoint()
cbkup = self.session.open_cursor('backup:', None, None)
# With the connection still open, copy files to new directory.
# Half the time use an unaligned copy.
aligned = (i % (self.freq * 2) != 0) or os.name == "nt"
copy_wiredtiger_home(olddir, newdir, aligned)
# Now simulate fsyncUnlock by closing the backup cursor.
cbkup.close()
# Open the new directory and verify
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
session.verify(self.uri)
conn.close()
def durable_check(self, check_value, uri, ds, nrows):
# Checkpoint and backup so as to simulate recovery
self.session.checkpoint()
newdir = "BACKUP"
copy_wiredtiger_home('.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
cursor = session.open_cursor(uri, None)
# Skip the initial rows, which were not updated
for i in range(0, nrows + 1):
self.assertEquals(cursor.next(), 0)
#print "Check value : " + str(check_value)
#print "value : " + str(cursor.get_value())
self.assertTrue(check_value == cursor.get_value())
cursor.close()
session.close()
conn.close()
def check_manual_backup(self, i, olddir, newdir):
''' Simulate a manual backup from olddir and restart in newdir. '''
self.session.checkpoint()
cbkup = self.session.open_cursor('backup:', None, None)
# With the connection still open, copy files to new directory.
# Half the time use an unaligned copy.
aligned = (i % (self.freq * 2) != 0) or os.name == "nt"
copy_wiredtiger_home(olddir, newdir, aligned)
# Now simulate fsyncUnlock by closing the backup cursor.
cbkup.close()
# Open the new directory and verify
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
session.verify(self.uri)
conn.close()
def test_bug014(self):
# Populate a table with 1000 keys on small pages.
uri = 'table:test_bug014'
ds = SimpleDataSet(self,
uri,
1000,
key_format=self.key_format,
value_format=self.value_format,
config='allocation_size=512,leaf_page_max=512')
ds.populate()
# Reopen it so we can fast-delete pages.
self.reopen_conn()
# Truncate a chunk of the key/value pairs inside a transaction.
self.session.begin_transaction(None)
start = self.session.open_cursor(uri, None)
start.set_key(ds.key(250))
end = self.session.open_cursor(uri, None)
end.set_key(ds.key(500))
self.session.truncate(None, start, end, None)
start.close()
end.close()
# With the truncation uncommitted, checkpoint the database.
ckpt_session = self.conn.open_session()
ckpt_session.checkpoint(None)
ckpt_session.close()
# Simulate a crash by copying to a new directory.
copy_wiredtiger_home(self, ".", "RESTART")
# Open the new directory.
conn = self.setUpConnectionOpen("RESTART")
session = self.setUpSessionOpen(conn)
cursor = session.open_cursor(uri)
# Confirm all of the records are there.
for i in range(1, 1001):
cursor.set_key(ds.key(i))
self.assertEqual(cursor.search(), 0)
conn.close()
def test_bug018(self):
'''Test closing multiple tables'''
basename = 'bug018.'
baseuri = 'file:' + basename
c1 = self.create_table(baseuri + '01.wt')
c2 = self.create_table(baseuri + '02.wt')
self.session.begin_transaction()
c1['key'] = 'value'
c2['key'] = 'value'
self.session.commit_transaction()
# Simulate a write failure by closing the file descriptor for the second
# table out from underneath WiredTiger. We do this right before
# closing the connection so that the write error happens during close
# when writing out the final data. Allow table 1 to succeed and force
# an error writing out table 2.
#
# This is Linux-specific code to figure out the file descriptor.
for f in os.listdir('/proc/self/fd'):
try:
if os.readlink('/proc/self/fd/' + f).endswith(basename + '02.wt'):
os.close(int(f))
except OSError:
pass
# Expect an error and messages, so turn off stderr checking.
with self.expectedStderrPattern(''):
try:
self.close_conn()
except wiredtiger.WiredTigerError:
self.conn = None
# Make a backup for forensics in case something goes wrong.
backup_dir = 'BACKUP'
copy_wiredtiger_home('.', backup_dir, True)
# After reopening and running recovery both tables should be in
# sync even though table 1 was successfully written and table 2
# had an error on close.
self.open_conn()
c1 = self.session.open_cursor(baseuri + '01.wt')
c2 = self.session.open_cursor(baseuri + '02.wt')
self.assertEqual(list(c1), list(c2))
def test_export(self):
uri_a = self.type + "exporta"
uri_b = self.type + "exportb"
uri_c = self.type + "exportc"
# Create a few tables.
self.session.create(uri_a)
self.session.create(uri_b)
self.session.create(uri_c)
# Insert some records.
c1 = self.session.open_cursor(uri_a)
c1["k1"] = "v1"
c1.close()
c2 = self.session.open_cursor(uri_b)
c2["k2"] = "v2"
c2.close()
c3 = self.session.open_cursor(uri_c)
c3["k3"] = "v3"
c3.close()
self.session.checkpoint()
if self.is_tiered_scenario():
self.session.flush_tier(None)
# Open a special backup cursor for export operation.
export_cursor = self.session.open_cursor('backup:export', None, None)
os.mkdir(self.dir)
copy_wiredtiger_home(self, '.', self.dir)
self.assertTrue(os.path.isfile("WiredTiger.export"))
export_cursor.close()
# The export file should be removed from the home directory.
self.assertFalse(os.path.isfile("WiredTiger.export"))
# The export file should exist in the backup directory.
self.assertTrue(
os.path.isfile(os.path.join(self.dir, "WiredTiger.export")))
def check_manual_backup(self, i, olddir, newdir):
''' Simulate a manual backup from olddir and restart in newdir. '''
self.session.checkpoint()
cbkup = self.session.open_cursor('backup:', None, None)
# With the connection still open, copy files to new directory.
# Half the time use an unaligned copy.
even = i % (self.freq * 2) == 0
aligned = even or os.name == "nt"
copy_wiredtiger_home(olddir, newdir, aligned)
# Half the time try to rename a table and the other half try
# to remove a table. They should fail.
if not even:
self.assertRaises(wiredtiger.WiredTigerError,
lambda: self.session.rename(
self.emptyuri, self.newuri, None))
else:
self.assertRaises(wiredtiger.WiredTigerError,
lambda: self.session.drop(self.emptyuri, None))
# Now simulate fsyncUnlock by closing the backup cursor.
cbkup.close()
# Once the backup cursor is closed we should be able to perform
# schema operations. Test that and then reset the files to their
# expected initial names.
if not even:
self.session.rename(self.emptyuri, self.newuri, None)
self.session.drop(self.newuri, None)
self.session.create(self.emptyuri, self.create_params)
else:
self.session.drop(self.emptyuri, None)
self.session.create(self.emptyuri, self.create_params)
# Open the new directory and verify
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
session.verify(self.uri)
conn.close()
def backup_check(self, check_value, valcnt, valcnt2, valcnt3):
newdir = "BACKUP"
copy_wiredtiger_home('.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
c = session.open_cursor(self.uri + self.tablename, None)
c2 = session.open_cursor(self.uri + self.tablename2, None)
c3 = session.open_cursor(self.uri + self.tablename3, None)
# Count how many times the second value is present
count = 0
for k, v in c:
if check_value in str(v):
# print "check_value found in key " + str(k)
count += 1
c.close()
# Count how many times the second value is present in the
# non-timestamp table.
count2 = 0
for k, v in c2:
if check_value in str(v):
# print "check_value found in key " + str(k)
count2 += 1
c2.close()
# Count how many times the second value is present in the
# logged timestamp table.
count3 = 0
for k, v in c3:
if check_value in str(v):
count3 += 1
c3.close()
conn.close()
# print "CHECK BACKUP: Count " + str(count) + " Count2 " + str(count2) + " Count3 " + str(count3)
# print "CHECK BACKUP: Expect value2 count " + str(valcnt)
# print "CHECK BACKUP: 2nd table Expect value2 count " + str(valcnt2)
# print "CHECK BACKUP: 3rd table Expect value2 count " + str(valcnt3)
# print "CHECK BACKUP: config " + str(self.ckptcfg)
self.assertEqual(count, valcnt)
self.assertEqual(count2, valcnt2)
self.assertEqual(count3, valcnt3)
def copy_and_restore(self, backup_cursor, last_doc_in_backup, last_doc_in_data):
log_files_to_copy = 0
os.mkdir(self.backup_dir)
if self.all_log_files:
helper.copy_wiredtiger_home('.', self.backup_dir)
log_files_copied = filter(lambda x: x.startswith('WiredTigerLog.'), os.listdir(self.backup_dir))
self.assertEqual(len(log_files_copied), 2)
else:
while True:
ret = backup_cursor.next()
if ret != 0:
break
shutil.copy(backup_cursor.get_key(), self.backup_dir)
if backup_cursor.get_key().startswith('WiredTigerLog.'):
log_files_to_copy += 1
self.assertEqual(ret, wiredtiger.WT_NOTFOUND)
self.assertEqual(log_files_to_copy, 1)
backup_conn = self.wiredtiger_open(self.backup_dir, self.conn_config)
if self.all_log_files:
self.captureout.checkAdditionalPattern(self, 'Both WiredTiger.turtle and WiredTiger.backup exist.*')
session = backup_conn.open_session()
cursor = session.open_cursor(self.uri)
if self.all_log_files:
doc_cnt = 0
for key, val in cursor:
doc_cnt += 1
self.assertLessEqual(key, last_doc_in_data)
self.assertEqual(doc_cnt, last_doc_in_data)
else:
doc_cnt = 0
for key, val in cursor:
doc_cnt += 1
self.assertLessEqual(key, last_doc_in_backup)
self.assertEqual(doc_cnt, last_doc_in_backup)
def check_manual_backup(self, i, olddir, newdir):
''' Simulate a manual backup from olddir and restart in newdir. '''
self.session.checkpoint()
cbkup = self.session.open_cursor('backup:', None, None)
# With the connection still open, copy files to new directory.
# Half the time use an unaligned copy.
even = i % (self.freq * 2) == 0
aligned = even or os.name == "nt"
copy_wiredtiger_home(self, olddir, newdir, aligned)
# Half the time try to rename a table and the other half try
# to remove a table. They should fail.
if not even:
self.assertRaises(
wiredtiger.WiredTigerError,
lambda: self.session.rename(self.emptyuri, self.newuri, None))
else:
self.assertRaises(wiredtiger.WiredTigerError,
lambda: self.session.drop(self.emptyuri, None))
# Now simulate fsyncUnlock by closing the backup cursor.
cbkup.close()
# Once the backup cursor is closed we should be able to perform
# schema operations. Test that and then reset the files to their
# expected initial names.
if not even:
self.session.rename(self.emptyuri, self.newuri, None)
self.session.drop(self.newuri, None)
self.session.create(self.emptyuri, self.create_params)
else:
self.session.drop(self.emptyuri, None)
self.session.create(self.emptyuri, self.create_params)
# Open the new directory and verify
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
session.verify(self.uri)
conn.close()
def backup_check(self, check_value, valcnt, valcnt2, valcnt3):
newdir = "BACKUP"
copy_wiredtiger_home('.', newdir, True)
conn = self.setUpConnectionOpen(newdir)
session = self.setUpSessionOpen(conn)
c = session.open_cursor(self.uri + self.tablename, None)
c2 = session.open_cursor(self.uri + self.tablename2, None)
c3 = session.open_cursor(self.uri + self.tablename3, None)
# Count how many times the second value is present
count = 0
for k, v in c:
if check_value in str(v):
# print "check_value found in key " + str(k)
count += 1
c.close()
# Count how many times the second value is present in the
# non-timestamp table.
count2 = 0
for k, v in c2:
if check_value in str(v):
# print "check_value found in key " + str(k)
count2 += 1
c2.close()
# Count how many times the second value is present in the
# logged timestamp table.
count3 = 0
for k, v in c3:
if check_value in str(v):
count3 += 1
c3.close()
conn.close()
# print "CHECK BACKUP: Count " + str(count) + " Count2 " + str(count2) + " Count3 " + str(count3)
# print "CHECK BACKUP: Expect value2 count " + str(valcnt)
# print "CHECK BACKUP: 2nd table Expect value2 count " + str(valcnt2)
# print "CHECK BACKUP: 3rd table Expect value2 count " + str(valcnt3)
# print "CHECK BACKUP: config " + str(self.ckptcfg)
self.assertEqual(count, valcnt)
self.assertEqual(count2, valcnt2)
self.assertEqual(count3, valcnt3)
def test_bug014(self):
# Populate a table with 1000 keys on small pages.
uri = 'table:test_bug014'
ds = SimpleDataSet(self, uri, 1000,
config='allocation_size=512,leaf_page_max=512')
ds.populate()
# Reopen it so we can fast-delete pages.
self.reopen_conn()
# Truncate a chunk of the key/value pairs inside a transaction.
self.session.begin_transaction(None)
start = self.session.open_cursor(uri, None)
start.set_key(ds.key(250))
end = self.session.open_cursor(uri, None)
end.set_key(ds.key(500))
self.session.truncate(None, start, end, None)
start.close()
end.close()
# With the truncation uncommitted, checkpoint the database.
ckpt_session = self.conn.open_session()
ckpt_session.checkpoint(None)
ckpt_session.close()
# Simulate a crash by copying to a new directory.
copy_wiredtiger_home(".", "RESTART")
# Open the new directory.
conn = self.setUpConnectionOpen("RESTART")
session = self.setUpSessionOpen(conn)
cursor = session.open_cursor(uri)
# Confirm all of the records are there.
for i in range(1, 1001):
cursor.set_key(ds.key(i))
self.assertEqual(cursor.search(), 0)
conn.close()
def test_export_restart(self):
uri_a = self.type + "exporta"
uri_b = self.type + "exportb"
uri_c = self.type + "exportc"
# Create two tables.
self.session.create(uri_a)
self.session.create(uri_b)
# Insert some records.
c4 = self.session.open_cursor(uri_a)
c4["k4"] = "v4"
c4.close()
c5 = self.session.open_cursor(uri_b)
c5["k5"] = "v5"
c5.close()
self.session.checkpoint()
if self.is_tiered_scenario():
self.session.flush_tier(None)
# Open a special backup cursor for export operation.
main_cursor = self.session.open_cursor('backup:export', None, None)
# Copy the file so that we have more information if WT-9203 ever happens again.
shutil.copyfile('WiredTiger.export', 'WiredTiger.export.original')
# Copy the main database to another directory, including the WiredTiger.export file.
os.mkdir(self.dir)
copy_wiredtiger_home(self, '.', self.dir)
main_cursor.close()
self.close_conn()
# Open a connection and session on the directory copy.
self.conn = self.setUpConnectionOpen(self.dir)
self.session = self.setUpSessionOpen(self.conn)
# Create a third table and drop the second table.
self.session.create(uri_c)
c6 = self.session.open_cursor(uri_c)
c6["k6"] = "k6"
c6.close()
self.session.checkpoint()
if self.is_tiered_scenario():
self.session.flush_tier(None)
self.session.drop(uri_b)
# Open an export cursor on the database copy.
wt_export_path = os.path.join(self.dir, "WiredTiger.export")
export_cursor = self.session.open_cursor('backup:export', None, None)
# Copy the file so that we have more information if WT-9203 ever happens again.
shutil.copyfile(wt_export_path,
os.path.join(self.dir, "WiredTiger.export.backup"))
self.assertTrue(os.path.isfile(wt_export_path))
# The information for the third table should exist in the WiredTiger.export file
# but the information for the second table should not exist in the file.
with open(wt_export_path, "r") as export_file:
export_file_string = export_file.read()
self.assertFalse("exportb" in export_file_string)
self.assertTrue("exportc" in export_file_string)
export_cursor.close()
def test_rollback_to_stable(self):
nrows = 10
# Create two tables.
uri_1 = "table:rollback_to_stable35_1"
ds_1 = SimpleDataSet(self,
uri_1,
0,
key_format=self.key_format,
value_format=self.value_format)
ds_1.populate()
uri_2 = "table:rollback_to_stable35_2"
ds_2 = SimpleDataSet(self,
uri_2,
0,
key_format=self.key_format,
value_format=self.value_format)
ds_2.populate()
if self.value_format == '8t':
valuea = 97
valueb = 98
valuec = 99
else:
valuea = "aaaaa" * 100
valueb = "bbbbb" * 100
valuec = "ccccc" * 100
self.large_updates(uri_1, uri_2, valuea, ds_1, ds_2, nrows)
self.check(valuea, uri_1, uri_2, nrows)
# Start a long running transaction and keep it open.
session_2 = self.conn.open_session()
session_2.begin_transaction()
self.large_updates(uri_1, uri_2, valueb, ds_1, ds_2, nrows)
self.check(valueb, uri_1, uri_2, nrows)
# Create a checkpoint thread
done = threading.Event()
ckpt = checkpoint_thread(self.conn, done)
try:
ckpt.start()
# Wait for checkpoint to start before committing.
ckpt_started = 0
while not ckpt_started:
stat_cursor = self.session.open_cursor('statistics:', None,
None)
ckpt_started = stat_cursor[stat.conn.txn_checkpoint_running][2]
stat_cursor.close()
time.sleep(1)
self.large_updates(uri_1, uri_2, valuec, ds_1, ds_2, nrows)
self.check(valuec, uri_1, uri_2, nrows)
# Evict the data.
self.evict_cursor(uri_1, nrows, valuec)
# Wait for checkpoint stop timing stress to copy the database.
ckpt_stop_timing_stress = 0
while not ckpt_stop_timing_stress:
time.sleep(1)
stat_cursor = self.session.open_cursor('statistics:', None,
None)
ckpt_stop_timing_stress = stat_cursor[
stat.conn.txn_checkpoint_stop_stress_active][2]
stat_cursor.close()
copy_wiredtiger_home(self, '.', "RESTART")
finally:
done.set()
ckpt.join()
self.session.checkpoint()
# Clear all running transactions before rollback to stable.
session_2.commit_transaction()
session_2.close()
# Open the new directory
self.close_conn()
self.conn_config = 'cache_size=50MB,statistics=(all),log=(enabled)'
conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(conn)
self.check(valuec, uri_1, uri_2, nrows)
stat_cursor = self.session.open_cursor('statistics:', None, None)
calls = stat_cursor[stat.conn.txn_rts][2]
upd_aborted = stat_cursor[stat.conn.txn_rts_upd_aborted][2]
hs_removed = stat_cursor[stat.conn.txn_rts_hs_removed][2]
keys_removed = stat_cursor[stat.conn.txn_rts_keys_removed][2]
keys_restored = stat_cursor[stat.conn.txn_rts_keys_restored][2]
pages_visited = stat_cursor[stat.conn.txn_rts_pages_visited][2]
stat_cursor.close()
self.assertEqual(calls, 0)
self.assertEqual(keys_removed, 0)
self.assertEqual(keys_restored, 0)
self.assertEqual(pages_visited, 0)
self.assertEqual(upd_aborted, 0)
self.assertGreaterEqual(hs_removed, 0)
def test_rollback_to_stable_prepare(self):
nrows = 1000
# Create a table without logging.
self.pr("create/populate tables")
uri_1 = "table:rollback_to_stable10_1"
ds_1 = SimpleDataSet(
self, uri_1, 0, key_format="i", value_format="S", config='log=(enabled=false)')
ds_1.populate()
# Create another table without logging.
uri_2 = "table:rollback_to_stable10_2"
ds_2 = SimpleDataSet(
self, uri_2, 0, key_format="i", value_format="S", config='log=(enabled=false)')
ds_2.populate()
# Pin oldest and stable to timestamp 10.
self.conn.set_timestamp('oldest_timestamp=' + self.timestamp_str(10) +
',stable_timestamp=' + self.timestamp_str(10))
value_a = "aaaaa" * 100
value_b = "bbbbb" * 100
value_c = "ccccc" * 100
value_d = "ddddd" * 100
value_e = "eeeee" * 100
# Perform several updates.
self.pr("large updates")
self.large_updates(uri_1, value_d, ds_1, nrows, self.prepare, 20)
self.large_updates(uri_1, value_c, ds_1, nrows, self.prepare, 30)
self.large_updates(uri_1, value_b, ds_1, nrows, self.prepare, 40)
self.large_updates(uri_1, value_a, ds_1, nrows, self.prepare, 50)
self.large_updates(uri_2, value_d, ds_2, nrows, self.prepare, 20)
self.large_updates(uri_2, value_c, ds_2, nrows, self.prepare, 30)
self.large_updates(uri_2, value_b, ds_2, nrows, self.prepare, 40)
self.large_updates(uri_2, value_a, ds_2, nrows, self.prepare, 50)
# Verify data is visible and correct.
self.check(value_d, uri_1, nrows, 20)
self.check(value_c, uri_1, nrows, 30)
self.check(value_b, uri_1, nrows, 40)
self.check(value_a, uri_1, nrows, 50)
self.check(value_d, uri_2, nrows, 20)
self.check(value_c, uri_2, nrows, 30)
self.check(value_b, uri_2, nrows, 40)
self.check(value_a, uri_2, nrows, 50)
# Pin stable to timestamp 60 if prepare otherwise 50.
if self.prepare:
self.conn.set_timestamp('stable_timestamp=' + self.timestamp_str(60))
else:
self.conn.set_timestamp('stable_timestamp=' + self.timestamp_str(50))
# Here's the update operations we'll perform, encapsulated so we can easily retry
# it if we get a rollback. Rollbacks may occur when checkpoint is running.
def prepare_range_updates(session, cursor, ds, value, nrows, prepare_config):
self.pr("updates")
for i in range(1, nrows):
key = ds.key(i)
cursor.set_key(key)
cursor.set_value(value)
self.assertEquals(cursor.update(), 0)
self.pr("prepare")
session.prepare_transaction(prepare_config)
# Create a checkpoint thread
done = threading.Event()
ckpt = checkpoint_thread(self.conn, done)
try:
self.pr("start checkpoint")
ckpt.start()
# Sleep for sometime so that checkpoint starts.
time.sleep(2)
# Perform several updates in parallel with checkpoint.
session_p1 = self.conn.open_session()
cursor_p1 = session_p1.open_cursor(uri_1)
session_p1.begin_transaction('isolation=snapshot')
self.retry_rollback('update ds1', session_p1,
lambda: prepare_range_updates(
session_p1, cursor_p1, ds_1, value_e, nrows,
'prepare_timestamp=' + self.timestamp_str(69)))
self.evict_cursor(uri_1, nrows, value_a)
# Perform several updates in parallel with checkpoint.
session_p2 = self.conn.open_session()
cursor_p2 = session_p2.open_cursor(uri_2)
session_p2.begin_transaction('isolation=snapshot')
self.retry_rollback('update ds2', session_p2,
lambda: prepare_range_updates(
session_p2, cursor_p2, ds_2, value_e, nrows,
'prepare_timestamp=' + self.timestamp_str(69)))
self.evict_cursor(uri_2, nrows, value_a)
finally:
done.set()
ckpt.join()
# Check that the history store file has been used and has non-zero size before the simulated
# crash.
stat_cursor = self.session.open_cursor('statistics:', None, None)
cache_hs_ondisk = stat_cursor[stat.conn.cache_hs_ondisk][2]
stat_cursor.close()
self.assertGreater(cache_hs_ondisk, 0)
# Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(self, ".", "RESTART")
# Commit the prepared transaction.
session_p1.commit_transaction('commit_timestamp=' + self.timestamp_str(70) + ',durable_timestamp=' + self.timestamp_str(71))
session_p2.commit_transaction('commit_timestamp=' + self.timestamp_str(70) + ',durable_timestamp=' + self.timestamp_str(71))
session_p1.close()
session_p2.close()
# Open the new directory.
self.pr("restart")
self.conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(self.conn)
self.pr("restart complete")
# The history store file size should be greater than zero after the restart.
stat_cursor = self.session.open_cursor('statistics:', None, None)
cache_hs_ondisk = stat_cursor[stat.conn.cache_hs_ondisk][2]
stat_cursor.close()
self.assertGreater(cache_hs_ondisk, 0)
# Check that the correct data is seen at and after the stable timestamp.
self.check(value_a, uri_1, nrows, 50)
self.check(value_a, uri_1, nrows, 80)
self.check(value_b, uri_1, nrows, 40)
self.check(value_c, uri_1, nrows, 30)
self.check(value_d, uri_1, nrows, 20)
# Check that the correct data is seen at and after the stable timestamp.
self.check(value_a, uri_2, nrows, 50)
self.check(value_a, uri_2, nrows, 80)
self.check(value_b, uri_2, nrows, 40)
self.check(value_c, uri_2, nrows, 30)
self.check(value_d, uri_2, nrows, 20)
stat_cursor = self.session.open_cursor('statistics:', None, None)
calls = stat_cursor[stat.conn.txn_rts][2]
hs_removed = stat_cursor[stat.conn.txn_rts_hs_removed][2]
hs_sweep = stat_cursor[stat.conn.txn_rts_sweep_hs_keys][2]
keys_removed = stat_cursor[stat.conn.txn_rts_keys_removed][2]
keys_restored = stat_cursor[stat.conn.txn_rts_keys_restored][2]
pages_visited = stat_cursor[stat.conn.txn_rts_pages_visited][2]
upd_aborted = stat_cursor[stat.conn.txn_rts_upd_aborted][2]
stat_cursor.close()
self.assertEqual(calls, 0)
self.assertEqual(keys_removed, 0)
self.assertEqual(keys_restored, 0)
self.assertGreaterEqual(upd_aborted, 0)
self.assertGreater(pages_visited, 0)
self.assertGreaterEqual(hs_removed, 0)
self.assertGreater(hs_sweep, 0)
# The test may output the following message in eviction under cache pressure. Ignore that.
self.ignoreStdoutPatternIfExists("oldest pinned transaction ID rolled back for eviction")
def prepare_updates(self, ds, nrows, nsessions, nkeys):
# Insert some records with commit timestamp, corrupt file and call salvage, verify before checkpoint.
# Commit some updates to get eviction and history store fired up
commit_value = b"bbbbb" * 100
cursor = self.session.open_cursor(self.uri)
for i in range(1, nsessions * nkeys):
self.session.begin_transaction('isolation=snapshot')
cursor.set_key(ds.key(nrows + i))
cursor.set_value(commit_value)
self.assertEquals(cursor.insert(), 0)
self.session.commit_transaction('commit_timestamp=' + timestamp_str(1))
cursor.close()
# Corrupt the table, Call salvage to recover data from the corrupted table and call verify
self.corrupt_salvage_verify()
# Call checkpoint
self.session.checkpoint()
hs_writes_start = self.get_stat(stat.conn.cache_write_hs)
# Have prepared updates in multiple sessions. This should ensure writing
# prepared updates to the history store
sessions = [0] * nsessions
cursors = [0] * nsessions
prepare_value = b"ccccc" * 100
for j in range (0, nsessions):
sessions[j] = self.conn.open_session()
sessions[j].begin_transaction('isolation=snapshot')
cursors[j] = sessions[j].open_cursor(self.uri)
# Each session will update many consecutive keys.
start = (j * nkeys)
end = start + nkeys
for i in range(start, end):
cursors[j].set_key(ds.key(nrows + i))
cursors[j].set_value(prepare_value)
self.assertEquals(cursors[j].insert(), 0)
sessions[j].prepare_transaction('prepare_timestamp=' + timestamp_str(4))
hs_writes = self.get_stat(stat.conn.cache_write_hs) - hs_writes_start
# Assert if not writing anything to the history store.
self.assertGreaterEqual(hs_writes, 0)
# Test if we can read prepared updates from the history store.
cursor = self.session.open_cursor(self.uri)
self.session.begin_transaction('read_timestamp=' + timestamp_str(3))
for i in range(1, nsessions * nkeys):
cursor.set_key(ds.key(nrows + i))
# The search should pass.
self.assertEqual(cursor.search(), 0)
# Correctness Test - commit_value should be visible
self.assertEquals(cursor.get_value(), commit_value)
# Correctness Test - prepare_value should NOT be visible
self.assertNotEquals(cursor.get_value(), prepare_value)
cursor.close()
# Close all cursors and sessions, this will cause prepared updates to be
# rollback-ed
for j in range (0, nsessions):
cursors[j].close()
sessions[j].close()
# Corrupt the table, Call salvage to recover data from the corrupted table and call verify
self.corrupt_salvage_verify()
self.session.commit_transaction()
self.session.checkpoint()
# Corrupt the table, Call salvage to recover data from the corrupted table and call verify
self.corrupt_salvage_verify()
# Finally, search for the keys inserted with commit timestamp
cursor = self.session.open_cursor(self.uri)
self.pr('Read Keys')
self.session.begin_transaction('read_timestamp=' + timestamp_str(4))
for i in range(1, nkeys):
cursor.set_key(ds.key(nrows + i))
# The search should pass
self.assertEqual(cursor.search(), 0)
# Correctness Test - commit_value should be visible
self.assertEquals(cursor.get_value(), commit_value)
cursor.close()
self.session.commit_transaction()
self.session.checkpoint()
# Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(".", "RESTART")
# Open the new directory.
self.conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(self.conn)
cursor = self.session.open_cursor(self.uri)
# Search the keys inserted with commit timestamp after crash
self.session.begin_transaction('read_timestamp=' + timestamp_str(4))
for i in range(1, nkeys):
cursor.set_key(ds.key(nrows + i))
# The search should pass
self.assertEqual(cursor.search(), 0)
# Correctness Test - commit_value should be visible
self.assertEquals(cursor.get_value(), commit_value)
# Correctness Test - prepare_value should NOT be visible
self.assertNotEquals(cursor.get_value(), prepare_value)
cursor.close()
self.session.commit_transaction()
# After simulating a crash, corrupt the table, call salvage to recover data from the corrupted table
# and call verify
self.corrupt_salvage_verify()
def test_truncate09(self):
# Create a large table with lots of pages.
uri = "table:test_truncate09"
format = 'key_format={},value_format=S'.format(self.key_format)
self.session.create(uri,
'allocation_size=512,leaf_page_max=512,' + format)
cursor = self.session.open_cursor(uri)
for i in range(1, 80000):
cursor[simple_key(cursor, i)] = simple_value(cursor, i)
cursor.close()
# Force to disk.
self.reopen_conn()
# Set the oldest timestamp and the stable timestamp.
self.conn.set_timestamp('oldest_timestamp=' + self.timestamp_str(100))
self.conn.set_timestamp('stable_timestamp=' + self.timestamp_str(100))
# Start a transaction.
self.session.begin_transaction()
# Truncate a chunk.
c1 = self.session.open_cursor(uri, None)
c1.set_key(simple_key(c1, 20000))
c2 = self.session.open_cursor(uri, None)
c2.set_key(simple_key(c1, 40000))
self.session.truncate(None, c1, c2, None)
# Commit the transaction.
self.session.timestamp_transaction('commit_timestamp=' +
self.timestamp_str(150))
self.session.commit_transaction()
# Move the stable timestamp to make the previous truncate operation permanent.
self.conn.set_timestamp('stable_timestamp=' + self.timestamp_str(200))
# Checkpoint
self.session.checkpoint()
# Start a transaction.
self.session.begin_transaction()
# Truncate a chunk.
c1.set_key(simple_key(c1, 50000))
c2.set_key(simple_key(c1, 70000))
self.session.truncate(None, c1, c2, None)
# Remove a single row.
c1.set_key(simple_key(c1, 75000))
c1.remove()
# Commit the transaction.
self.session.timestamp_transaction('commit_timestamp=' +
self.timestamp_str(250))
self.session.commit_transaction()
# Checkpoint
self.session.checkpoint()
# Restart, testing RTS on the copy.
copy_wiredtiger_home(self, ".", "RESTART")
simulate_crash_restart(self, ".", "RESTART")
# Search for a key in the truncated range which is stabilised, hence should not find it.
cursor = self.session.open_cursor(uri)
cursor.set_key(simple_key(cursor, 30000))
self.assertNotEqual(cursor.search(), 0)
# Search for a key in the truncated range which is not stabilised, hence should find it.
cursor.set_key(simple_key(cursor, 60000))
self.assertEqual(cursor.search(), 0)
# Search for a removed key which is not stabilised, hence should find it.
cursor.set_key(simple_key(cursor, 75000))
self.assertEqual(cursor.search(), 0)
def prepare_updates(self, ds):
# Set oldest and stable timestamp for the database.
self.conn.set_timestamp('oldest_timestamp=' + timestamp_str(1))
self.conn.set_timestamp('stable_timestamp=' + timestamp_str(1))
# Commit some updates to get eviction and history store fired up.
# Insert a key at timestamp 1.
commit_key = "C"
commit_value = b"bbbbb" * 100
cursor = self.session.open_cursor(self.uri)
for i in range(1, self.nsessions * self.nkeys):
self.session.begin_transaction('isolation=snapshot')
key = commit_key + ds.key(self.nrows + i)
cursor.set_key(key)
cursor.set_value(commit_value)
self.assertEquals(cursor.insert(), 0)
self.session.commit_transaction('commit_timestamp=' +
timestamp_str(1))
cursor.close()
# Call checkpoint.
self.session.checkpoint()
cursor = self.session.open_cursor(self.uri)
for i in range(1, self.nsessions * self.nkeys):
self.session.begin_transaction('isolation=snapshot')
key = commit_key + ds.key(self.nrows + i)
cursor.set_key(key)
self.assertEquals(cursor.remove(), 0)
self.session.commit_transaction('commit_timestamp=' +
timestamp_str(10))
cursor.close()
# Move the stable timestamp to match the timestamp for the last update.
self.conn.set_timestamp('stable_timestamp=' + timestamp_str(10))
hs_writes_start = self.get_stat(stat.conn.cache_write_hs)
# Have prepared updates in multiple sessions. This should ensure writing prepared updates to
# the data store. Insert the same key at timestamp 20, but with prepare updates.
sessions = [0] * self.nsessions
cursors = [0] * self.nsessions
prepare_value = b"ccccc" * 100
for j in range(0, self.nsessions):
sessions[j] = self.conn.open_session()
sessions[j].begin_transaction('isolation=snapshot')
cursors[j] = sessions[j].open_cursor(self.uri)
# Each session will update many consecutive keys.
start = (j * self.nkeys)
end = start + self.nkeys
for i in range(start, end):
cursors[j].set_key(commit_key + ds.key(self.nrows + i))
cursors[j].set_value(prepare_value)
self.assertEquals(cursors[j].insert(), 0)
sessions[j].prepare_transaction('prepare_timestamp=' +
timestamp_str(20))
hs_writes = self.get_stat(stat.conn.cache_write_hs) - hs_writes_start
# Assert if not writing anything to the history store.
self.assertGreaterEqual(hs_writes, 0)
txn_config = 'read_timestamp=' + timestamp_str(
5) + ',ignore_prepare=true'
# Search keys with timestamp 5, ignore_prepare=true and expect the cursor search to return 0 (key found)
self.search_keys_timestamp_and_ignore(ds, txn_config, commit_value)
txn_config = 'read_timestamp=' + timestamp_str(
20) + ',ignore_prepare=true'
# Search keys with timestamp 20, ignore_prepare=true, expect the cursor to return wiredtiger.WT_NOTFOUND
self.search_keys_timestamp_and_ignore(ds, txn_config, None)
prepare_conflict_msg = '/conflict with a prepared update/'
txn_config = 'read_timestamp=' + timestamp_str(
20) + ',ignore_prepare=false'
# Search keys with timestamp 20, ignore_prepare=false and expect the cursor the cursor search to return prepare conflict message
self.search_keys_timestamp_and_ignore(ds, txn_config,
prepare_conflict_msg, True)
# If commit is True then commit the transactions and simulate a crash which would
# eventualy rollback transactions.
if self.commit == True:
# Commit the prepared_transactions with timestamp 30.
for j in range(0, self.nsessions):
sessions[j].commit_transaction('commit_timestamp=' +
timestamp_str(30) +
',durable_timestamp=' +
timestamp_str(30))
# Move the stable timestamp to match the durable timestamp for prepared updates.
self.conn.set_timestamp('stable_timestamp=' + timestamp_str(30))
self.session.checkpoint()
# Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(self, ".", "RESTART")
# Open the new directory.
self.conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(self.conn)
# After simulating a crash, search for the keys inserted. Rollback-to-stable as part of recovery
# will try restoring values according to the last known stable timestamp.
# Search keys with timestamp 5, ignore_prepare=true and expect the cursor search to return
# value committed before prepared update.
txn_config = 'read_timestamp=' + timestamp_str(
5) + ',ignore_prepare=false'
self.search_keys_timestamp_and_ignore(ds, txn_config, commit_value)
# Search keys with timestamp 20, ignore_prepare=true and expect the cursor search to return
# WT_NOTFOUND.
txn_config = 'read_timestamp=' + timestamp_str(
20) + ',ignore_prepare=true'
self.search_keys_timestamp_and_ignore(ds, txn_config, None)
# Search keys with timestamp 20, ignore_prepare=false and expect the cursor search to return WT_NOTFOUND.
txn_config = 'read_timestamp=' + timestamp_str(
20) + ',ignore_prepare=false'
self.search_keys_timestamp_and_ignore(ds, txn_config, None)
# If commit is true then the commit_tramsactions was called and we will expect prepare_value.
if self.commit == True:
txn_config = 'read_timestamp=' + timestamp_str(
30) + ',ignore_prepare=true'
# Search keys with timestamp 30, ignore_prepare=true and expect the cursor value to be prepare_value.
self.search_keys_timestamp_and_ignore(ds, txn_config,
prepare_value)
else:
# Commit is false and we simulated a crash/restart which would have rolled-back the transactions, therefore we expect the
# cursor search to return WT_NOTFOUND.
txn_config = 'read_timestamp=' + timestamp_str(
30) + ',ignore_prepare=true'
# Search keys with timestamp 30, ignore_prepare=true and expect the cursor value to return WT_NOTFOUND.
self.search_keys_timestamp_and_ignore(ds, txn_config, None)
if self.commit == True:
txn_config = 'read_timestamp=' + timestamp_str(
30) + ',ignore_prepare=false'
# Search keys with timestamp 30, ignore_prepare=false and expect the cursor value to be prepare_value.
self.search_keys_timestamp_and_ignore(ds, txn_config,
prepare_value)
else:
# Commit is false and we simulated a crash/restart which would have rolled-back the transactions, therefore we expect the
# cursor search to return WT_NOTFOUND.
txn_config = 'read_timestamp=' + timestamp_str(
30) + ',ignore_prepare=false'
# Search keys with timestamp 30, ignore_prepare=false and expect the cursor value to return WT_NOTFOUND.
self.search_keys_timestamp_and_ignore(ds, txn_config, None)
def test_rollback_to_stable_prepare(self):
nrows = 1000
# Create a table without logging.
self.pr("create/populate tables")
uri_1 = "table:rollback_to_stable10_1"
ds_1 = SimpleDataSet(self,
uri_1,
0,
key_format="i",
value_format="S",
config='log=(enabled=false)')
ds_1.populate()
# Create another table without logging.
uri_2 = "table:rollback_to_stable10_2"
ds_2 = SimpleDataSet(self,
uri_2,
0,
key_format="i",
value_format="S",
config='log=(enabled=false)')
ds_2.populate()
# Pin oldest and stable to timestamp 10.
self.conn.set_timestamp('oldest_timestamp=' + timestamp_str(10) +
',stable_timestamp=' + timestamp_str(10))
value_a = "aaaaa" * 100
value_b = "bbbbb" * 100
value_c = "ccccc" * 100
value_d = "ddddd" * 100
value_e = "eeeee" * 100
value_f = "fffff" * 100
# Perform several updates.
self.pr("large updates")
self.large_updates(uri_1, value_d, ds_1, nrows, 20)
self.large_updates(uri_1, value_c, ds_1, nrows, 30)
self.large_updates(uri_1, value_b, ds_1, nrows, 40)
self.large_updates(uri_1, value_a, ds_1, nrows, 50)
self.large_updates(uri_2, value_d, ds_2, nrows, 20)
self.large_updates(uri_2, value_c, ds_2, nrows, 30)
self.large_updates(uri_2, value_b, ds_2, nrows, 40)
self.large_updates(uri_2, value_a, ds_2, nrows, 50)
# Verify data is visible and correct.
self.check(value_d, uri_1, nrows, 20)
self.check(value_c, uri_1, nrows, 30)
self.session.breakpoint()
self.check(value_b, uri_1, nrows, 40)
self.check(value_a, uri_1, nrows, 50)
self.check(value_d, uri_2, nrows, 20)
self.check(value_c, uri_2, nrows, 30)
self.session.breakpoint()
self.check(value_b, uri_2, nrows, 40)
self.check(value_a, uri_2, nrows, 50)
# Pin stable to timestamp 60 if prepare otherwise 50.
if self.prepare:
self.conn.set_timestamp('stable_timestamp=' + timestamp_str(60))
else:
self.conn.set_timestamp('stable_timestamp=' + timestamp_str(50))
# Here's the update operation we'll perform, encapsulated so we can easily retry
# it if we get a rollback. Rollbacks may occur when checkpoint is running.
def simple_update(cursor, key, value):
cursor.set_key(key)
cursor.set_value(value)
self.assertEquals(cursor.update(), 0)
# Create a checkpoint thread
done = threading.Event()
ckpt = checkpoint_thread(self.conn, done)
try:
self.pr("start checkpoint")
ckpt.start()
# Perform several updates in parallel with checkpoint.
session_p1 = self.conn.open_session()
cursor_p1 = session_p1.open_cursor(uri_1)
session_p1.begin_transaction('isolation=snapshot')
self.pr("updates 1")
for i in range(1, nrows):
retry_rollback(
self, 'update ds1',
lambda: simple_update(cursor_p1, ds_1.key(i), value_e))
self.pr("prepare 1")
session_p1.prepare_transaction('prepare_timestamp=' +
timestamp_str(69))
# Perform several updates in parallel with checkpoint.
session_p2 = self.conn.open_session()
cursor_p2 = session_p2.open_cursor(uri_2)
session_p2.begin_transaction('isolation=snapshot')
self.pr("updates 2")
for i in range(1, nrows):
retry_rollback(
self, 'update ds2',
lambda: simple_update(cursor_p2, ds_2.key(i), value_e))
self.pr("prepare 2")
session_p2.prepare_transaction('prepare_timestamp=' +
timestamp_str(69))
finally:
done.set()
ckpt.join()
# Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(".", "RESTART")
# Commit the prepared transaction.
session_p1.commit_transaction('commit_timestamp=' + timestamp_str(70) +
',durable_timestamp=' +
timestamp_str(71))
session_p2.commit_transaction('commit_timestamp=' + timestamp_str(70) +
',durable_timestamp=' +
timestamp_str(71))
session_p1.close()
session_p2.close()
# Open the new directory.
self.pr("restart")
self.conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(self.conn)
self.pr("restart complete")
# Check that the correct data is seen at and after the stable timestamp.
self.check(value_a, uri_1, nrows, 50)
self.check(value_a, uri_1, nrows, 80)
self.check(value_b, uri_1, nrows, 40)
self.check(value_c, uri_1, nrows, 30)
self.check(value_d, uri_1, nrows, 20)
# Check that the correct data is seen at and after the stable timestamp.
self.check(value_a, uri_2, nrows, 50)
self.check(value_a, uri_2, nrows, 80)
self.check(value_b, uri_2, nrows, 40)
self.check(value_c, uri_2, nrows, 30)
self.check(value_d, uri_2, nrows, 20)
stat_cursor = self.session.open_cursor('statistics:', None, None)
calls = stat_cursor[stat.conn.txn_rts][2]
hs_removed = stat_cursor[stat.conn.txn_rts_hs_removed][2]
hs_sweep = stat_cursor[stat.conn.txn_rts_sweep_hs_keys][2]
keys_removed = stat_cursor[stat.conn.txn_rts_keys_removed][2]
keys_restored = stat_cursor[stat.conn.txn_rts_keys_restored][2]
pages_visited = stat_cursor[stat.conn.txn_rts_pages_visited][2]
upd_aborted = stat_cursor[stat.conn.txn_rts_upd_aborted][2]
stat_cursor.close()
self.assertEqual(calls, 0)
self.assertEqual(keys_removed, 0)
self.assertEqual(keys_restored, 0)
self.assertGreaterEqual(upd_aborted, 0)
self.assertGreater(pages_visited, 0)
self.assertGreaterEqual(hs_removed, 0)
self.assertGreater(hs_sweep, 0)
def test_rollback_to_stable_prepare(self):
nrows = 1000
# Create a table.
self.pr("create/populate tables")
uri_1 = "table:rollback_to_stable10_1"
ds_1 = SimpleDataSet(
self, uri_1, 0, key_format=self.key_format, value_format=self.value_format,
config=self.prepare_extraconfig)
ds_1.populate()
# Create another table.
uri_2 = "table:rollback_to_stable10_2"
ds_2 = SimpleDataSet(
self, uri_2, 0, key_format=self.key_format, value_format=self.value_format,
config=self.prepare_extraconfig)
ds_2.populate()
if self.value_format == '8t':
nrows *= 2
value_a = 97
value_b = 98
value_c = 99
value_d = 100
value_e = 101
else:
value_a = "aaaaa" * 100
value_b = "bbbbb" * 100
value_c = "ccccc" * 100
value_d = "ddddd" * 100
value_e = "eeeee" * 100
# Pin oldest and stable to timestamp 10.
self.conn.set_timestamp('oldest_timestamp=' + self.timestamp_str(10) +
',stable_timestamp=' + self.timestamp_str(10))
# Perform several updates.
self.pr("large updates")
self.large_updates(uri_1, value_d, ds_1, nrows, self.prepare, 20)
self.large_updates(uri_1, value_c, ds_1, nrows, self.prepare, 30)
self.large_updates(uri_1, value_b, ds_1, nrows, self.prepare, 40)
self.large_updates(uri_1, value_a, ds_1, nrows, self.prepare, 50)
self.large_updates(uri_2, value_d, ds_2, nrows, self.prepare, 20)
self.large_updates(uri_2, value_c, ds_2, nrows, self.prepare, 30)
self.large_updates(uri_2, value_b, ds_2, nrows, self.prepare, 40)
self.large_updates(uri_2, value_a, ds_2, nrows, self.prepare, 50)
# Verify data is visible and correct.
self.check(value_d, uri_1, nrows, None, 20)
self.check(value_c, uri_1, nrows, None, 30)
self.check(value_b, uri_1, nrows, None, 40)
self.check(value_a, uri_1, nrows, None, 50)
self.check(value_d, uri_2, nrows, None, 20)
self.check(value_c, uri_2, nrows, None, 30)
self.check(value_b, uri_2, nrows, None, 40)
self.check(value_a, uri_2, nrows, None, 50)
# Pin stable to timestamp 60 if prepare otherwise 50.
if self.prepare:
self.conn.set_timestamp('stable_timestamp=' + self.timestamp_str(60))
else:
self.conn.set_timestamp('stable_timestamp=' + self.timestamp_str(50))
# Do an explicit checkpoint first, before starting the background checkpointer.
# Otherwise (depending on timing and load) because there's a lot to write for the
# first checkpoint there's a tendency for the background checkpointer to only
# manage to do the one checkpoint; and sometimes (especially on FLCS) it ends up
# not containing any of the concurrent updates, and then the test fails because
# RTS correctly notices it has no work to do and doesn't visit any of the pages
# or update anything in the history store.
self.session.checkpoint()
# Here's the update operations we'll perform, encapsulated so we can easily retry
# it if we get a rollback. Rollbacks may occur when checkpoint is running.
def prepare_range_updates(session, cursor, ds, value, nrows, prepare_config):
self.pr("updates")
for i in range(1, nrows):
key = ds.key(i)
cursor.set_key(key)
cursor.set_value(value)
self.assertEquals(cursor.update(), 0)
self.pr("prepare")
session.prepare_transaction(prepare_config)
# Create a checkpoint thread
done = threading.Event()
ckpt = checkpoint_thread(self.conn, done)
try:
self.pr("start checkpoint")
ckpt.start()
# Sleep for some time so that checkpoint starts.
time.sleep(5)
# Perform several updates in parallel with checkpoint.
session_p1 = self.conn.open_session()
cursor_p1 = session_p1.open_cursor(uri_1)
session_p1.begin_transaction()
self.retry_rollback('update ds1', session_p1,
lambda: prepare_range_updates(
session_p1, cursor_p1, ds_1, value_e, nrows,
'prepare_timestamp=' + self.timestamp_str(69)))
self.evict_cursor(uri_1, nrows, value_a)
# Perform several updates in parallel with checkpoint.
session_p2 = self.conn.open_session()
cursor_p2 = session_p2.open_cursor(uri_2)
session_p2.begin_transaction()
self.retry_rollback('update ds2', session_p2,
lambda: prepare_range_updates(
session_p2, cursor_p2, ds_2, value_e, nrows,
'prepare_timestamp=' + self.timestamp_str(69)))
self.evict_cursor(uri_2, nrows, value_a)
finally:
done.set()
ckpt.join()
# Check that the history store file has been used and has non-zero size before the simulated
# crash.
stat_cursor = self.session.open_cursor('statistics:', None, None)
cache_hs_ondisk = stat_cursor[stat.conn.cache_hs_ondisk][2]
stat_cursor.close()
self.assertGreater(cache_hs_ondisk, 0)
# Simulate a crash by copying to a new directory(RESTART).
copy_wiredtiger_home(self, ".", "RESTART")
# Commit the prepared transaction.
session_p1.commit_transaction('commit_timestamp=' + self.timestamp_str(70) + ',durable_timestamp=' + self.timestamp_str(71))
session_p2.commit_transaction('commit_timestamp=' + self.timestamp_str(70) + ',durable_timestamp=' + self.timestamp_str(71))
session_p1.close()
session_p2.close()
# Open the new directory.
self.pr("restart")
self.conn = self.setUpConnectionOpen("RESTART")
self.session = self.setUpSessionOpen(self.conn)
self.pr("restart complete")
# The history store file size should be greater than zero after the restart.
stat_cursor = self.session.open_cursor('statistics:', None, None)
cache_hs_ondisk = stat_cursor[stat.conn.cache_hs_ondisk][2]
stat_cursor.close()
self.assertGreater(cache_hs_ondisk, 0)
# Check that the correct data is seen at and after the stable timestamp.
self.check(value_a, uri_1, nrows, None, 50)
self.check(value_a, uri_1, nrows, None, 80)
self.check(value_b, uri_1, nrows, None, 40)
self.check(value_c, uri_1, nrows, None, 30)
self.check(value_d, uri_1, nrows, None, 20)
# Check that the correct data is seen at and after the stable timestamp.
self.check(value_a, uri_2, nrows, None, 50)
self.check(value_a, uri_2, nrows, None, 80)
self.check(value_b, uri_2, nrows, None, 40)
self.check(value_c, uri_2, nrows, None, 30)
self.check(value_d, uri_2, nrows, None, 20)
stat_cursor = self.session.open_cursor('statistics:', None, None)
calls = stat_cursor[stat.conn.txn_rts][2]
hs_removed = stat_cursor[stat.conn.txn_rts_hs_removed][2]
hs_sweep = stat_cursor[stat.conn.txn_rts_sweep_hs_keys][2]
keys_removed = stat_cursor[stat.conn.txn_rts_keys_removed][2]
keys_restored = stat_cursor[stat.conn.txn_rts_keys_restored][2]
pages_visited = stat_cursor[stat.conn.txn_rts_pages_visited][2]
upd_aborted = stat_cursor[stat.conn.txn_rts_upd_aborted][2]
stat_cursor.close()
self.assertEqual(calls, 0)
self.assertEqual(keys_removed, 0)
self.assertEqual(keys_restored, 0)
self.assertGreaterEqual(upd_aborted, 0)
self.assertGreater(pages_visited, 0)
# Each row that gets processed by RTS can be counted by either hs_removed or hs_sweep,
# but not both. If the data store page for the row appears in the last checkpoint, it
# gets counted in hs_removed; if not, it gets counted in hs_sweep, unless the history
# store page for the row didn't make it out, in which case nothing gets counted at all.
# We expect at least some history store pages to appear, so assert that some rows get
# processed, but the balance between the two counts depends on test timing and we
# should not depend on it.
self.assertGreater(hs_removed + hs_sweep, 0)
# The test may output the following message in eviction under cache pressure. Ignore that.
self.ignoreStdoutPatternIfExists("oldest pinned transaction ID rolled back for eviction")
