def __init__(self, provider_config, node_ids, cluster_name):
self.node_ids = set(node_ids)
self.cluster_name = cluster_name
self.provider_config = provider_config
region = provider_config["region"]
self.ec2_client = client_cache("ec2", region)
self.ssm_client = client_cache("ssm", region)
self.cloudwatch_client = client_cache("cloudwatch", region)
def __init__(self, provider_config, node_ids, cluster_name):
# dedupe and sort node IDs to support deterministic unit test stubs
self.node_ids = sorted(list(set(node_ids)))
self.cluster_name = cluster_name
self.provider_config = provider_config
region = provider_config["region"]
self.ec2_client = client_cache("ec2", region)
self.ssm_client = client_cache("ssm", region)
self.cloudwatch_client = client_cache("cloudwatch", region)
def __init__(self, provider_config: Dict[str, Any], node_id: str,
cluster_name: str) -> None:
self.node_id = node_id
self.cluster_name = cluster_name
self.provider_config = provider_config
region = provider_config["region"]
self.ec2_resource = resource_cache("ec2", region)
self.ec2_client = self.ec2_resource.meta.client
self.ssm_client = client_cache("ssm", region)
cloudwatch_resource = resource_cache("cloudwatch", region)
self.cloudwatch_client = cloudwatch_resource.meta.client
self.CLOUDWATCH_CONFIG_TYPE_TO_CONFIG_VARIABLE_REPLACE_FUNC = {
CloudwatchConfigType.AGENT: self._replace_cwa_config_variables,
CloudwatchConfigType.DASHBOARD:
self._replace_dashboard_config_variables,
CloudwatchConfigType.ALARM: self._load_config_file,
}
self.CLOUDWATCH_CONFIG_TYPE_TO_UPDATE_FUNC_HEAD_NODE = {
CloudwatchConfigType.AGENT: self._restart_cloudwatch_agent,
CloudwatchConfigType.DASHBOARD: self._put_cloudwatch_dashboard,
CloudwatchConfigType.ALARM: self._put_cloudwatch_alarm,
}
self.CLOUDWATCH_CONFIG_TYPE_TO_UPDATE_FUNC_WORKER_NODE = {
CloudwatchConfigType.AGENT: self._restart_cloudwatch_agent,
CloudwatchConfigType.ALARM: self._put_cloudwatch_alarm,
}
def list_ec2_instances(
region: str,
aws_credentials: Dict[str, Any] = None) -> List[Dict[str, Any]]:
"""Get all instance-types/resources available in the user's AWS region.
Args:
region (str): the region of the AWS provider. e.g., "us-west-2".
Returns:
final_instance_types: a list of instances. An example of one element in
the list:
{'InstanceType': 'm5a.xlarge', 'ProcessorInfo':
{'SupportedArchitectures': ['x86_64'], 'SustainedClockSpeedInGhz':
2.5},'VCpuInfo': {'DefaultVCpus': 4, 'DefaultCores': 2,
'DefaultThreadsPerCore': 2, 'ValidCores': [2],
'ValidThreadsPerCore': [1, 2]}, 'MemoryInfo': {'SizeInMiB': 16384},
...}
"""
final_instance_types = []
aws_credentials = aws_credentials or {}
ec2 = client_cache("ec2", region, BOTO_MAX_RETRIES, **aws_credentials)
instance_types = ec2.describe_instance_types()
final_instance_types.extend(copy.deepcopy(instance_types["InstanceTypes"]))
while "NextToken" in instance_types:
instance_types = ec2.describe_instance_types(
NextToken=instance_types["NextToken"])
final_instance_types.extend(
copy.deepcopy(instance_types["InstanceTypes"]))
return final_instance_types
def resolve_policy_arns(config: Dict[str, Any], iam: Any,
default_policy_arns: List[str]) -> List[str]:
"""Attach necessary AWS policies for CloudWatch related operations.
Args:
config: provider section of cluster config file.
iam: AWS iam resource.
default_policy_arns: List of default ray AWS policies.
Returns:
list of policy arns including additional policies for CloudWatch
related operations if cloudwatch agent config is specifed in
cluster config file.
"""
cwa_cfg_exists = CloudwatchHelper.cloudwatch_config_exists(
config, "agent")
if cwa_cfg_exists:
cloudwatch_managed_policy = {
"Version":
"2012-10-17",
"Statement": [{
"Effect":
"Allow",
"Action": [
"ssm:SendCommand",
"ssm:ListCommandInvocations",
"iam:PassRole",
],
"Resource":
"*",
}],
}
iam_client = iam.meta.client
iam_client.create_policy(
PolicyName="CloudwatchManagedPolicies",
PolicyDocument=json.dumps(cloudwatch_managed_policy),
)
sts_client = client_cache("sts", config["region"])
account_id = sts_client.get_caller_identity().get("Account")
managed_policy_arn = (
"arn:aws:iam::{}:policy/CloudwatchManagedPolicies".format(
account_id))
policy_waiter = iam_client.get_waiter("policy_exists")
policy_waiter.wait(
PolicyArn=managed_policy_arn,
WaiterConfig={
"Delay": 2,
"MaxAttempts": 200
},
)
new_policy_arns = copy.copy(default_policy_arns)
new_policy_arns.extend([
"arn:aws:iam::aws:policy/CloudWatchAgentAdminPolicy",
"arn:aws:iam::aws:policy/AmazonSSMManagedInstanceCore",
managed_policy_arn,
])
return new_policy_arns
else:
return default_policy_arns
def __init__(self, provider_config: Dict[str, Any], node_id: str,
cluster_name: str) -> None:
self.node_id = node_id
self.cluster_name = cluster_name
self.provider_config = provider_config
region = provider_config["region"]
self.ec2_resource = resource_cache("ec2", region)
self.ec2_client = self.ec2_resource.meta.client
self.ssm_client = client_cache("ssm", region)
cloudwatch_resource = resource_cache("cloudwatch", region)
self.cloudwatch_client = cloudwatch_resource.meta.client
def __init__(self, provider_config: Dict[str, Any], node_ids: List[str],
cluster_name: str) -> None:
# dedupe and sort node IDs to support deterministic unit test stubs
self.node_ids = sorted(set(node_ids))
self.cluster_name = cluster_name
self.provider_config = provider_config
region = provider_config["region"]
self.ec2_resource = resource_cache("ec2", region)
self.ec2_client = self.ec2_resource.meta.client
self.ssm_client = client_cache("ssm", region)
cloudwatch_resource = resource_cache("cloudwatch", region)
self.cloudwatch_client = cloudwatch_resource.meta.client
def ssm_client_stub():
client = client_cache("ssm", "us-west-2")
with Stubber(client) as stubber:
yield stubber
stubber.assert_no_pending_responses()
