def on_update(event, context):
log.info("Handle resource update event %s" % json.dumps(event, indent=2))
# get parameters passed in
props = event["ResourceProperties"]
nodes_per_zone = int(props["NodesPerZone"])
zone_count = int(props["NumberOfZones"])
parent_stack_name = props["ParentStackName"]
parent_stack_id = props["ParentStackId"]
subnets = props["Subnets"]
security_group_id = props["SecurityGroup"]
# old properties
old_props = event["OldResourceProperties"]
old_nodes_per_zone = int(old_props["NodesPerZone"])
old_zone_count = int(old_props["NumberOfZones"])
# validate diff and handle special case
if old_zone_count != zone_count:
reason = "Updating number of zones is not supported"
log.error(reason)
return cfn_failure_response(event, reason)
if old_nodes_per_zone == nodes_per_zone:
return cfn_success_response(event, reuse_physical_id=True)
if old_nodes_per_zone > nodes_per_zone and nodes_per_zone != 0:
reason = "Scaling down the number of nodes per zone by updating the stack is not recommended. " \
"Please manually remove unused network interface."
log.warning(reason)
return cfn_failure_response(event, reason)
if nodes_per_zone == 0:
log.info("Hibernating the cluster, retain network interfaces")
return cfn_success_response(event, reuse_physical_id=True)
# prepare ENI meta information
id_hash = hashlib.md5(parent_stack_id.encode()).hexdigest()
eni_tag_prefix = parent_stack_name + "-" + id_hash + "_"
eni_idx = old_zone_count * old_nodes_per_zone
for i in range(zone_count):
for j in range(nodes_per_zone):
if j < old_nodes_per_zone:
continue
tag = eni_tag_prefix + str(eni_idx)
eni_idx += 1
eni_id = create_eni(subnets[i], security_group_id, tag)
if not eni_id:
reason = "Failed to create network interface with tag %s" % tag
log.warning(reason)
continue
return cfn_success_response(event, reuse_physical_id=True)
def on_create(event, context):
log.info("Handle resource create event %s" % json.dumps(event, indent=2))
# get parameters passed in
props = event["ResourceProperties"]
vpc = props["Vpc"]
service_name = props["ServiceName"]
security_group = props["SecurityGroup"]
subnets = props["Subnets"]
# create endpoint
try:
response = ec2_client.create_vpc_endpoint(
VpcEndpointType='Interface',
VpcId=vpc,
ServiceName=service_name,
SubnetIds=subnets,
SecurityGroupIds=[security_group]
)
except ClientError as e:
reason="Failed to create vpc endpoint for service %s" % service_name
log.exception(reason)
time.sleep(5) # sleep for 5 seconds to allow exception info being sent to CloudWatch
return cfn_failure_response(event, reason)
endpoint_id = response["VpcEndpoint"]["VpcEndpointId"]
log.info(
"Creating vpc endpoint %s for service %s" % (
endpoint_id, service_name
)
)
log.info("Waiting for vpc endpoint %s creation finish" % endpoint_id)
endpoint_wait_for_creation(endpoint_id)
return cfn_success_response(event)
def on_create(event, context):
log.info("Handle resource create event %s" % json.dumps(event, indent=2))
# get parameters passed in
props = event["ResourceProperties"]
nodes_per_zone = int(props["NodesPerZone"])
zone_count = int(props["NumberOfZones"])
parent_stack_name = props["ParentStackName"]
parent_stack_id = props["ParentStackId"]
subnets = props["Subnets"]
security_group_id = props["SecurityGroup"]
# prepare ENI meta information
id_hash = hashlib.md5(parent_stack_id.encode()).hexdigest()
eni_tag_prefix = parent_stack_name + "-" + id_hash + "_"
dns = []
# craete ENIs
for i in range(0, zone_count):
for j in range(0, nodes_per_zone):
eni_idx = i * nodes_per_zone + j
tag = eni_tag_prefix + str(eni_idx)
eni_info = create_eni(subnets[i], security_group_id, tag)
if not eni_info:
reason = "Failed to create network interface with tag %s" % tag
log.warning(reason)
continue
eni_id = eni_info["NetworkInterfaceId"]
eni_dns = eni_info["PrivateDnsName"]
eni_assign_tag(eni_id=eni_id, tag=tag)
dns.append(eni_dns)
return cfn_success_response(event, data={"Addresses": ",".join(dns)})
def on_delete(event, context):
log.info("Handle resource delete event %s " % json.dumps(event, indent=2))
# get parameters passed in
props = event["ResourceProperties"]
nodes_per_zone = int(props["NodesPerZone"])
zone_count = int(props["NumberOfZones"])
parent_stack_name = props["ParentStackName"]
parent_stack_id = props["ParentStackId"]
# prepare ENI meta information
id_hash = hashlib.md5(parent_stack_id.encode()).hexdigest()
eni_tag_prefix = parent_stack_name + "-" + id_hash + "_"
# delete ENIs
for i in range(0, zone_count):
for j in range(0, nodes_per_zone):
eni_idx = i * nodes_per_zone + j
tag = eni_tag_prefix + str(eni_idx)
log.info("Querying EC2 for ENI with tag %s" % tag)
# query
response = None
try:
response = ec2_client.describe_network_interfaces(
# TODO AWS SDK bug #1450
# Filters=[{
# "Name": "tag:cluster-eni-id",
# "Values": [tag]
# }]
Filters=[{
"Name": "description",
"Values": [tag]
}])
except ClientError as e:
reason = "Failed to describe network interface with tag %s" % tag
log.exception(reason)
time.sleep(5)
continue
for eni_info in response["NetworkInterfaces"]:
eni_id = eni_info["NetworkInterfaceId"]
log.info("Found network interface %s " % eni_id)
# detach
if "Attachment" in eni_info and (
eni_info["Attachment"]["Status"] == "attached"
or eni_info["Attachment"]["Status"] == "attaching"):
attachment_id = eni_info["Attachment"]["AttachmentId"]
if not detach_eni(eni_id, attachment_id):
reason = "Failed to detach network interface %s" % eni_id
log.error(reason)
try:
ec2_client.delete_network_interface(
NetworkInterfaceId=eni_id)
log.info("Deleting network interface %s" % eni_id)
except ClientError as e:
reason = "Failed to delete network interface %s" % eni_id
log.exception(reason)
return cfn_success_response(event)
def on_delete(event, context):
log.info("Handle resource delete event %s" % json.dumps(event, indent=2))
# get parameters passed in
props = event["ResourceProperties"]
service_name = props["ServiceName"]
vpc = props["Vpc"]
# find endpoint
response = None
try:
response = ec2_client.describe_vpc_endpoints(
Filters=[
{
"Name": "service-name",
"Values": [service_name]
},
{
"Name": "vpc-id",
"Values": [vpc]
}
]
)
except ClientError as e:
reason = "Failed to describe vpc endpoint for service %s" % service_name
log.exception(reason)
time.sleep(5) # sleep for 5 seconds to allow exception info being sent to CloudWatch
return cfn_failure_response(event, reason)
if len(response["VpcEndpoints"]) <= 0:
log.info("No endpoint found for service %s" % service_name)
else:
endpoint_id = response["VpcEndpoints"][0]["VpcEndpointId"]
try:
response = ec2_client.delete_vpc_endpoints(
VpcEndpointIds=[endpoint_id]
)
log.info("Deleting endpoint %s" % endpoint_id)
except ClientError as e:
reason = "Failed to delete vpc endpoint %s" % endpoint_id
log.exception(reason)
time.sleep(5)
return cfn_failure_response(event, reason)
if "Unsuccessful" in response and len(response["Unsuccessful"]) > 0:
reason = "Failed to delete vpc endpoint %s: %s" % (
endpoint_id,
response["Unsuccessful"][0]["Error"]["Message"]
)
log.error(reason)
time.sleep(5)
return cfn_failure_response(event, reason)
return cfn_success_response(event)