def main() -> None:
# Determine the internal IP address of this node.
my_internal_ip = utils.detect_ip()
log.info('My internal IP address is `{}`'.format(my_internal_ip))
master_node_count = get_expected_master_node_count()
log.info('Expected number of DC/OS master nodes: %s', master_node_count)
set_num_replicas(my_internal_ip, master_node_count)
def main() -> None:
# Determine the internal IP address of this node.
my_internal_ip = utils.detect_ip()
log.info('My internal IP address is `{}`'.format(my_internal_ip))
master_node_count = get_expected_master_node_count()
log.info('Expected number of DC/OS master nodes: %s', master_node_count)
set_num_replicas(my_internal_ip, master_node_count)
# We are running CockroachDB v2.0.x so pass '2.0'.
set_cluster_version(my_internal_ip, '2.0')
def main() -> None:
# Determine the internal IP address of this node.
my_internal_ip = utils.detect_ip()
log.info('My internal IP address is `{}`'.format(my_internal_ip))
master_node_count = get_expected_master_node_count()
log.info('Expected number of DC/OS master nodes: %s', master_node_count)
set_num_replicas(my_internal_ip, master_node_count)
# We are running CockroachDB v2.0.x so pass '2.0'.
set_cluster_version(my_internal_ip, '2.0')
def main() -> None:
# Determine the internal IP address of this node.
my_internal_ip = utils.detect_ip()
log.info('My internal IP address is `{}`'.format(my_internal_ip))
args = _parse_args()
if args.backup_file_path:
log.info('Write backup to: {}'.format(args.backup_file_path))
else:
log.info('Write backup to: STDOUT')
if args.backup_file_path:
with open(args.backup_file_path, 'wb') as f:
dump_database(my_internal_ip=my_internal_ip, out=f)
else:
dump_database(my_internal_ip=my_internal_ip, out=sys.stdout)
def main() -> None:
# Determine the internal IP address of this node.
my_internal_ip = utils.detect_ip()
log.info('My internal IP address is `{}`'.format(my_internal_ip))
args = _parse_args()
log.info('Backup filepath: {}'.format(args.backup_file_path))
log.info('Begin IAM database restore procedure.')
# Add sub-second timestamp resolution to the suffix so that this script can
# be run twice within a second.
# See https://jira.mesosphere.com/browse/DCOS-42407
try:
recover_database(
my_internal_ip=my_internal_ip,
backup_file_path=args.backup_file_path,
db_suffix=datetime.now().strftime('%Y%m%d_%H%M%S_%f'),
)
except subprocess.CalledProcessError:
log.error("Failed to restore IAM database.")
sys.exit(1)
log.info('IAM database restored successfully.')
def main() -> None:
logging.basicConfig(format='[%(levelname)s] %(message)s', level='INFO')
# Determine our internal IP.
my_ip = utils.detect_ip()
log.info("My IP is `{}`".format(my_ip))
# Connect to ZooKeeper.
log.info("Connecting to ZooKeeper.")
zk_user = os.environ.get('DATASTORE_ZK_USER')
zk_secret = os.environ.get('DATASTORE_ZK_SECRET')
zk = zk_connect(zk_user=zk_user, zk_secret=zk_secret)
# We are connected to ZooKeeper.
# Ensure that the ZNodes exist.
zk.ensure_path("/cockroach")
zk.ensure_path("/cockroach/nodes")
zk.ensure_path("/cockroach/locking")
# Determine whether the cluster has been bootstrapped already by
# checking whether the `ZK_NODES_PATH` ZNode has children. This is
# best-effort as we aren't holding the lock, but we do call
# `zk.sync()` which is supposed to ensure that we read the latest
# value from ZK.
nodes = _get_registered_nodes(zk=zk, zk_path=ZK_NODES_PATH)
if nodes:
# The cluster has already been initialized. Dump the node IPs to
# `NODES_FILE_PATH` and exit.
log.info("Cluster has members registered already: {}".format(nodes))
if my_ip not in nodes:
log.info(
"IP not found in list of nodes. Registering cluster membership."
)
with _zk_lock(zk=zk,
lock_path=ZK_LOCK_PATH,
contender_id=LOCK_CONTENDER_ID,
timeout=ZK_LOCK_TIMEOUT):
nodes = _register_cluster_membership(zk=zk,
zk_path=ZK_NODES_PATH,
ip=my_ip)
_dump_nodes_to_file(nodes, NODES_FILE_PATH)
log.info("Registration complete. ")
return
# No cockroachdb nodes have been registered with ZK yet. We
# assume that we need to bootstrap the cluster so we take the ZK
# lock and hold it until the cluster is bootstrapped and our IP
# has been successfully registered with ZK.
#
# The lock needs to be held around the entire cockroachdb startup
# procedure as only the first instance should start without the
# --join parameter (and thereby bootstrap the cluster.) This lock
# prevents multiple instances from starting without --join at the
# same time.
#
# If we fail to acquire the lock it means a peer is already
# bootstrapping the cluster. We should crash and when we get
# restarted by systemd, we expect to see that the cluster has been
# bootstrapped and will enter that alternative code path which
# leads to an eventually converged cluster.
with _zk_lock(zk=zk,
lock_path=ZK_LOCK_PATH,
contender_id=LOCK_CONTENDER_ID,
timeout=ZK_LOCK_TIMEOUT):
# We check that the cluster hasn't been bootstrapped since we
# first read the list of nodes from ZK.
log.info("Checking for registered nodes while holding lock.")
nodes = _get_registered_nodes(zk=zk, zk_path=ZK_NODES_PATH)
if nodes:
# The cluster has been bootstrapped since we checked. We join the
# existing cluster and dump the node IPs.
log.info("Cluster has already been initialized: {}".format(nodes))
nodes = _register_cluster_membership(zk=zk,
zk_path=ZK_NODES_PATH,
ip=my_ip)
_dump_nodes_to_file(nodes, NODES_FILE_PATH)
return
else:
log.info("Cluster has not been initialized yet.")
# The cluster still has not been bootstrapped. We start
# cockroachdb without a list of cluster IPs to join,
# which will cause it to bootstrap the cluster.
_init_cockroachdb_cluster(ip=my_ip)
# Only now that the CockroachDB cluster has been initialized, we
# add our IP to the list of nodes that have successfully joined the
# cluster at one stage or another.
#
# If this fails the fact that a cluster was initialized will be
# ignored by subsequent runs as our IP won't be present in ZK.
nodes = _register_cluster_membership(zk=zk,
zk_path=ZK_NODES_PATH,
ip=my_ip)
_dump_nodes_to_file(nodes, NODES_FILE_PATH)
log.info("Successfully initialized cluster.")
return
def main() -> None:
logging.basicConfig(format='[%(levelname)s] %(message)s', level='INFO')
# Determine our internal IP.
my_ip = utils.detect_ip()
log.info("My IP is `{}`".format(my_ip))
# Connect to ZooKeeper.
log.info("Connecting to ZooKeeper.")
zk_user = os.environ.get('DATASTORE_ZK_USER')
zk_secret = os.environ.get('DATASTORE_ZK_SECRET')
zk = zk_connect(zk_user=zk_user, zk_secret=zk_secret)
# We are connected to ZooKeeper.
# Ensure that the ZNodes exist.
zk.ensure_path("/cockroach")
zk.ensure_path("/cockroach/nodes")
zk.ensure_path("/cockroach/locking")
# Determine whether the cluster has been bootstrapped already by
# checking whether the `ZK_NODES_PATH` ZNode has children. This is
# best-effort as we aren't holding the lock, but we do call
# `zk.sync()` which is supposed to ensure that we read the latest
# value from ZK.
nodes = _get_registered_nodes(zk=zk, zk_path=ZK_NODES_PATH)
if nodes:
# The cluster has already been initialized. Dump the node IPs to
# `NODES_FILE_PATH` and exit.
log.info("Cluster has members registered already: {}".format(nodes))
if my_ip not in nodes:
log.info("IP not found in list of nodes. Registering cluster membership.")
with _zk_lock(zk=zk, lock_path=ZK_LOCK_PATH, contender_id=LOCK_CONTENDER_ID, timeout=ZK_LOCK_TIMEOUT):
nodes = _register_cluster_membership(zk=zk, zk_path=ZK_NODES_PATH, ip=my_ip)
_dump_nodes_to_file(nodes, NODES_FILE_PATH)
log.info("Registration complete. ")
return
# No cockroachdb nodes have been registered with ZK yet. We
# assume that we need to bootstrap the cluster so we take the ZK
# lock and hold it until the cluster is bootstrapped and our IP
# has been successfully registered with ZK.
#
# The lock needs to be held around the entire cockroachdb startup
# procedure as only the first instance should start without the
# --join parameter (and thereby bootstrap the cluster.) This lock
# prevents multiple instances from starting without --join at the
# same time.
#
# If we fail to acquire the lock it means a peer is already
# bootstrapping the cluster. We should crash and when we get
# restarted by systemd, we expect to see that the cluster has been
# bootstrapped and will enter that alternative code path which
# leads to an eventually converged cluster.
with _zk_lock(zk=zk, lock_path=ZK_LOCK_PATH, contender_id=LOCK_CONTENDER_ID, timeout=ZK_LOCK_TIMEOUT):
# We check that the cluster hasn't been bootstrapped since we
# first read the list of nodes from ZK.
log.info("Checking for registered nodes while holding lock.")
nodes = _get_registered_nodes(zk=zk, zk_path=ZK_NODES_PATH)
if nodes:
# The cluster has been bootstrapped since we checked. We join the
# existing cluster and dump the node IPs.
log.info("Cluster has already been initialized: {}".format(nodes))
nodes = _register_cluster_membership(zk=zk, zk_path=ZK_NODES_PATH, ip=my_ip)
_dump_nodes_to_file(nodes, NODES_FILE_PATH)
return
else:
log.info("Cluster has not been initialized yet.")
# The cluster still has not been bootstrapped. We start
# cockroachdb without a list of cluster IPs to join,
# which will cause it to bootstrap the cluster.
_init_cockroachdb_cluster(ip=my_ip)
# Only now that the CockroachDB cluster has been initialized, we
# add our IP to the list of nodes that have successfully joined the
# cluster at one stage or another.
#
# If this fails the fact that a cluster was initialized will be
# ignored by subsequent runs as our IP won't be present in ZK.
nodes = _register_cluster_membership(zk=zk, zk_path=ZK_NODES_PATH, ip=my_ip)
_dump_nodes_to_file(nodes, NODES_FILE_PATH)
log.info("Successfully initialized cluster.")
return