def delete_network():
existing_vpcs = u.get_vpc_dict()
if VPC_NAME in existing_vpcs:
vpc = ec2.Vpc(existing_vpcs[VPC_NAME].id)
print("Deleting VPC %s (%s) subresources:" % (VPC_NAME, vpc.id))
for subnet in vpc.subnets.all():
try:
sys.stdout.write("Deleting subnet %s ... " % (subnet.id))
sys.stdout.write(response_type(subnet.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
for gateway in vpc.internet_gateways.all():
sys.stdout.write("Deleting gateway %s ... " % (gateway.id))
# todo: if instances are using VPC, this fails with
# botocore.exceptions.ClientError: An error occurred (DependencyViolation) when calling the DetachInternetGateway operation: Network vpc-ca4abab3 has some mapped public address(es). Please unmap those public address(es) before detaching the gateway.
sys.stdout.write('detached ... ' if u.is_good_response(
gateway.detach_from_vpc(VpcId=vpc.id)) else ' detach_failed ')
sys.stdout.write('deleted ' if u.is_good_response(gateway.delete(
)) else ' delete_failed ')
sys.stdout.write('\n')
def desc(route_table):
return "%s (%s)" % (route_table.id, u.get_name(route_table.tags))
for route_table in vpc.route_tables.all():
sys.stdout.write("Deleting route table %s ... " %
(desc(route_table)))
try:
sys.stdout.write(response_type(route_table.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
def desc(security_group):
return "%s (%s, %s)" % (security_group.id,
u.get_name(security_group.tags),
security_group.group_name)
# TODO: this tries to remove default security group, maybe not remove it?
for security_group in vpc.security_groups.all():
sys.stdout.write('Deleting security group %s ... ' %
(desc(security_group)))
try:
sys.stdout.write(response_type(security_group.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
sys.stdout.write("Deleting VPC %s ... " % (vpc.id))
try:
sys.stdout.write(response_type(vpc.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
def make_job(self,
role_name,
num_tasks=1,
skip_existing_job_validation=False,
**kwargs):
"""skip_existing_job_validation: if True, doesn't check that existing job on server has same number of tasks as requested."""
# u.maybe_create_resources()
assert num_tasks >= 0
# TODO: document launch parameters
job_name = u.format_job_name(role_name, self.name)
instance_type = kwargs['instance_type']
instances = u.lookup_aws_instances(job_name,
instance_type=instance_type)
kwargs = u.merge_kwargs(kwargs, self.kwargs)
ami = kwargs.get('ami', '')
ami_name = kwargs.get('ami_name', '')
availability_zone = kwargs.get('availability_zone', '')
if not availability_zone:
availability_zone = os.environ['ZONE']
placement_group = kwargs.get('placement_group', '')
# automatically generated placement_group_name
use_placement_group = kwargs.get('use_placement_group', False)
assert use_placement_group == False or placement_group == ''
if use_placement_group:
placement_group = self.placement_group_name
install_script = kwargs.get('install_script', '')
skip_efs_mount = kwargs.get('skip_efs_mount', False)
linux_type = kwargs.get('linux_type', 'ubuntu')
# TODO: use heuristics to tell linux type from AMI name
user_data = kwargs.get('user_data', '')
if user_data:
assert user_data.startswith('#!/bin/bash')
ebs = kwargs.get('ebs', '')
use_spot = kwargs.get('use_spot', False)
monitoring = kwargs.get('monitoring', True)
# always install tmux on Amazon linux types
# TODO: has no effect for some reason
# https://console.aws.amazon.com/support/v1?region=us-west-2#/case/?displayId=5256445351&language=en
if linux_type == 'amazon':
user_data += 'sudo yum install tmux -y'
if user_data:
user_data += '\necho userdata_ok >> /tmp/is_initialized\n'
# print("Using user_data", user_data)
# TODO: also make sure instance type is the same
if instances:
if not skip_existing_job_validation:
assert len(instances) == num_tasks, (
"Found job with same name %s(%s), but number of tasks %d doesn't match requested %d, kill job manually."
%
(job_name, instances[0].state, len(instances), num_tasks))
print("Found existing job " + job_name)
starting_instances = False
for i in instances:
if i.state['Name'] == 'stopped':
i.start()
starting_instances = True
# TODO: replace with proper wait loop
if starting_instances:
while True:
print("Waiting forever for instances to start")
time.sleep(10)
print(instances)
else:
print("Launching new job %s into VPC %s" %
(job_name, u.get_resource_name()))
assert not (
ami and ami_name
), "Must have only one of ami and ami_name, got " + ami + ", " + ami_name
assert ami or ami_name, "Must specify at least one of ami and ami_name"
if ami_name:
ami = u.lookup_ami_id(ami_name).id
security_group = u.get_security_group_dict()[u.get_resource_name()]
keypair = u.get_keypair_dict()[u.get_keypair_name()]
vpc = u.get_vpc_dict()[u.get_resource_name()]
subnet_dict = u.get_subnet_dict(vpc)
region = u.get_region()
assert availability_zone in subnet_dict, "Availability zone %s is not in subnet dict for current AWS default region %s, available subnets are %s. (hint, set AWS_DEFAULT_REGION=%s)" % (
availability_zone, region, ', '.join(
subnet_dict.keys()), availability_zone[:-1])
subnet = subnet_dict[availability_zone]
ec2 = u.create_ec2_resource()
u.maybe_create_placement_group(placement_group)
self.log("Requesting %d %s" % (num_tasks, instance_type))
args = {
'ImageId': ami,
'InstanceType': instance_type,
'MinCount': num_tasks,
'MaxCount': num_tasks,
'KeyName': keypair.name
}
# storage setup
if ebs: args['BlockDeviceMappings'] = ebs
# network setup
# TODO: get rid of zone? Zone seems to be required for constructor
# that allows to enable AssociatePublicIpAddress field
args['NetworkInterfaces'] = [{
'SubnetId': subnet.id,
'DeviceIndex': 0,
'AssociatePublicIpAddress': True,
'Groups': [security_group.id]
}]
placement_arg = {'AvailabilityZone': availability_zone}
if placement_group: placement_arg['GroupName'] = placement_group
args['Placement'] = placement_arg
if monitoring: args['Monitoring'] = {'Enabled': True}
args['UserData'] = user_data
if use_spot: instances = u.create_spot_instances(args)
else:
try:
instances = ec2.create_instances(**args)
except Exception as e:
print(f"Instance creation failed with ({e})")
print("Account number: ", u.get_account_number())
print("Region: ", u.get_region())
sys.exit()
assert instances
assert len(instances) == num_tasks
# TODO: make instances match their launch indices. This way
# tasks can figure out which # they are
for (task_num, instance) in enumerate(instances):
while True:
try:
# sometimes get "An error occurred (InvalidInstanceID.NotFound)"
# task_name = u.format_task_name(instance.ami_launch_index, role_name,
# self.name)
task_name = u.format_task_name(task_num, job_name)
instance.create_tags(Tags=u.make_name(task_name))
break
except Exception as e:
self.log(
"create_tags failed with %s, retrying in %d seconds"
% (str(e), TIMEOUT_SEC))
time.sleep(TIMEOUT_SEC)
job = Job(self,
job_name,
instances=instances,
install_script=install_script,
linux_type=linux_type,
user_data=user_data,
skip_efs_mount=skip_efs_mount)
self.jobs.append(job)
return job
def network_setup():
"""Creates VPC if it doesn't already exists, configures it for public
internet access, returns vpc, subnet, security_group"""
# from https://gist.github.com/nguyendv/8cfd92fc8ed32ebb78e366f44c2daea6
ec2 = u.create_ec2_resource()
existing_vpcs = u.get_vpc_dict()
zones = u.get_available_zones()
if VPC_NAME in existing_vpcs:
print("Reusing VPC " + VPC_NAME)
vpc = existing_vpcs[VPC_NAME]
subnets = list(vpc.subnets.all())
assert len(subnets) == len(
zones
), "Has %s subnets, but %s zones, something went wrong during resource creation, try delete_resources.py/create_resources.py" % (
len(subnets), len(zones))
else:
print("Creating VPC " + VPC_NAME)
vpc = ec2.create_vpc(CidrBlock='192.168.0.0/16')
# enable DNS on the VPC
response = vpc.modify_attribute(EnableDnsHostnames={"Value": True})
assert u.is_good_response(response)
response = vpc.modify_attribute(EnableDnsSupport={"Value": True})
assert u.is_good_response(response)
vpc.create_tags(Tags=u.make_name(VPC_NAME))
vpc.wait_until_available()
gateways = u.get_gateway_dict(vpc)
if DEFAULT_NAME in gateways:
print("Reusing gateways " + DEFAULT_NAME)
else:
print("Creating gateway " + DEFAULT_NAME)
ig = ec2.create_internet_gateway()
ig.attach_to_vpc(VpcId=vpc.id)
ig.create_tags(Tags=u.make_name(DEFAULT_NAME))
# check that attachment succeeded
# TODO: sometimes get
# AssertionError: vpc vpc-33d0804b is in state None
attach_state = u.get1(ig.attachments, State=-1, VpcId=vpc.id)
assert attach_state == 'available', "vpc %s is in state %s" % (
vpc.id, attach_state)
route_table = vpc.create_route_table()
route_table.create_tags(Tags=u.make_name(ROUTE_TABLE_NAME))
dest_cidr = '0.0.0.0/0'
route = route_table.create_route(DestinationCidrBlock=dest_cidr,
GatewayId=ig.id)
# check success
for route in route_table.routes:
# result looks like this
# ec2.Route(route_table_id='rtb-a8b438cf',
# destination_cidr_block='0.0.0.0/0')
if route.destination_cidr_block == dest_cidr:
break
else:
# sometimes get
# AssertionError: Route for 0.0.0.0/0 not found in [ec2.Route(route_table_id='rtb-cd9153b0', destination_cidr_block='192.168.0.0/16')]
# TODO: add a wait/retry?
assert False, "Route for %s not found in %s" % (dest_cidr,
route_table.routes)
assert len(zones) <= 16 # for cidr/20 to fit into cidr/16
ip = 0
for zone in zones:
cidr_block = '192.168.%d.0/20' % (ip, )
ip += 16
print("Creating subnet %s in zone %s" % (cidr_block, zone))
subnet = vpc.create_subnet(CidrBlock=cidr_block,
AvailabilityZone=zone)
subnet.create_tags(Tags=[{
'Key': 'Name',
'Value': f'{VPC_NAME}-subnet'
}, {
'Key': 'Region',
'Value': zone
}])
u.wait_until_available(subnet)
route_table.associate_with_subnet(SubnetId=subnet.id)
# Creates security group if necessary
existing_security_groups = u.get_security_group_dict()
if SECURITY_GROUP_NAME in existing_security_groups:
print("Reusing security group " + SECURITY_GROUP_NAME)
security_group = existing_security_groups[SECURITY_GROUP_NAME]
else:
print("Creating security group " + SECURITY_GROUP_NAME)
security_group = ec2.create_security_group(
GroupName=SECURITY_GROUP_NAME,
Description=SECURITY_GROUP_NAME,
VpcId=vpc.id)
security_group.create_tags(Tags=[{
"Key": "Name",
"Value": SECURITY_GROUP_NAME
}])
# allow ICMP access for public ping
security_group.authorize_ingress(CidrIp='0.0.0.0/0',
IpProtocol='icmp',
FromPort=-1,
ToPort=-1)
# open public ports
# always include SSH port which is required for basic functionality
assert 22 in PUBLIC_TCP_RANGES, "Must enable SSH access"
for port in PUBLIC_TCP_RANGES:
if u.is_list_or_tuple(port):
assert len(port) == 2
from_port, to_port = port
else:
from_port, to_port = port, port
response = security_group.authorize_ingress(IpProtocol="tcp",
CidrIp="0.0.0.0/0",
FromPort=from_port,
ToPort=to_port)
assert u.is_good_response(response)
for port in PUBLIC_UDP_RANGES:
if u.is_list_or_tuple(port):
assert len(port) == 2
from_port, to_port = port
else:
from_port, to_port = port, port
response = security_group.authorize_ingress(IpProtocol="udp",
CidrIp="0.0.0.0/0",
FromPort=from_port,
ToPort=to_port)
assert u.is_good_response(response)
# allow ingress within security group
# Authorizing ingress doesn't work with names in a non-default VPC,
# so must use more complicated syntax
# https://github.com/boto/boto3/issues/158
for protocol in ['icmp']:
try:
rule = {
'FromPort': -1,
'IpProtocol': protocol,
'IpRanges': [],
'PrefixListIds': [],
'ToPort': -1,
'UserIdGroupPairs': [{
'GroupId': security_group.id
}]
}
security_group.authorize_ingress(IpPermissions=[rule])
except Exception as e:
if e.response['Error'][
'Code'] == 'InvalidPermission.Duplicate':
print("Warning, got " + str(e))
else:
assert False, "Failed while authorizing ingress with " + str(
e)
for protocol in ['tcp', 'udp']:
try:
rule = {
'FromPort': 0,
'IpProtocol': protocol,
'IpRanges': [],
'PrefixListIds': [],
'ToPort': 65535,
'UserIdGroupPairs': [{
'GroupId': security_group.id
}]
}
security_group.authorize_ingress(IpPermissions=[rule])
except Exception as e:
if e.response['Error'][
'Code'] == 'InvalidPermission.Duplicate':
print("Warning, got " + str(e))
else:
assert False, "Failed while authorizing ingress with " + str(
e)
return vpc, security_group
def make_job(self, role_name, num_tasks=1, **kwargs):
assert num_tasks >= 0
# TODO: document launch parameters
job_name = u.format_job_name(role_name, self.name)
instances = u.lookup_aws_instances(job_name)
kwargs = u.merge_kwargs(kwargs, self.kwargs)
ami = kwargs['ami']
instance_type = kwargs['instance_type']
availability_zone = kwargs['availability_zone']
placement_group = kwargs.get('placement_group', '')
install_script = kwargs.get('install_script', '')
skip_efs_mount = kwargs.get('skip_efs_mount', False)
linux_type = kwargs.get('linux_type', 'ubuntu')
user_data = kwargs.get('user_data', '')
if user_data:
user_data += '\necho userdata_ok >> /tmp/is_initialized\n'
# print("Using user_data", user_data)
# TODO: also make sure instance type is the same
if instances:
assert len(instances) == num_tasks, (
"Found job with same name, but number of tasks %d doesn't match requested %d, kill job manually."
% (len(instances), num_tasks))
print("Found existing job " + job_name)
else:
print("Launching new job %s into VPC %s" %
(job_name, u.get_resource_name()))
security_group = u.get_security_group_dict()[u.get_resource_name()]
keypair = u.get_keypair_dict()[u.get_keypair_name()]
vpc = u.get_vpc_dict()[u.get_resource_name()]
subnet_dict = u.get_subnet_dict(vpc)
region = u.get_region()
assert availability_zone in subnet_dict, "Availability zone %s is not in subnet dict for current AWS default region %s, available subnets are %s. (hint, set AWS_DEFAULT_REGION)" % (
availability_zone, region, ', '.join(subnet_dict.keys()))
subnet = subnet_dict[availability_zone]
ec2 = u.create_ec2_resource()
u.maybe_create_placement_group(placement_group)
self.log("Requesting %d %s" % (num_tasks, instance_type))
args = {
'ImageId': ami,
'InstanceType': instance_type,
'MinCount': num_tasks,
'MaxCount': num_tasks,
'KeyName': keypair.name
}
# network setup
args['NetworkInterfaces'] = [{
'SubnetId': subnet.id,
'DeviceIndex': 0,
'AssociatePublicIpAddress': True,
'Groups': [security_group.id]
}]
placement_arg = {'AvailabilityZone': availability_zone}
if placement_group: placement_arg['GroupName'] = placement_group
args['Placement'] = placement_arg
args['UserData'] = user_data
instances = ec2.create_instances(**args)
assert len(instances) == num_tasks
# assign proper names to tasks
for instance in instances:
while True:
try:
# sometimes get "An error occurred (InvalidInstanceID.NotFound)"
task_name = u.format_task_name(
instance.ami_launch_index, role_name, self.name)
# TODO: use instance.create_tags instead like in create_resources.py
ec2.create_tags(Resources=[instance.id],
Tags=u.make_name(task_name))
break
except Exception as e:
self.log(
"create_tags failed with %s, retrying in %d seconds"
% (str(e), TIMEOUT_SEC))
time.sleep(TIMEOUT_SEC)
job = Job(self,
job_name,
instances=instances,
install_script=install_script,
linux_type=linux_type,
user_data=user_data,
skip_efs_mount=skip_efs_mount)
self.jobs.append(job)
return job
def server_job(name,
num_tasks=1,
instance_type=None,
install_script='',
placement_group='',
ami='',
availability_zone='',
linux_type=DEFAULT_LINUX_TYPE):
"""Creates a job on AWS cluster with publicly facing ports.
Reuse requires that that job launched previous under same name has identical
settings (number of tasks/instace type/placement group)
"""
global SSH_KEY_PATH
DEFAULT_NAME = u.RESOURCE_NAME
security_group = u.get_security_group_dict()[DEFAULT_NAME]
keypair = u.get_keypair_dict()[DEFAULT_NAME]
# get availability zone -> subnet dictionary
vpc = u.get_vpc_dict()[DEFAULT_NAME]
subnet_dict = {}
for subnet in vpc.subnets.all():
zone = subnet.availability_zone
assert zone not in subnet_dict, "More than one subnet in %s, why?" % (
zone, )
subnet_dict[zone] = subnet
subnet = subnet_dict[availability_zone]
global ROOT_INSTALL_SCRIPT
if linux_type == 'ubuntu':
ROOT_INSTALL_SCRIPT = ROOT_INSTALL_SCRIPT_UBUNTU
elif linux_type == 'amazon':
ROOT_INSTALL_SCRIPT = ROOT_INSTALL_SCRIPT_AMAZON
else:
assert False, "Unknown linux type '%s', expected 'ubuntu' or 'amazon'."
if instance_type is None:
instance_type = 'c5.large'
instances = lookup_aws_instances(name)
# todo: get rid of this global variable?
SSH_KEY_PATH = "%s/%s-%s.pem" % (
os.environ["HOME"],
DEFAULT_NAME,
os.environ['AWS_DEFAULT_REGION'],
)
if instances:
assert len(instances) == num_tasks, (
"Found job with same name, but number"
" of tasks %d doesn't match requested %d, kill job manually." %
(len(instances), num_tasks))
print("Found existing job " + name)
else:
print("Launching new job %s into VPC %s" % (name, DEFAULT_NAME))
ec2 = boto3.resource('ec2')
if placement_group:
_maybe_create_placement_group(placement_group)
print("Requesting %d %s" % (num_tasks, instance_type))
if not ami:
ami = os.environ.get('AMI', '')
assert ami, "No AMI specified, need AMI env-var or explicit parameter"
args = {
'ImageId': ami,
'InstanceType': instance_type,
'MinCount': num_tasks,
'MaxCount': num_tasks,
'KeyName': keypair.name
}
# network setup
args['NetworkInterfaces'] = [{
'SubnetId': subnet.id,
'DeviceIndex': 0,
'AssociatePublicIpAddress': True,
'Groups': [security_group.id]
}]
placement_arg = {'AvailabilityZone': availability_zone}
if placement_group: placement_arg['GroupName'] = placement_group
args['Placement'] = placement_arg
instances = ec2.create_instances(**args)
# todo: use task index in name
for instance in instances:
while True:
try:
# sometimes get "An error occurred (InvalidInstanceID.NotFound)"
tag = ec2.create_tags(Resources=[instance.id],
Tags=[{
'Key': 'Name',
'Value': name
}])
break
except Exception as e:
self.log(
"create_tags failed with %s, retrying in %d seconds" %
(str(e), TIMEOUT_SEC))
time.sleep(TIMEOUT_SEC)
assert len(instances) == num_tasks
print('{} Instances created'.format(len(instances)))
job = Job(name,
instances=instances,
install_script=install_script,
linux_type=linux_type)
return job
def main():
# TODO: also bring down all the instances and wait for them to come down
region = os.environ['AWS_DEFAULT_REGION']
if DEFAULT_NAME == 'nexus':
print("Nexus resources are protected, don't delete them")
sys.exit()
print("Deleting %s resources in region %s" % (
DEFAULT_NAME,
region,
))
existing_vpcs = u.get_vpc_dict()
client = u.create_ec2_client()
ec2 = u.create_ec2_resource()
def response_type(response):
return 'ok' if u.is_good_response(response) else 'failed'
# delete EFS
efss = u.get_efs_dict()
efs_id = efss.get(DEFAULT_NAME, '')
efs_client = u.create_efs_client()
if efs_id:
try:
# delete mount targets first
print("About to delete %s (%s)" % (efs_id, DEFAULT_NAME))
response = efs_client.describe_mount_targets(FileSystemId=efs_id)
assert u.is_good_response(response)
for mount_response in response['MountTargets']:
subnet = ec2.Subnet(mount_response['SubnetId'])
zone = subnet.availability_zone
state = mount_response['LifeCycleState']
id = mount_response['MountTargetId']
ip = mount_response['IpAddress']
sys.stdout.write('Deleting mount target %s ... ' % (id, ))
sys.stdout.flush()
response = efs_client.delete_mount_target(MountTargetId=id)
print(response_type(response))
sys.stdout.write('Deleting EFS %s (%s)... ' %
(efs_id, DEFAULT_NAME))
sys.stdout.flush()
u.delete_efs_id(efs_id)
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
if VPC_NAME in existing_vpcs:
vpc = ec2.Vpc(existing_vpcs[VPC_NAME].id)
print("Deleting VPC %s (%s) subresources:" % (VPC_NAME, vpc.id))
for subnet in vpc.subnets.all():
try:
sys.stdout.write("Deleting subnet %s ... " % (subnet.id))
sys.stdout.write(response_type(subnet.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
for gateway in vpc.internet_gateways.all():
sys.stdout.write("Deleting gateway %s ... " % (gateway.id))
# todo: if instances are using VPC, this fails with
# botocore.exceptions.ClientError: An error occurred (DependencyViolation) when calling the DetachInternetGateway operation: Network vpc-ca4abab3 has some mapped public address(es). Please unmap those public address(es) before detaching the gateway.
sys.stdout.write('detached ... ' if u.is_good_response(
gateway.detach_from_vpc(VpcId=vpc.id)) else ' detach_failed ')
sys.stdout.write('deleted ' if u.is_good_response(gateway.delete(
)) else ' delete_failed ')
sys.stdout.write('\n')
def desc(route_table):
return "%s (%s)" % (route_table.id, u.get_name(route_table.tags))
for route_table in vpc.route_tables.all():
sys.stdout.write("Deleting route table %s ... " %
(desc(route_table)))
try:
sys.stdout.write(response_type(route_table.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
def desc(security_group):
return "%s (%s, %s)" % (security_group.id,
u.get_name(security_group.tags),
security_group.group_name)
# TODO: this tries to remove default security group, maybe not remove it?
for security_group in vpc.security_groups.all():
sys.stdout.write('Deleting security group %s ... ' %
(desc(security_group)))
try:
sys.stdout.write(response_type(security_group.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
sys.stdout.write("Deleting VPC %s ... " % (vpc.id))
sys.stdout.write(response_type(vpc.delete()) + '\n')
# delete keypair
keypairs = u.get_keypair_dict()
keypair = keypairs.get(DEFAULT_NAME, '')
if keypair:
try:
sys.stdout.write("Deleting keypair %s (%s) ... " %
(keypair.key_name, DEFAULT_NAME))
sys.stdout.write(response_type(keypair.delete()) + '\n')
except Exception as e:
sys.stdout.write('failed\n')
u.loge(str(e) + '\n')
keypair_fn = u.get_keypair_fn(KEYPAIR_NAME)
if os.path.exists(keypair_fn):
print("Deleting local keypair file %s" % (keypair_fn, ))
os.system('rm -f ' + keypair_fn)