update logic of create_and_deploy_minh.py

pull/37/head
Minh Doan 6 years ago
parent f7fdc9c317
commit cbba5732fe
  1. 351
      aws-experiment-launch/create_and_deploy_minh.py

@ -8,19 +8,14 @@ from threading import Thread
from Queue import Queue
import base64
from utils import utils
class InstanceResource:
ON_DEMAND = 1
SPOT_INSTANCE = 2
SPOT_FLEET = 3
REGION_NAME = 'region_name'
REGION_KEY = 'region_key'
REGION_SECURITY_GROUP = 'region_security_group'
REGION_HUMAN_NAME = 'region_human_name'
INSTANCE_TYPE = 't2.small'
REGION_AMI = 'region_ami'
with open("user-data.sh", "r") as userdata_file:
USER_DATA = userdata_file.read()
@ -36,350 +31,49 @@ CURRENT_SESSION = datetime.datetime.fromtimestamp(
PLACEMENT_GROUP = "PLACEMENT-" + CURRENT_SESSION
NODE_NAME_SUFFIX = "NODE-" + CURRENT_SESSION
"""
TODO:
save NODE to disk, so that you can selectively only run deploy (not recreate instances).
Right now all instances have "NODE" so this has uninted consequences of running on instances that were previous created.
Build (argparse,functions) support for
1. run only create instance (multiple times)
2. run only codedeploy (multiple times)
3. run create instance followed by codedeploy
"""
### CREATE INSTANCES ###
def run_one_region_instances(config, region_number, number_of_instances, instance_resource=InstanceResource.ON_DEMAND):
#todo: explore the use ec2 resource and not client. e.g. create_instances -- Might make for better code.
"""
e.g. ec2.create_instances
"""
region_name = config[region_number][REGION_NAME]
region_name = config[region_number][utils.REGION_NAME]
session = boto3.Session(region_name=region_name)
ec2_client = session.client('ec2')
print utils.get_one_availability_zone(ec2_client)
if instance_resource == InstanceResource.ON_DEMAND:
NODE_NAME = create_instances(
config, ec2_client, region_number, int(number_of_instances))
print("Created %s in region %s"%(NODE_NAME,region_number)) ##REPLACE ALL print with logger
elif instance_resource == InstanceResource.SPOT_INSTANCE:
response = request_spot_instances(
response, node_name_tag = create_instances(
config, ec2_client, region_number, int(number_of_instances))
print("Created %s in region %s" % (node_name_tag, region_number))
else:
response = request_spot_fleet(
config, ec2_client, region_number, int(number_of_instances))
return session
return None, node_name_tag
return response, node_name_tag
def create_instances(config, ec2_client, region_number, number_of_instances):
NODE_NAME = region_number + "-" + NODE_NAME_SUFFIX
node_name_tag = region_number + "-" + NODE_NAME_SUFFIX
response = ec2_client.run_instances(
MinCount=number_of_instances,
MaxCount=number_of_instances,
ImageId=config[region_number][REGION_AMI],
ImageId=config[region_number][utils.REGION_AMI],
Placement={
'AvailabilityZone': get_one_availability_zone(ec2_client),
'AvailabilityZone': utils.get_one_availability_zone(ec2_client),
},
SecurityGroups=[config[region_number][REGION_SECURITY_GROUP]],
SecurityGroups=[config[region_number][utils.REGION_SECURITY_GROUP]],
IamInstanceProfile={
'Name': IAM_INSTANCE_PROFILE
},
KeyName=config[region_number][REGION_KEY],
KeyName=config[region_number][utils.REGION_KEY],
UserData=USER_DATA,
InstanceType=INSTANCE_TYPE,
InstanceType=utils.INSTANCE_TYPE,
TagSpecifications=[
{
'ResourceType': 'instance',
'Tags': [
{
'Key': 'Name',
'Value': NODE_NAME
'Value': node_name_tag
},
]
},
],
# We can also request spot instances this way but this way will block the
# process until spot requests are fulfilled, otherwise it will throw exception
# after 4 failed re-try.
# InstanceMarketOptions= {
# 'MarketType': 'spot',
# 'SpotOptions': {
# 'SpotInstanceType': 'one-time',
# 'BlockDurationMinutes': 60,
# }
# }
)
return NODE_NAME
def request_spot_instances(config, ec2_client, region_number, number_of_instances):
NODE_NAME = region_number + "-" + NODE_NAME_SUFFIX
response = ec2_client.request_spot_instances(
# DryRun=True,
BlockDurationMinutes=60,
InstanceCount=number_of_instances,
LaunchSpecification={
'SecurityGroups': [config[region_number][REGION_SECURITY_GROUP]],
'IamInstanceProfile': {
'Name': IAM_INSTANCE_PROFILE
},
'UserData': USER_DATA_BASE64,
'ImageId': config[region_number][REGION_AMI],
'InstanceType': INSTANCE_TYPE,
'KeyName': config[region_number][REGION_KEY],
'Placement': {
'AvailabilityZone': get_one_availability_zone(ec2_client)
}
}
)
return response
def request_spot_fleet(config, ec2_client, region_number, number_of_instances):
NODE_NAME = region_number + "-" + NODE_NAME_SUFFIX
# https://boto3.readthedocs.io/en/latest/reference/services/ec2.html#EC2.Client.request_spot_fleet
response = ec2_client.request_spot_fleet(
# DryRun=True,
SpotFleetRequestConfig={
# https://docs.aws.amazon.com/AWSEC2/latest/UserGuide/spot-fleet.html#spot-fleet-allocation-strategy
'AllocationStrategy': 'diversified',
# 'IamFleetRole': IAM_INSTANCE_PROFILE, // TODO@ricl, create fleet role.
'LaunchSpecifications': [
{
'SecurityGroups': [
{
'GroupName': config[region_number][REGION_SECURITY_GROUP]
}
],
'IamInstanceProfile': {
'Name': IAM_INSTANCE_PROFILE
},
'ImageId': config[region_number][REGION_AMI],
'InstanceType': INSTANCE_TYPE,
'KeyName': config[region_number][REGION_KEY],
'Placement': {
'AvailabilityZone': get_one_availability_zone(ec2_client)
},
'UserData': USER_DATA,
# 'WeightedCapacity': 123.0,
'TagSpecifications': [
{
'ResourceType': 'instance',
'Tags': [
{
'Key': 'Name',
'Value': NODE_NAME
},
]
}
]
},
],
# 'SpotPrice': 'string', # The maximum price per unit hour that you are willing to pay for a Spot Instance. The default is the On-Demand price.
'TargetCapacity': 1,
'OnDemandTargetCapacity': 0,
'Type': 'maintain',
}
)
return response
def get_availability_zones(ec2_client):
response = ec2_client.describe_availability_zones()
all_zones = []
if response.get('AvailabilityZones', None):
region_info = response.get('AvailabilityZones')
for info in region_info:
if info['State'] == 'available':
all_zones.append(info['ZoneName'])
return all_zones
def get_one_availability_zone(ec2_client):
all_zones = get_availability_zones(ec2_client)
if len(all_zones) > 0:
return all_zones[0]
else:
print("No availability zone for this region")
sys.exit()
#### CODEDEPLOY ###
def run_one_region_codedeploy(region_number, commitId):
#todo: explore the use ec2 resource and not client. e.g. create_instances -- Might make for better code.
"""
for getting instance ids:---
ec2 = boto3.resource('ec2', region_name=region_name])
result = ec2.instances.filter(Filters=[{'Name': 'instance-state-name', 'Values': ['running']}])
for instance in result:
instances.append(instance.id)
for getting public ips : --
ec2 = boto3.resource('ec2')
instance
"""
region_name = config[region_number][REGION_NAME]
NODE_NAME = region_number + "-" + NODE_NAME_SUFFIX
session = boto3.Session(region_name=region_name)
ec2_client = session.client('ec2')
filters = [{'Name': 'tag:Name','Values': [NODE_NAME]}]
instance_ids = get_instance_ids(ec2_client.describe_instances(Filters=filters))
print("Number of instances: %d" % len(instance_ids))
print("Waiting for all %d instances in region %s to start running"%(len(instance_ids),region_number))
waiter = ec2_client.get_waiter('instance_running')
waiter.wait(InstanceIds=instance_ids)
print("Waiting for all %d instances in region %s to be INSTANCE STATUS OK"%(len(instance_ids),region_number))
waiter = ec2_client.get_waiter('instance_status_ok')
waiter.wait(InstanceIds=instance_ids)
print("Waiting for all %d instances in region %s to be SYSTEM STATUS OK"%(len(instance_ids),region_number))
waiter = ec2_client.get_waiter('system_status_ok')
waiter.wait(InstanceIds=instance_ids)
codedeploy = session.client('codedeploy')
application_name = APPLICATION_NAME
deployment_group = APPLICATION_NAME + "-" + str(commitId)[6] + "-" + CURRENT_SESSION
repo = REPO
print("Setting up to deploy commitId %s on region %s"%(commitId,region_number))
response = get_application(codedeploy, application_name)
deployment_group = get_deployment_group(
codedeploy, region_number, application_name, deployment_group)
depId = deploy(codedeploy, application_name,
deployment_group, repo, commitId)
return region_number, depId
def get_deployment_group(codedeploy, region_number,application_name, deployment_group):
NODE_NAME = region_number + "-" + NODE_NAME_SUFFIX
response = codedeploy.create_deployment_group(
applicationName=application_name,
deploymentGroupName=deployment_group,
deploymentConfigName='CodeDeployDefault.AllAtOnce',
serviceRoleArn='arn:aws:iam::656503231766:role/BenchMarkCodeDeployServiceRole',
deploymentStyle={
'deploymentType': 'IN_PLACE',
'deploymentOption': 'WITHOUT_TRAFFIC_CONTROL'
},
ec2TagFilters = [
{
'Key': 'Name',
'Value': NODE_NAME,
'Type': 'KEY_AND_VALUE'
}
]
)
return deployment_group
def get_application(codedeploy, application_name):
response = codedeploy.list_applications()
if application_name in response['applications']:
return response
else:
response = codedeploy.create_application(
applicationName=application_name,
computePlatform='Server'
)
return response
def deploy(codedeploy, application_name, deployment_group, repo, commitId):
"""Deploy new code at specified revision to instance.
arguments:
- repo: GitHub repository path from which to get the code
- commitId: commit ID to be deployed
- wait: wait until the CodeDeploy finishes
"""
print("Launching CodeDeploy with commit " + commitId)
res = codedeploy.create_deployment(
applicationName=application_name,
deploymentGroupName=deployment_group,
deploymentConfigName='CodeDeployDefault.AllAtOnce',
description='benchmark experiments',
revision={
'revisionType': 'GitHub',
'gitHubLocation': {
'repository': repo,
'commitId': commitId,
}
}
)
depId = res["deploymentId"]
print("Deployment ID: " + depId)
# The deployment is launched at this point, so exit unless asked to wait
# until it finishes
info = {'status': 'Created'}
start = time.time()
while info['status'] not in ('Succeeded', 'Failed', 'Stopped',) and (time.time() - start < 600.0):
info = codedeploy.get_deployment(deploymentId=depId)['deploymentInfo']
print(info['status'])
time.sleep(15)
if info['status'] == 'Succeeded':
print("\nDeploy Succeeded")
return depId
else:
print("\nDeploy Failed")
print(info)
return depId
def run_one_region_codedeploy_wrapper(region_number, commitId, queue):
region_number, depId = run_one_region_codedeploy(region_number, commitId)
queue.put((region_number, depId))
def launch_code_deploy(region_list, commitId):
queue = Queue()
jobs = []
for i in range(len(region_list)):
region_number = region_list[i]
my_thread = Thread(target=run_one_region_codedeploy_wrapper, args=(
region_number, commitId, queue))
my_thread.start()
jobs.append(my_thread)
for my_thread in jobs:
my_thread.join()
results = [queue.get() for job in jobs]
return results
##### UTILS ####
def get_instance_ids(describe_instances_response):
instance_ids = []
for reservation in describe_instances_response["Reservations"]:
for instance in reservation["Instances"]:
instance_ids.append(instance["InstanceId"])
return instance_ids
def read_configuration_file(filename):
config = {}
with open(filename, 'r') as f:
for myline in f:
mylist = myline.strip().split(',')
region_num = mylist[0]
config[region_num] = {}
config[region_num][REGION_NAME] = mylist[1]
config[region_num][REGION_KEY] = mylist[2]
config[region_num][REGION_SECURITY_GROUP] = mylist[3]
config[region_num][REGION_HUMAN_NAME] = mylist[4]
config[region_num][REGION_AMI] = mylist[5]
return config
def get_commitId(commitId):
if commitId is None:
commitId = run("git rev-list --max-count=1 HEAD",
hide=True).stdout.strip()
print("Got newest commitId as " + commitId)
return commitId
##### UTILS ####
return response, node_name_tag
if __name__ == "__main__":
parser = argparse.ArgumentParser(
@ -390,20 +84,19 @@ if __name__ == "__main__":
default='1', help='number of instances')
parser.add_argument('--configuration', type=str,
dest='config', default='configuration.txt')
parser.add_argument('--commitId', type=str, dest='commitId',
default='1f7e6e7ca7cf1c1190cedec10e791c01a29971cf')
args = parser.parse_args()
config = read_configuration_file(args.config)
config = utils.read_region_config(args.config)
region_list = args.regions.split(',')
instances_list = args.numInstances.split(',')
assert len(region_list) == len(instances_list), "number of regions: %d != number of instances per region: %d" % (
len(region_list), len(instances_list))
commitId = args.commitId
assert len(region_list) == len(instances_list), "number of regions: %d != number of instances per region: %d" % (len(region_list), len(instances_list))
for i in range(len(region_list)):
region_number = region_list[i]
number_of_instances = instances_list[i]
session = run_one_region_instances(
response, node_name_tag = run_one_region_instances(
config, region_number, number_of_instances, InstanceResource.ON_DEMAND)
if response:
print utils.get_ip_list(response)
# Enable the code below later.
# results = launch_code_deploy(region_list, commitId)

Loading…
Cancel
Save