def backup_instance(self):
""" Back up a replica instance to s3 in csv """
host_lock_handle = None
try:
log.info('Backup for instance {i} started at {t}'
''.format(t=str(self.timestamp), i=self.instance))
log.info('Checking heartbeat to make sure replicaiton is not too '
'lagged.')
self.check_replication_for_backup()
log.info('Taking host backup lock')
host_lock_handle = host_utils.take_flock_lock(
backup.BACKUP_LOCK_FILE)
log.info('Setting up export directory structure')
self.setup_and_get_tmp_path()
log.info('Will temporarily dump inside of {path}'
''.format(path=self.dump_base_path))
log.info('Releasing any invalid shard backup locks')
self.ensure_backup_locks_sanity()
log.info('Deleting old expired locks')
self.purge_old_expired_locks()
log.info('Stopping replication SQL thread to get a snapshot')
mysql_lib.stop_replication(self.instance,
mysql_lib.REPLICATION_THREAD_SQL)
workers = []
for _ in range(multiprocessing.cpu_count() / 2):
proc = multiprocessing.Process(
target=self.mysql_backup_csv_dbs)
proc.daemon = True
proc.start()
workers.append(proc)
# throw in a sleep to make sure all threads have started dumps
time.sleep(2)
log.info('Restarting replication')
mysql_lib.start_replication(self.instance,
mysql_lib.REPLICATION_THREAD_SQL)
for worker in workers:
worker.join()
if not self.dbs_to_backup.empty():
raise Exception('All worker processes have completed, but '
'work remains in the queue')
log.info('CSV backup is complete, will run a check')
mysql_backup_status.verify_csv_backup(self.instance.replica_type,
self.datestamp,
self.instance)
finally:
if host_lock_handle:
log.info('Releasing general host backup lock')
host_utils.release_flock_lock(host_lock_handle)
def mysql_backup(instance,
backup_type=backup.BACKUP_TYPE_XBSTREAM,
initial_build=False):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
backup_type - backup.BACKUP_TYPE_LOGICAL or backup.BACKUP_TYPE_XBSTREAM
initial_build - Boolean, if this is being created right after the server
was built
"""
log.info('Confirming sanity of replication (if applicable)')
zk = host_utils.MysqlZookeeper()
try:
(_, replica_type) = zk.get_replica_set_from_instance(instance)
except:
# instance is not in production
replica_type = None
if replica_type and replica_type != host_utils.REPLICA_ROLE_MASTER:
mysql_lib.assert_replication_sanity(instance)
log.info('Logging initial status to mysqlops')
start_timestamp = time.localtime()
lock_handle = None
backup_id = mysql_lib.start_backup_log(instance, backup_type,
start_timestamp)
# Take a lock to prevent multiple backups from running concurrently
try:
log.info('Taking backup lock')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
# Actually run the backup
log.info('Running backup')
if backup_type == backup.BACKUP_TYPE_XBSTREAM:
backup_file = backup.xtrabackup_instance(instance, start_timestamp,
initial_build)
elif backup_type == backup.BACKUP_TYPE_LOGICAL:
backup_file = backup.logical_backup_instance(
instance, start_timestamp, initial_build)
else:
raise Exception('Unsupported backup type {backup_type}'
''.format(backup_type=backup_type))
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
# Update database with additional info now that backup is done.
if backup_id:
log.info("Updating database log entry with final backup info")
mysql_lib.finalize_backup_log(backup_id, backup_file)
else:
log.info("The backup is complete, but we were not able to "
"write to the central log DB.")
def backup_instance(self):
""" Back up a replica instance to s3 in csv """
host_lock_handle = None
try:
log.info('Backup for instance {i} started at {t}'
''.format(t=str(self.timestamp),
i=self.instance))
log.info('Checking heartbeat to make sure replicaiton is not too '
'lagged.')
self.check_replication_for_backup()
log.info('Taking host backup lock')
host_lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info('Setting up export directory structure')
self.setup_and_get_tmp_path()
log.info('Will temporarily dump inside of {path}'
''.format(path=self.dump_base_path))
log.info('Releasing any invalid shard backup locks')
self.ensure_backup_locks_sanity()
log.info('Deleting old expired locks')
self.purge_old_expired_locks()
log.info('Stopping replication SQL thread to get a snapshot')
mysql_lib.stop_replication(self.instance, mysql_lib.REPLICATION_THREAD_SQL)
workers = []
for _ in range(multiprocessing.cpu_count() / 2):
proc = multiprocessing.Process(target=self.mysql_backup_csv_dbs)
proc.daemon = True
proc.start()
workers.append(proc)
# throw in a sleep to make sure all threads have started dumps
time.sleep(2)
log.info('Restarting replication')
mysql_lib.start_replication(self.instance, mysql_lib.REPLICATION_THREAD_SQL)
for worker in workers:
worker.join()
if not self.dbs_to_backup.empty():
raise Exception('All worker processes have completed, but '
'work remains in the queue')
log.info('CSV backup is complete, will run a check')
mysql_backup_status.verify_csv_backup(self.instance.replica_type,
self.datestamp,
self.instance)
finally:
if host_lock_handle:
log.info('Releasing general host backup lock')
host_utils.release_flock_lock(host_lock_handle)
def mysql_backup(instance, backup_type=backup.BACKUP_TYPE_XBSTREAM):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
"""
log.info('Logging initial status to mysqlops')
start_timestamp = time.localtime()
lock_handle = None
backup_id = mysql_lib.start_backup_log(instance, backup_type,
start_timestamp)
# Take a lock to prevent multiple backups from running concurrently
try:
log.info('Taking backup lock')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info('Cleaning up old backups')
purge_mysql_backups.purge_mysql_backups(instance, skip_lock=True)
# Actually run the backup
log.info('Running backup')
if backup_type == backup.BACKUP_TYPE_XBSTREAM:
backup_file = backup.xtrabackup_instance(instance, start_timestamp)
elif backup_type == backup.BACKUP_TYPE_LOGICAL:
backup_file = backup.logical_backup_instance(
instance, start_timestamp)
else:
raise Exception('Unsupported backup type {backup_type}'
''.format(backup_type=backup_type))
# Upload file to s3
log.info('Uploading file to s3')
backup.s3_upload(backup_file)
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
# Update database with additional info now that backup is done.
if backup_id:
log.info("Updating database log entry with final backup info")
mysql_lib.finalize_backup_log(backup_id,
backup_file,
size=os.stat(backup_file).st_size)
else:
log.info("The backup is complete, but we were not able to "
"write to the central log DB.")
# Running purge again
log.info('Purging backups again')
purge_mysql_backups.purge_mysql_backups(instance)
def mysql_backup(instance, backup_type=backup.BACKUP_TYPE_XBSTREAM, initial_build=False):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
backup_type - backup.BACKUP_TYPE_LOGICAL or backup.BACKUP_TYPE_XBSTREAM
initial_build - Boolean, if this is being created right after the server
was built
"""
log.info('Confirming sanity of replication (if applicable)')
zk = host_utils.MysqlZookeeper()
try:
(_, replica_type) = zk.get_replica_set_from_instance(instance)
except:
# instance is not in production
replica_type = None
if replica_type and replica_type != host_utils.REPLICA_ROLE_MASTER:
mysql_lib.assert_replication_sanity(instance)
log.info('Logging initial status to mysqlops')
start_timestamp = time.localtime()
lock_handle = None
backup_id = mysql_lib.start_backup_log(instance, backup_type,
start_timestamp)
# Take a lock to prevent multiple backups from running concurrently
try:
log.info('Taking backup lock')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
# Actually run the backup
log.info('Running backup')
if backup_type == backup.BACKUP_TYPE_XBSTREAM:
backup_file = backup.xtrabackup_instance(instance, start_timestamp, initial_build)
elif backup_type == backup.BACKUP_TYPE_LOGICAL:
backup_file = backup.logical_backup_instance(instance, start_timestamp, initial_build)
else:
raise Exception('Unsupported backup type {backup_type}'
''.format(backup_type=backup_type))
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
# Update database with additional info now that backup is done.
if backup_id:
log.info("Updating database log entry with final backup info")
mysql_lib.finalize_backup_log(backup_id, backup_file)
else:
log.info("The backup is complete, but we were not able to "
"write to the central log DB.")
def archive_mysql_binlogs(port, dry_run):
""" Flush logs and upload all binary logs that don't exist to s3
Arguments:
port - Port of the MySQL instance on which to act
dry_run - Display output but do not uplad
"""
binlog_rotator.rotate_binlogs_if_needed(port, dry_run)
zk = host_utils.MysqlZookeeper()
instance = host_utils.HostAddr(':'.join((host_utils.HOSTNAME,
str(port))))
if zk.get_replica_set_from_instance(instance)[0] is None:
log.info('Instance is not in production, exiting')
return
lock_handle = None
ensure_binlog_archiving_table_sanity(instance)
try:
log.info('Taking binlog archiver lock')
lock_handle = host_utils.take_flock_lock(BINLOG_LOCK_FILE)
log_bin_dir = host_utils.get_cnf_setting('log_bin', port)
bin_logs = mysql_lib.get_master_logs(instance)
logged_uploads = get_logged_binlog_uploads(instance)
for binlog in bin_logs[:-1]:
err_count = 0
local_file = os.path.join(os.path.dirname(log_bin_dir),
binlog['Log_name'])
if already_uploaded(instance, local_file, logged_uploads):
continue
success = False
while not success:
try:
upload_binlog(instance, local_file, dry_run)
success = True
except:
if err_count > MAX_ERRORS:
log.error('Error count in thread > MAX_THREAD_ERROR. '
'Aborting :(')
raise
log.error('error: {e}'.format(e=traceback.format_exc()))
err_count = err_count + 1
time.sleep(err_count*2)
log.info('Archiving complete')
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
def mysql_backup(instance, backup_type=backup.BACKUP_TYPE_XBSTREAM):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
"""
log.info("Logging initial status to mysqlops")
start_timestamp = time.localtime()
lock_handle = None
backup_id = mysql_lib.start_backup_log(instance, backup_type, start_timestamp)
# Take a lock to prevent multiple backups from running concurrently
try:
log.info("Taking backup lock")
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info("Cleaning up old backups")
purge_mysql_backups.purge_mysql_backups(instance, skip_lock=True)
# Actually run the backup
log.info("Running backup")
if backup_type == backup.BACKUP_TYPE_XBSTREAM:
backup_file = backup.xtrabackup_instance(instance, start_timestamp)
elif backup_type == backup.BACKUP_TYPE_LOGICAL:
backup_file = backup.logical_backup_instance(instance, start_timestamp)
else:
raise Exception("Unsupported backup type {backup_type}" "".format(backup_type=backup_type))
# Upload file to s3
log.info("Uploading file to s3")
backup.s3_upload(backup_file)
finally:
if lock_handle:
log.info("Releasing lock")
host_utils.release_flock_lock(lock_handle)
# Update database with additional info now that backup is done.
if backup_id:
log.info("Updating database log entry with final backup info")
mysql_lib.finalize_backup_log(backup_id, backup_file, size=os.stat(backup_file).st_size)
else:
log.info("The backup is complete, but we were not able to " "write to the central log DB.")
# Running purge again
log.info("Purging backups again")
purge_mysql_backups.purge_mysql_backups(instance)
def purge_mysql_backups(instance, skip_lock=False):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
skip_lock - Don't take out a lock against other backup related functions
running
"""
lock_handle = None
try:
if not skip_lock:
log.info('Taking backup lock')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
(temp_path, target_path) = backup.get_paths(str(instance.port))
log.info("Cleaning up any partial backups")
backup.remove_backups(temp_path,
keep_newest=0,
extension=('.xbstream', '.sql.gz'))
log.info("Purge xtrabackup backups to the "
"last {cnt}".format(cnt=KEEP_OLD_XTRABACKUP))
backup.remove_backups(target_path,
keep_newest=KEEP_OLD_XTRABACKUP,
extension=('.xbstream'))
log.info("Purge logical backup backups to the "
"last {cnt}".format(cnt=KEEP_OLD_LOGICAL))
backup.remove_backups(target_path,
keep_newest=KEEP_OLD_LOGICAL,
extension=('.sql.gz'))
log.info("Chmod'ing {target}".format(target=backup.TARGET_DIR))
host_utils.change_perms(backup.TARGET_DIR, 777)
log.info("Chmod'ing {temp}".format(temp=backup.TEMP_DIR))
host_utils.change_perms(backup.TEMP_DIR, 777)
finally:
if not skip_lock and lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
def purge_mysql_backups(instance, skip_lock=False):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
skip_lock - Don't take out a lock against other backup related functions
running
"""
lock_handle = None
try:
if not skip_lock:
log.info('Taking backup lock')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
(temp_path, target_path) = backup.get_paths(str(instance.port))
log.info("Cleaning up any partial backups")
backup.remove_backups(temp_path,
keep_newest=0,
extension=('.xbstream', '.sql.gz'))
log.info("Purge xtrabackup backups to the "
"last {cnt}".format(cnt=KEEP_OLD_XTRABACKUP))
backup.remove_backups(target_path,
keep_newest=KEEP_OLD_XTRABACKUP,
extension=('.xbstream'))
log.info("Purge logical backup backups to the "
"last {cnt}".format(cnt=KEEP_OLD_LOGICAL))
backup.remove_backups(target_path,
keep_newest=KEEP_OLD_LOGICAL,
extension=('.sql.gz'))
log.info("Chmod'ing {target}".format(target=backup.TARGET_DIR))
host_utils.change_perms(backup.TARGET_DIR, 777)
log.info("Chmod'ing {temp}".format(temp=backup.TEMP_DIR))
host_utils.change_perms(backup.TEMP_DIR, 777)
finally:
if not skip_lock and lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
def restore_instance(restore_source, destination, restore_type,
restore_file, no_repl, date,
add_to_zk, skip_production_check,
test_restore):
""" Restore a MySQL backup on to localhost
Args:
restore_source - A hostaddr object for where to pull a backup from
destination - A hostaddr object for where to restore the backup
restore_type - How to pull the backup, options are 's3', 'remote_server'
and 'local_file'
no_repl - Should replication be not started. It will always be setup.
date - What date should the backup be from
add_to_zk - Should the instnace be added to zk. If so, the log from the
host being launched will be consulted.
skip_production_check - Do not check if the host is already in zk for
production use.
test_restore - Use less ram and shutdown the instance after going
through the motions of a restore.
"""
(temp_dir, target_dir) = backup.get_paths(str(destination.port))
log.info('Supplied source is {source}'.format(source=restore_source))
log.info('Supplied destination is {dest}'.format(dest=destination))
log.info('Restore type is {rest}'.format(rest=restore_type))
log.info('Local restore file is {file}'.format(file=restore_file))
log.info('Desired date of restore {date}'.format(date=date))
if test_restore == 'test':
log.info('Running restore in test mode')
# Try to prevent unintentional destruction of prod servers
zk = host_utils.MysqlZookeeper()
try:
(_, replica_type) = zk.get_replica_set_from_instance(destination)
except:
# instance is not in production
replica_type = None
if replica_type == host_utils.REPLICA_ROLE_MASTER:
# If the instance, we will refuse to run. No ifs, ands, or buts/
raise Exception('Restore script must never run on a master')
if replica_type:
if skip_production_check:
log.info('Ignoring production check. We hope you know what you '
'are doing and we will try to take a backup in case '
'you are wrong.')
try:
mysql_backup.mysql_backup(destination)
except Exception as e:
log.error(e)
log.warning('Unable to take a backup. We will give you {time} '
'seconds to change your mind and ^c.'
''.format(time=SCARY_TIMEOUT))
time.sleep(SCARY_TIMEOUT)
else:
raise Exception("It appears {instance} is in use. This is"
" very dangerous!".format(instance=destination))
# Take a lock to prevent multiple restores from running concurrently
log.info('Taking a flock to block another restore from starting')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info('Rebuilding cnf files just in case')
mysql_cnf_builder.build_cnf()
mysql_init_server.create_and_chown_dirs(destination.port)
# load some data from the mysql conf file
datadir = host_utils.get_cnf_setting('datadir', destination.port)
# Where will we look for a backup?
if restore_type != 'local_file':
(restore_type, restore_source,
restore_file, restore_size) = find_a_backup_to_restore(restore_type, restore_source,
destination, date)
# Not using an if/else because find_a_backup_to_restore could set to
# local_file if the file has already been downloaded.
if restore_type == 'local_file':
restore_source = backup.get_host_from_backup(restore_file)
# restore_size will be computed in the unpack function
restore_size = None
log.info('Detected the source of backup as {src}'.format(src=restore_source))
if restore_source.get_zk_replica_set():
replica_set = restore_source.get_zk_replica_set()[0]
master = zk.get_mysql_instance_from_replica_set(replica_set, host_utils.REPLICA_ROLE_MASTER)
else:
# ZK has no idea what this replica set is, probably a new replica set.
master = restore_source
# Start logging
row_id = backup.start_restore_log(master, {'restore_type': restore_type,
'test_restore': test_restore,
'restore_source': restore_source,
'restore_port': destination.port,
'restore_file': restore_file,
'source_instance': destination.hostname,
'restore_date': date,
'replication': no_repl,
'zookeeper': add_to_zk})
# Giant try to allow logging if anything goes wrong.
try:
# If we hit an exception, this status will be used. If not, it will
# be overwritten
restore_log_update = {'restore_status': 'BAD'}
log.info('Quick sanity check')
mysql_init_server.basic_host_sanity()
log.info('Shutting down MySQL')
host_utils.stop_mysql(destination.port)
log.info('Removing any existing MySQL data')
mysql_init_server.delete_mysql_data(destination.port)
log.info('Unpacking {rfile} into {ddir}'.format(rfile=restore_file,
ddir=datadir))
backup.xbstream_unpack(restore_file, destination.port,
restore_source, restore_type, restore_size)
log.info('Decompressing files in {path}'.format(path=datadir))
backup.innobackup_decompress(destination.port)
log.info('Applying logs')
if test_restore == 'test':
# We don't really need a lot of memory if we're just
# verifying that it works.
backup.apply_log(destination.port, memory='1G')
else:
backup.apply_log(destination.port, memory='10G')
log.info('Removing old innodb redo logs')
mysql_init_server.delete_innodb_log_files(destination.port)
log.info('Setting permissions for MySQL on {dir}'.format(dir=datadir))
host_utils.change_owner(datadir, 'mysql', 'mysql')
log.info('Starting MySQL')
host_utils.upgrade_auth_tables(destination.port)
restore_log_update = {'restore_status': 'OK'}
log.info('Running MySQL upgrade')
host_utils.start_mysql(destination.port,
options=host_utils.DEFAULTS_FILE_EXTRA_ARG.format(defaults_file=host_utils.MYSQL_NOREPL_CNF_FILE))
if master == backup.get_host_from_backup(restore_file):
log.info('Pulling replication info from restore to backup source')
(binlog_file, binlog_pos) = backup.parse_xtrabackup_binlog_info(datadir)
else:
log.info('Pulling replication info from restore to '
'master of backup source')
(binlog_file, binlog_pos) = backup.parse_xtrabackup_slave_info(datadir)
log.info('Setting up MySQL replication')
restore_log_update['replication'] = 'FAIL'
# Since we haven't started the slave yet, make sure we've got these
# plugins installed, whether we use them or not.
mysql_lib.setup_semisync_plugins(destination)
# Try to configure replication. If this was just a test restore,
# don't wait for it to catch up - don't even start the slave.
if test_restore == 'test':
mysql_lib.change_master(destination,
master,
binlog_file,
binlog_pos,
no_start=True)
backup.quick_test_replication(destination)
else:
mysql_lib.change_master(destination,
master,
binlog_file,
binlog_pos,
no_start=(no_repl == 'SKIP'))
mysql_lib.wait_replication_catch_up(destination)
host_utils.restart_pt_daemons(destination.port)
restore_log_update['replication'] = 'OK'
mysql_lib.setup_response_time_metrics(destination)
except Exception as e:
log.error(e)
if row_id is not None:
restore_log_update['status_message'] = e
restore_log_update['finished_at'] = True
raise
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
backup.update_restore_log(master, row_id, restore_log_update)
# If this was a test restore, we don't need to keep the 3307
# instance running, so let's shut it off.
if test_restore == 'test':
log.info('Shutting down MySQL backup/restore test instance')
host_utils.stop_mysql(destination.port)
backup.update_restore_log(master, row_id, {'finished_at': True})
return
try:
if add_to_zk == 'REQ':
log.info('Adding instance to zk')
modify_mysql_zk.auto_add_instance_to_zk(destination, dry_run=False)
backup.update_restore_log(master, row_id, {'zookeeper': 'OK'})
else:
log.info('add_to_zk is not set, therefore not adding to zk')
except Exception as e:
log.warning("An exception occurred: {e}".format(e=e))
log.warning("If this is a DB issue, that's fine. "
"Otherwise, you should check ZK.")
backup.update_restore_log(master, row_id, {'finished_at': True})
log.info('Starting a new backup')
mysql_backup.mysql_backup(destination)
def restore_instance(backup_type, restore_source, destination,
no_repl, date,
add_to_zk, skip_production_check):
""" Restore a MySQL backup on to localhost
Args:
backup_type - Type of backup to restore
restore_source - A hostaddr object for where to pull a backup from
destination - A hostaddr object for where to restore the backup
no_repl - Should replication be not started. It will always be setup.
date - What date should the backup be from
add_to_zk - Should the instnace be added to zk. If so, the log from the
host being launched will be consulted.
skip_production_check - Do not check if the host is already in zk for
production use.
"""
log.info('Supplied source is {source}'.format(source=restore_source))
log.info('Supplied destination is {dest}'.format(dest=destination))
log.info('Desired date of restore {date}'.format(date=date))
zk = host_utils.MysqlZookeeper()
# Try to prevent unintentional destruction of prod servers
log.info('Confirming no prod instances running on destination')
prod_check(destination, skip_production_check)
# Take a lock to prevent multiple restores from running concurrently
log.info('Taking a flock to block another restore from starting')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info('Looking for a backup to restore')
if restore_source:
possible_sources = [restore_source]
else:
possible_sources = get_possible_sources(destination, backup_type)
backup_key = find_a_backup_to_restore(possible_sources, destination,
backup_type, date)
# Figure out what what we use to as the master when we setup replication
(restore_source, _) = backup.get_metadata_from_backup_file(backup_key.name)
if restore_source.get_zk_replica_set():
replica_set = restore_source.get_zk_replica_set()[0]
master = zk.get_mysql_instance_from_replica_set(replica_set, host_utils.REPLICA_ROLE_MASTER)
else:
# ZK has no idea what this replica set is, probably a new replica set.
master = restore_source
# Start logging
row_id = backup.start_restore_log(master, {'restore_source': restore_source,
'restore_port': destination.port,
'restore_file': backup_key.name,
'source_instance': destination.hostname,
'restore_date': date,
'replication': no_repl,
'zookeeper': add_to_zk})
# Giant try to allow logging if anything goes wrong.
try:
# If we hit an exception, this status will be used. If not, it will
# be overwritten
restore_log_update = {'restore_status': 'BAD'}
# This also ensures that all needed directories exist
log.info('Rebuilding local mysql instance')
mysql_init_server.mysql_init_server(destination, skip_production_check=True,
skip_backup=True, skip_locking=True)
if backup_type == backup.BACKUP_TYPE_XBSTREAM:
xbstream_restore(backup_key, destination.port)
if master == restore_source:
log.info('Pulling replication info from restore to backup source')
(binlog_file, binlog_pos) = backup.parse_xtrabackup_binlog_info(destination.port)
else:
log.info('Pulling replication info from restore to '
'master of backup source')
(binlog_file, binlog_pos) = backup.parse_xtrabackup_slave_info(destination.port)
elif backup_type == backup.BACKUP_TYPE_LOGICAL:
logical_restore(backup_key, destination)
host_utils.stop_mysql(destination.port)
log.info('Running MySQL upgrade')
host_utils.upgrade_auth_tables(destination.port)
log.info('Starting MySQL')
host_utils.start_mysql(destination.port,
options=host_utils.DEFAULTS_FILE_EXTRA_ARG.format(defaults_file=host_utils.MYSQL_NOREPL_CNF_FILE))
# Since we haven't started the slave yet, make sure we've got these
# plugins installed, whether we use them or not.
mysql_lib.setup_semisync_plugins(destination)
restore_log_update = {'restore_status': 'OK'}
# Try to configure replication.
log.info('Setting up MySQL replication')
restore_log_update['replication'] = 'FAIL'
if backup_type == backup.BACKUP_TYPE_XBSTREAM:
mysql_lib.change_master(destination,
master,
binlog_file,
binlog_pos,
no_start=(no_repl == 'SKIP'))
elif backup_type == backup.BACKUP_TYPE_LOGICAL:
if no_repl == 'SKIP':
log.info('As requested, not starting replication.')
else:
mysql_lib.restart_replication(destination)
if no_repl == 'REQ':
mysql_lib.wait_replication_catch_up(destination)
restore_log_update['replication'] = 'OK'
host_utils.restart_pt_daemons(destination.port)
mysql_lib.setup_response_time_metrics(destination)
except Exception as e:
log.error(e)
if row_id is not None:
restore_log_update['status_message'] = e
restore_log_update['finished_at'] = True
raise
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
backup.update_restore_log(master, row_id, restore_log_update)
try:
if add_to_zk == 'REQ':
log.info('Adding instance to zk')
modify_mysql_zk.auto_add_instance_to_zk(destination.port,
dry_run=False)
backup.update_restore_log(master, row_id, {'zookeeper': 'OK'})
else:
log.info('add_to_zk is not set, therefore not adding to zk')
except Exception as e:
log.warning("An exception occurred: {e}".format(e=e))
log.warning("If this is a DB issue, that's fine. "
"Otherwise, you should check ZK.")
backup.update_restore_log(master, row_id, {'finished_at': True})
if no_repl == 'REQ':
log.info('Starting a new backup')
mysql_backup.mysql_backup(destination, initial_build=True)
def mysql_init_server(instance,
skip_production_check=False,
skip_locking=False,
skip_backup=True):
""" Remove any data and initialize a MySQL instance
Args:
instance - A hostaddr object pointing towards localhost to act upon
skip_production_check - Dangerous! will not run safety checks to protect
production data
skip_locking - Do not take a lock on localhost. Useful when the caller has
already has taken the lock (ie mysql_restore_xtrabackup)
skip_backup - Don't run a backup after the instance is setup
"""
lock_handle = None
if not skip_locking:
# Take a lock to prevent multiple restores from running concurrently
log.info('Taking a flock to block race conditions')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
try:
# sanity check
zk = host_utils.MysqlZookeeper()
if (not skip_production_check
and instance in zk.get_all_mysql_instances()):
raise Exception("It appears {instance} is in use. This is"
" very dangerous!".format(instance=instance))
log.info('Checking host for mounts, etc...')
basic_host_sanity()
log.info('(re)Generating MySQL cnf files')
mysql_cnf_builder.build_cnf()
log.info('Creating any missing directories')
create_and_chown_dirs(instance.port)
log.info('Shutting down MySQL (if applicable)')
host_utils.stop_mysql(instance.port)
log.info('Deleting existing MySQL data')
delete_mysql_data(instance.port)
log.info('Creating MySQL privileges tables')
init_privileges_tables(instance.port)
log.info('Clearing innodb log files')
delete_innodb_log_files(instance.port)
log.info('Starting up instance')
host_utils.start_mysql(instance.port)
log.info('Importing MySQL users')
mysql_grants.manage_mysql_grants(instance, 'nuke_then_import')
log.info('Creating test database')
mysql_lib.create_db(instance, 'test')
log.info('Setting up query response time plugins')
mysql_lib.setup_response_time_metrics(instance)
log.info('Setting up semi-sync replication plugins')
mysql_lib.setup_semisync_plugins(instance)
log.info('Restarting pt daemons')
host_utils.restart_pt_daemons(instance.port)
log.info('MySQL initalization complete')
finally:
if not skip_locking and lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
if not skip_backup:
log.info('Taking a backup')
mysql_backup.mysql_backup(instance)
def archive_mysql_binlogs(port, dry_run):
""" Flush logs and upload all binary logs that don't exist to s3
Arguments:
port - Port of the MySQL instance on which to act
dry_run - Display output but do not uplad
"""
lock_handle = None
try:
log.info('Taking binlog archiver lock')
lock_handle = host_utils.take_flock_lock(BINLOG_LOCK_FILE)
log_bin_dir = host_utils.get_cnf_setting('log_bin', port)
instance = host_utils.HostAddr(':'.join((host_utils.HOSTNAME,
str(port))))
s3_conn = boto.connect_s3()
bucket = s3_conn.get_bucket(environment_specific.S3_BUCKET, validate=False)
mysql_conn = mysql_lib.connect_mysql(instance)
bin_logs = mysql_lib.get_master_logs(mysql_conn)
prefix = os.path.join(BINLOG_S3_DIR,
instance.hostname,
str(instance.port))
uploaded_binlogs = bucket.get_all_keys(prefix=prefix)
for binlog in bin_logs[:-1]:
compressed_file = ''.join((binlog['Log_name'], '.gz'))
local_file = os.path.join(os.path.dirname(log_bin_dir),
binlog['Log_name'])
local_file_gz = os.path.join(TMP_DIR, compressed_file)
remote_path = os.path.join(BINLOG_S3_DIR,
instance.hostname,
str(instance.port),
compressed_file)
log.info('Local file {local_file} will compress to {local_file_gz} '
'and upload to {remote_path}'.format(local_file=local_file,
local_file_gz=local_file_gz,
remote_path=remote_path))
new_key = boto.s3.key.Key(bucket)
new_key.key = remote_path
if already_uploaded(remote_path, uploaded_binlogs):
log.info('Binlog has already been uploaded')
continue
if dry_run:
log.info('In dry_run mode, skipping compression and upload')
continue
log.info('Compressing file')
f_in = open(local_file, 'r')
f_out = gzip.open(local_file_gz, 'w', compresslevel=2)
f_out.writelines(f_in)
f_out.close()
f_in.close()
log.info('Uploading file')
new_key.set_contents_from_filename(local_file_gz)
log.info('Deleting local compressed file')
os.remove(local_file_gz)
log.info('Archiving complete')
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
def xtrabackup_backup_instance(instance):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
"""
starttime_sql = time.strftime('%Y-%m-%d %H:%M:%S')
log.info('Logging initial status to mysqlops')
row_id = None
lock_handle = None
try:
reporting_conn = mysql_lib.get_mysqlops_connections()
cursor = reporting_conn.cursor()
sql = ("INSERT INTO mysqlops.mysql_backups "
"SET "
"hostname = %(hostname)s, "
"port = %(port)s, "
"started = %(started)s, "
"backup_type = 'xbstream' ")
metadata = {'hostname': instance.hostname,
'port': instance.port,
'started': starttime_sql}
cursor.execute(sql, metadata)
row_id = cursor.lastrowid
reporting_conn.commit()
except Exception as e:
log.warning("Unable to write log entry to "
"mysqlopsdb001: {e}".format(e=e))
log.warning("However, we will attempt to continue with the backup.")
# Take a lock to prevent multiple backups from running concurrently
try:
log.info('Taking backup lock')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info('Cleaning up old backups')
purge_mysql_backups.purge_mysql_backups(instance, skip_lock=True)
# Actually run the backup
log.info('Running backup')
backup_file = backup.xtrabackup_instance(instance)
finished = time.strftime('%Y-%m-%d %H:%M:%S')
# Upload file to s3
log.info('Uploading file to s3')
backup.s3_upload(backup_file)
# Update database with additional info now that backup is done.
if row_id is None:
log.info("The backup is complete, but we were not able to "
"write to the central log DB.")
else:
log.info("Updating database log entry with final backup info")
try:
sql = ("UPDATE mysqlops.mysql_backups "
"SET "
"filename = %(filename)s, "
"finished = %(finished)s, "
"size = %(size)s "
"WHERE id = %(id)s")
metadata = {'filename': backup_file,
'finished': finished,
'size': os.stat(backup_file).st_size,
'id': row_id}
cursor.execute(sql, metadata)
reporting_conn.commit()
reporting_conn.close()
except Exception as e:
log.warning("Unable to update mysqlopsdb with "
"backup status: {e}".format(e=e))
# Running purge again most for the chmod
purge_mysql_backups.purge_mysql_backups(instance, skip_lock=True)
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
def restore_instance(restore_source, destination, no_repl, date, add_to_zk,
skip_production_check):
""" Restore a MySQL backup on to localhost
Args:
restore_source - A hostaddr object for where to pull a backup from
destination - A hostaddr object for where to restore the backup
no_repl - Should replication be not started. It will always be setup.
date - What date should the backup be from
add_to_zk - Should the instnace be added to zk. If so, the log from the
host being launched will be consulted.
skip_production_check - Do not check if the host is already in zk for
production use.
"""
log.info('Supplied source is {source}'.format(source=restore_source))
log.info('Supplied destination is {dest}'.format(dest=destination))
log.info('Desired date of restore {date}'.format(date=date))
# Try to prevent unintentional destruction of prod servers
zk = host_utils.MysqlZookeeper()
try:
(_, replica_type) = zk.get_replica_set_from_instance(destination)
except:
# instance is not in production
replica_type = None
if replica_type == host_utils.REPLICA_ROLE_MASTER:
# If the instance, we will refuse to run. No ifs, ands, or buts/
raise Exception('Restore script must never run on a master')
if replica_type:
if skip_production_check:
log.info('Ignoring production check. We hope you know what you '
'are doing and we will try to take a backup in case '
'you are wrong.')
try:
mysql_backup.mysql_backup(destination)
except Exception as e:
log.error(e)
log.warning('Unable to take a backup. We will give you {time} '
'seconds to change your mind and ^c.'
''.format(time=SCARY_TIMEOUT))
time.sleep(SCARY_TIMEOUT)
else:
raise Exception("It appears {instance} is in use. This is"
" very dangerous!".format(instance=destination))
# Take a lock to prevent multiple restores from running concurrently
log.info('Taking a flock to block another restore from starting')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info('Rebuilding cnf files just in case')
mysql_cnf_builder.build_cnf()
mysql_init_server.create_and_chown_dirs(destination.port)
# load some data from the mysql conf file
datadir = host_utils.get_cnf_setting('datadir', destination.port)
(restore_source, restore_file,
restore_size) = find_a_backup_to_restore(restore_source, destination,
date)
if restore_source.get_zk_replica_set():
replica_set = restore_source.get_zk_replica_set()[0]
master = zk.get_mysql_instance_from_replica_set(
replica_set, host_utils.REPLICA_ROLE_MASTER)
else:
# ZK has no idea what this replica set is, probably a new replica set.
master = restore_source
# Start logging
row_id = backup.start_restore_log(
master, {
'restore_source': restore_source,
'restore_port': destination.port,
'restore_file': restore_file,
'source_instance': destination.hostname,
'restore_date': date,
'replication': no_repl,
'zookeeper': add_to_zk
})
# Giant try to allow logging if anything goes wrong.
try:
# If we hit an exception, this status will be used. If not, it will
# be overwritten
restore_log_update = {'restore_status': 'BAD'}
log.info('Quick sanity check')
mysql_init_server.basic_host_sanity()
log.info('Shutting down MySQL')
host_utils.stop_mysql(destination.port)
log.info('Removing any existing MySQL data')
mysql_init_server.delete_mysql_data(destination.port)
log.info('Unpacking {rfile} into {ddir}'.format(rfile=restore_file,
ddir=datadir))
backup.xbstream_unpack(restore_file, destination.port, restore_source,
restore_size)
log.info('Decompressing files in {path}'.format(path=datadir))
backup.innobackup_decompress(destination.port)
# Determine how much RAM to use for applying logs based on the
# system's total RAM size; all our boxes have 32G or more, so
# this will always be better than before, but not absurdly high.
log_apply_ram = psutil.phymem_usage()[0] / 1024 / 1024 / 1024 / 3
log.info('Applying logs')
backup.apply_log(destination.port, memory='{}G'.format(log_apply_ram))
log.info('Removing old innodb redo logs')
mysql_init_server.delete_innodb_log_files(destination.port)
log.info('Setting permissions for MySQL on {dir}'.format(dir=datadir))
host_utils.change_owner(datadir, 'mysql', 'mysql')
log.info('Starting MySQL')
host_utils.upgrade_auth_tables(destination.port)
restore_log_update = {'restore_status': 'OK'}
log.info('Running MySQL upgrade')
host_utils.start_mysql(
destination.port,
options=host_utils.DEFAULTS_FILE_EXTRA_ARG.format(
defaults_file=host_utils.MYSQL_NOREPL_CNF_FILE))
if master == backup.get_metadata_from_backup_file(restore_file)[0]:
log.info('Pulling replication info from restore to backup source')
(binlog_file,
binlog_pos) = backup.parse_xtrabackup_binlog_info(datadir)
else:
log.info('Pulling replication info from restore to '
'master of backup source')
(binlog_file,
binlog_pos) = backup.parse_xtrabackup_slave_info(datadir)
log.info('Setting up MySQL replication')
restore_log_update['replication'] = 'FAIL'
# Since we haven't started the slave yet, make sure we've got these
# plugins installed, whether we use them or not.
mysql_lib.setup_semisync_plugins(destination)
# Try to configure replication.
mysql_lib.change_master(destination,
master,
binlog_file,
binlog_pos,
no_start=(no_repl == 'SKIP'))
mysql_lib.wait_replication_catch_up(destination)
host_utils.restart_pt_daemons(destination.port)
restore_log_update['replication'] = 'OK'
mysql_lib.setup_response_time_metrics(destination)
except Exception as e:
log.error(e)
if row_id is not None:
restore_log_update['status_message'] = e
restore_log_update['finished_at'] = True
raise
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
backup.update_restore_log(master, row_id, restore_log_update)
try:
if add_to_zk == 'REQ':
log.info('Adding instance to zk')
modify_mysql_zk.auto_add_instance_to_zk(destination, dry_run=False)
backup.update_restore_log(master, row_id, {'zookeeper': 'OK'})
else:
log.info('add_to_zk is not set, therefore not adding to zk')
except Exception as e:
log.warning("An exception occurred: {e}".format(e=e))
log.warning("If this is a DB issue, that's fine. "
"Otherwise, you should check ZK.")
backup.update_restore_log(master, row_id, {'finished_at': True})
log.info('Starting a new backup')
mysql_backup.mysql_backup(destination)
def archive_mysql_binlogs(port, dry_run):
""" Flush logs and upload all binary logs that don't exist to s3
Arguments:
port - Port of the MySQL instance on which to act
dry_run - Display output but do not uplad
"""
lock_handle = None
try:
log.info('Taking binlog archiver lock')
lock_handle = host_utils.take_flock_lock(BINLOG_LOCK_FILE)
log_bin_dir = host_utils.get_cnf_setting('log_bin', port)
instance = host_utils.HostAddr(':'.join(
(host_utils.HOSTNAME, str(port))))
s3_conn = boto.connect_s3()
bucket = s3_conn.get_bucket(environment_specific.S3_BUCKET,
validate=False)
mysql_conn = mysql_lib.connect_mysql(instance)
bin_logs = mysql_lib.get_master_logs(mysql_conn)
prefix = os.path.join(BINLOG_S3_DIR, instance.hostname,
str(instance.port))
uploaded_binlogs = bucket.get_all_keys(prefix=prefix)
for binlog in bin_logs[:-1]:
compressed_file = ''.join((binlog['Log_name'], '.gz'))
local_file = os.path.join(os.path.dirname(log_bin_dir),
binlog['Log_name'])
local_file_gz = os.path.join(TMP_DIR, compressed_file)
remote_path = os.path.join(BINLOG_S3_DIR, instance.hostname,
str(instance.port), compressed_file)
log.info(
'Local file {local_file} will compress to {local_file_gz} '
'and upload to {remote_path}'.format(
local_file=local_file,
local_file_gz=local_file_gz,
remote_path=remote_path))
new_key = boto.s3.key.Key(bucket)
new_key.key = remote_path
if already_uploaded(remote_path, uploaded_binlogs):
log.info('Binlog has already been uploaded')
continue
if dry_run:
log.info('In dry_run mode, skipping compression and upload')
continue
log.info('Compressing file')
f_in = open(local_file, 'r')
f_out = gzip.open(local_file_gz, 'w', compresslevel=2)
f_out.writelines(f_in)
f_out.close()
f_in.close()
log.info('Uploading file')
new_key.set_contents_from_filename(local_file_gz)
log.info('Deleting local compressed file')
os.remove(local_file_gz)
log.info('Archiving complete')
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
def mysql_init_server(instance,
skip_production_check=False, skip_locking=False,
skip_backup=True):
""" Remove any data and initialize a MySQL instance
Args:
instance - A hostaddr object pointing towards localhost to act upon
skip_production_check - Dangerous! will not run safety checks to protect
production data
skip_locking - Do not take a lock on localhost. Useful when the caller has
already has taken the lock (ie mysql_restore_xtrabackup)
skip_backup - Don't run a backup after the instance is setup
"""
lock_handle = None
if not skip_locking:
# Take a lock to prevent multiple restores from running concurrently
log.info('Taking a flock to block race conditions')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
try:
# sanity check
zk = host_utils.MysqlZookeeper()
if (not skip_production_check and
instance in zk.get_all_mysql_instances()):
raise Exception("It appears {instance} is in use. This is"
" very dangerous!".format(instance=instance))
log.info('Checking host for mounts, etc...')
basic_host_sanity()
log.info('(re)Generating MySQL cnf files')
mysql_cnf_builder.build_cnf()
log.info('Creating any missing directories')
create_and_chown_dirs(instance.port)
log.info('Shutting down MySQL (if applicable)')
host_utils.stop_mysql(instance.port)
log.info('Deleting existing MySQL data')
delete_mysql_data(instance.port)
log.info('Creating MySQL privileges tables')
init_privileges_tables(instance.port)
log.info('Clearing innodb log files')
delete_innodb_log_files(instance.port)
log.info('Starting up instance')
host_utils.start_mysql(instance.port)
log.info('Importing MySQL users')
mysql_grants.manage_mysql_grants(instance, 'nuke_then_import')
log.info('Creating test database')
conn = mysql_lib.connect_mysql(instance)
mysql_lib.create_db(conn, 'test')
log.info('Setting up query response time plugins')
mysql_lib.setup_response_time_metrics(instance)
log.info('Setting up semi-sync replication plugins')
mysql_lib.setup_semisync_plugins(instance)
log.info('Restarting pt daemons')
host_utils.restart_pt_daemons(instance.port)
log.info('MySQL initalization complete')
finally:
if not skip_locking and lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)
if not skip_backup:
log.info('Taking a backup')
mysql_backup.mysql_backup(instance)
def xtrabackup_backup_instance(instance):
""" Run a file based backup on a supplied local instance
Args:
instance - A hostaddr object
"""
starttime_sql = time.strftime('%Y-%m-%d %H:%M:%S')
log.info('Logging initial status to mysqlops')
row_id = None
lock_handle = None
try:
reporting_conn = mysql_lib.get_mysqlops_connections()
cursor = reporting_conn.cursor()
sql = ("INSERT INTO mysqlops.mysql_backups "
"SET "
"hostname = %(hostname)s, "
"port = %(port)s, "
"started = %(started)s, "
"backup_type = 'xbstream' ")
metadata = {
'hostname': instance.hostname,
'port': instance.port,
'started': starttime_sql
}
cursor.execute(sql, metadata)
row_id = cursor.lastrowid
reporting_conn.commit()
except Exception as e:
log.warning("Unable to write log entry to "
"mysqlopsdb001: {e}".format(e=e))
log.warning("However, we will attempt to continue with the backup.")
# Take a lock to prevent multiple backups from running concurrently
try:
log.info('Taking backup lock')
lock_handle = host_utils.take_flock_lock(backup.BACKUP_LOCK_FILE)
log.info('Cleaning up old backups')
purge_mysql_backups.purge_mysql_backups(instance, skip_lock=True)
# Actually run the backup
log.info('Running backup')
backup_file = backup.xtrabackup_instance(instance)
finished = time.strftime('%Y-%m-%d %H:%M:%S')
# Upload file to s3
log.info('Uploading file to s3')
backup.s3_upload(backup_file)
# Update database with additional info now that backup is done.
if row_id is None:
log.info("The backup is complete, but we were not able to "
"write to the central log DB.")
else:
log.info("Updating database log entry with final backup info")
try:
sql = ("UPDATE mysqlops.mysql_backups "
"SET "
"filename = %(filename)s, "
"finished = %(finished)s, "
"size = %(size)s "
"WHERE id = %(id)s")
metadata = {
'filename': backup_file,
'finished': finished,
'size': os.stat(backup_file).st_size,
'id': row_id
}
cursor.execute(sql, metadata)
reporting_conn.commit()
reporting_conn.close()
except Exception as e:
log.warning("Unable to update mysqlopsdb with "
"backup status: {e}".format(e=e))
# Running purge again most for the chmod
purge_mysql_backups.purge_mysql_backups(instance, skip_lock=True)
finally:
if lock_handle:
log.info('Releasing lock')
host_utils.release_flock_lock(lock_handle)