def get_s3_backup(instance, date, backup_type):
""" Find xbstream file for an instance on s3 on a given day
Args:
instance - A hostaddr object for the desired instance
date - Desired date of restore file
backup_type - xbstream or mysqldump
Returns:
A list of s3 keys
"""
backup_keys = list()
prefixes = set()
try:
replica_set = instance.get_zk_replica_set()[0]
except:
log.debug('Instance {} is not in zk'.format(instance))
replica_set = None
if replica_set:
prefixes.add(BACKUP_SEARCH_PREFIX.format(
retention_policy=environment_specific.get_backup_retention_policy(instance),
backup_type=backup_type,
replica_set=replica_set,
hostname=instance.hostname,
port=instance.port,
timestamp=date))
prefixes.add(BACKUP_SEARCH_INITIAL_PREFIX.format(
backup_type=backup_type,
hostname=instance.hostname,
port=instance.port,
timestamp=date))
conn = boto.connect_s3()
for bucket in environment_specific.BACKUP_BUCKET_DOWNLOAD_MAP[host_utils.get_iam_role()]:
bucket_conn = conn.get_bucket(bucket, validate=False)
for prefix in prefixes:
log.info('Looking for backup with prefix '
's3://{bucket}/{prefix}'.format(bucket=bucket,
prefix=prefix))
bucket_items = bucket_conn.list(prefix=prefix)
for key in bucket_items:
if (key.size <= MINIMUM_VALID_BACKUP_SIZE_BYTES):
continue
backup_keys.append(key)
if not backup_keys:
msg = ''.join([NO_BACKUP, instance.__str__()])
raise Exception(msg)
return backup_keys
def logical_backup_instance(instance, timestamp, initial_build):
""" Take a compressed mysqldump backup
Args:
instance - A hostaddr instance
timestamp - A timestamp which will be used to create the backup filename
initial_build - Boolean, if this is being created right after the server
was built
Returns:
A string of the path to the finished backup
"""
backup_file = create_backup_file_name(instance, timestamp,
initial_build,
BACKUP_TYPE_LOGICAL)
(dump_user,
dump_pass) = mysql_lib.get_mysql_user_for_role(USER_ROLE_MYSQLDUMP)
dump_cmd = MYSQLDUMP_CMD.format(dump_user=dump_user,
dump_pass=dump_pass,
host=instance.hostname,
port=instance.port).split()
procs = dict()
try:
log.info(' '.join(dump_cmd + ['|']))
procs['mysqldump'] = subprocess.Popen(dump_cmd,
stdout=subprocess.PIPE)
procs['pv'] = create_pv_proc(procs['mysqldump'].stdout)
log.info(' '.join(PIGZ + ['|']))
procs['pigz'] = subprocess.Popen(PIGZ,
stdin=procs['pv'].stdout,
stdout=subprocess.PIPE)
log.info('Uploading backup to s3://{buk}/{key}'
''.format(buk=environment_specific.BACKUP_BUCKET_UPLOAD_MAP[host_utils.get_iam_role()],
key=backup_file))
safe_uploader.safe_upload(precursor_procs=procs,
stdin=procs['pigz'].stdout,
bucket=environment_specific.BACKUP_BUCKET_UPLOAD_MAP[host_utils.get_iam_role()],
key=backup_file)
log.info('mysqldump was successful')
return backup_file
except:
safe_uploader.kill_precursor_procs(procs)
raise
def xtrabackup_instance(instance, timestamp, initial_build):
""" Take a compressed mysql backup
Args:
instance - A hostaddr instance
timestamp - A timestamp which will be used to create the backup filename
initial_build - Boolean, if this is being created right after the server
was built
Returns:
A string of the path to the finished backup
"""
# Prevent issues with too many open files
resource.setrlimit(resource.RLIMIT_NOFILE, (131072, 131072))
backup_file = create_backup_file_name(instance, timestamp,
initial_build,
BACKUP_TYPE_XBSTREAM)
tmp_log = os.path.join(environment_specific.RAID_MOUNT,
'log', 'xtrabackup_{ts}.log'.format(
ts=time.strftime('%Y-%m-%d-%H:%M:%S', timestamp)))
tmp_log_handle = open(tmp_log, "w")
procs = dict()
try:
cmd = create_xtrabackup_command(instance, timestamp, tmp_log)
log.info(' '.join(cmd + [' 2> ', tmp_log, ' | ']))
procs['xtrabackup'] = subprocess.Popen(cmd,
stdout=subprocess.PIPE,
stderr=tmp_log_handle)
procs['pv'] = create_pv_proc(procs['xtrabackup'].stdout)
log.info('Uploading backup to s3://{buk}/{loc}'
''.format(buk=environment_specific.BACKUP_BUCKET_UPLOAD_MAP[host_utils.get_iam_role()],
loc=backup_file))
safe_uploader.safe_upload(precursor_procs=procs,
bucket=environment_specific.BACKUP_BUCKET_UPLOAD_MAP[host_utils.get_iam_role()],
stdin=procs['pv'].stdout,
key=backup_file,
check_func=check_xtrabackup_log,
check_arg=tmp_log)
log.info('Xtrabackup was successful')
return backup_file
except:
safe_uploader.kill_precursor_procs(procs)
raise
def logical_backup_instance(instance, timestamp, blackhole=False,
initial_build=False, databases=None):
""" Take a compressed mysqldump backup
Args:
instance - A hostaddr instance
timestamp - A timestamp which will be used to create the backup filename
blackhole - Boolean, if set will backup DBs as blackhole tables
with no indexes or data
initial_build - Boolean, if this is being created right after the server
was built
databases - List, if set backup only a subset of databases
Returns:
An S3 key of the backup.
"""
zk = host_utils.MysqlZookeeper()
try:
replica_type = zk.get_replica_type_from_instance(instance)
except:
# instance is not in production
replica_type = host_utils.REPLICA_ROLE_MASTER
arg_repl_coordinate = ARG_MASTER_DATA \
if replica_type == host_utils.REPLICA_ROLE_MASTER else ARG_SLAVE_DATA
arg_no_data = ARG_NO_DATA if blackhole else ''
if databases:
backup_type = BACKUP_TYPE_PARTIAL_LOGICAL
db_args = ARG_DATABASES.format(dbs=' '.join(databases))
else:
backup_type = BACKUP_TYPE_LOGICAL
db_args = ARG_ALL_DATABASES
arg_replace = ARG_REPLACE if databases == [mysql_lib.METADATA_DB] else ''
dump_user, dump_pass = mysql_lib.get_mysql_user_for_role(USER_ROLE_MYSQLDUMP)
dump_cmd = MYSQLDUMP_CMD.format(dump_user=dump_user,
dump_pass=dump_pass,
host=instance.hostname,
port=instance.port,
db_args=db_args,
net_buffer_length=MAX_INSERT_LENGTH,
arg_repl_coordinate=arg_repl_coordinate,
arg_replace=arg_replace,
arg_no_data=arg_no_data).split()
backup_file = create_backup_file_name(instance, timestamp,
initial_build,
backup_type)
procs = dict()
log.info(' '.join(dump_cmd + ['|']))
procs['mysqldump'] = subprocess.Popen(dump_cmd,
stdout=subprocess.PIPE)
if blackhole:
procs['innodb_to_blackhole'] = subprocess.Popen(
INNODB_TO_BLACKHOLE,
shell=True,
stdin=procs['mysqldump'].stdout,
stdout=subprocess.PIPE)
log.info(' '.join([INNODB_TO_BLACKHOLE, '|']))
# Blackhole only supports indexes up to 1k long, which is shorter
# than InnoDB. We are therefore removing indexes and
# auto_inc columns.
procs['remove_auto_inc_col_arg'] = subprocess.Popen(
REMOVE_AUTO_INC_COL_ARG,
shell=True,
stdin=procs['innodb_to_blackhole'].stdout,
stdout=subprocess.PIPE)
log.info(' '.join([REMOVE_AUTO_INC_COL_ARG, '|']))
procs['remove_auto_inc_start_value'] = subprocess.Popen(
REMOVE_AUTO_INC_START_VALUE,
shell=True,
stdin=procs['remove_auto_inc_col_arg'].stdout,
stdout=subprocess.PIPE)
log.info(' '.join([REMOVE_AUTO_INC_START_VALUE, '|']))
procs['remove_indexes'] = subprocess.Popen(
REMOVE_INDEXES,
shell=True,
stdin=procs['remove_auto_inc_start_value'].stdout,
stdout=subprocess.PIPE)
log.info(' '.join([REMOVE_INDEXES, '|']))
stdout = procs['remove_indexes'].stdout
elif databases == [mysql_lib.METADATA_DB]:
# If we are backing up the metadata db, we don't want to nuke
# existing data, but need to copy existing data over for rbr
# to work.
procs['create_if_not_exists_sed'] = subprocess.Popen(
CREATE_IF_NOT_EXISTS_SED,
shell=True,
stdin=procs['mysqldump'].stdout,
stdout=subprocess.PIPE)
log.info(' '.join([CREATE_IF_NOT_EXISTS_SED, '|']))
stdout = procs['create_if_not_exists_sed'].stdout
else:
stdout = procs['mysqldump'].stdout
log.info(' '.join(PIGZ + ['|']))
procs['pigz'] = subprocess.Popen(PIGZ,
stdin=stdout,
stdout=subprocess.PIPE)
key = safe_uploader.safe_upload(
precursor_procs=procs,
stdin=procs['pigz'].stdout,
bucket=environment_specific.BACKUP_BUCKET_UPLOAD_MAP[host_utils.get_iam_role()],
key=backup_file,
verbose=True)
log.info('mysqldump was successful')
return key
