def setup(num):
workspace_name = '%s-%s-%02d' % (workspace_prefix, location, num)
try:
ws = Workspace.get(
name=workspace_name,
subscription_id=subscription_id,
resource_group=resource_group)
print('Found existing workspace %s' % workspace_name)
except WorkspaceException:
print('Creating new workspace %s...' % workspace_name)
ws = Workspace.create(
name=workspace_name,
subscription_id=subscription_id,
resource_group=resource_group,
location=location)
try:
compute_target = AmlCompute(ws, compute_name)
print('Found existing compute %s' % compute_name)
compute_target.update(min_nodes=min_nodes, max_nodes=max_nodes)
except ComputeTargetException:
print('Creating new compute target %s...' % compute_name)
compute_config = AmlCompute.provisioning_configuration(vm_size=vm_size, min_nodes=min_nodes, max_nodes=max_nodes)
compute_target = ComputeTarget.create(ws, compute_name, compute_config)
compute_target.wait_for_completion(show_output=True, timeout_in_minutes=20)
ds = ws.get_default_datastore()
ds.upload("testdata")
dataset_name = 'sample_dataset'
if dataset_name not in ws.datasets:
data = Dataset.File.from_files(path=[(ds, 'testdata.txt')])
data.register(
workspace = ws,
name = dataset_name,
description = 'Sample data for load test')
print('Dataset successfully registered')
else:
print('Dataset already exists')
class ClusterConnector:
def __init__(
self,
workspace,
cluster_name,
ssh_key,
vm_type,
admin_username="******",
):
"""Thin wrapper class around azureml.core.compute.AmlCluster
Provides parallel ssh objects and helper for master node and all node commands
and file copies.
Usage:
>>> cc = ClusterConnector(workspace, "MyCluster", sshkey, "Standard_ND40rs_v2")
>>> cc.initialize(min_nodes=0, max_nodes=4, idle_timeout_secs=30)
>>> cluster = cc.cluster
>>> [print(node['name']) for node in cc.cluster.list_nodes()]
"""
self.cluster_name = cluster_name
self.workspace = workspace
self.ssh_key = ssh_key
self.vm_type = vm_type
self.admin_username = admin_username
enable_host_logger()
hlog = logging.getLogger("pssh.host_logger")
tstr = datetime.now().isoformat(timespec="minutes")
[
hlog.removeHandler(h) for h in hlog.handlers
if isinstance(h, logging.StreamHandler)
]
os.makedirs("clusterlogs", exist_ok=True)
self.logfile = "clusterlogs/{}_{}.log".format(self.workspace.name,
tstr)
hlog.addHandler(logging.FileHandler(self.logfile))
self.cluster = None
self._master_scp = None
self._master_ssh = None
self._all_ssh = None
def initialise(self, min_nodes=0, max_nodes=0, idle_timeout_secs=1800):
"""Initialise underlying AmlCompute cluster instance"""
self._create_or_update_cluster(min_nodes, max_nodes, idle_timeout_secs)
def _check_logs_emessage(self, host, port):
msg = "Remote command failed on {}:{}. For details see {}".format(
host, port, self.logfile)
return msg
def terminate(self):
print('Attempting to terminate cluster "{}"'.format(
colored(self.cluster_name, "green")))
try:
self.cluster.update(min_nodes=0,
max_nodes=0,
idle_seconds_before_scaledown=10)
self.cluster.wait_for_completion()
except ComputeTargetException as err:
raise RuntimeError(
"Failed to terminate cluster nodes ({})".format(err))
if len(self.cluster.list_nodes()):
raise RuntimeError(
"Failed to terminate cluster nodes (nodes still running)")
@property
def cluster_nodes(self):
self.cluster.refresh_state()
return sorted(self.cluster.list_nodes(), key=lambda n: n["port"])
def _create_or_update_cluster(self, min_nodes, max_nodes,
idle_timeout_secs):
try:
self.cluster = AmlCompute(workspace=self.workspace,
name=self.cluster_name)
print('Updating existing cluster "{}"'.format(
colored(self.cluster_name, "green")))
self.cluster.update(
min_nodes=min_nodes,
max_nodes=max_nodes,
idle_seconds_before_scaledown=idle_timeout_secs,
)
except ComputeTargetException:
print('Creating new cluster "{}"'.format(
colored(self.cluster_name, "green")))
cluster_config = AmlCompute.provisioning_configuration(
vm_size=self.vm_type,
min_nodes=min_nodes,
max_nodes=max_nodes,
idle_seconds_before_scaledown=idle_timeout_secs,
admin_username=self.admin_username,
admin_user_ssh_key=self.ssh_key,
remote_login_port_public_access="Enabled",
)
self.cluster = AmlCompute.create(self.workspace, self.cluster_name,
cluster_config)
self.cluster.wait_for_completion()
if len(self.cluster_nodes) < min_nodes:
sleep(30)
if len(self.cluster_nodes) < min_nodes:
raise RuntimeError("Failed to provision sufficient nodes")
def _copy_nodefile_to_nodes(self):
if len(self.cluster_nodes) == 1:
cprint("Single node cluster -- skipping IB config", "yellow")
return
print("Collecting cluster IB info")
outputs = self._all_ssh.run_command(
r'ifconfig ib0 | grep -oe "inet[^6][adr: ]*[0-9.]*" | cut -d" " -f2',
shell="bash -c",
)
self._all_ssh.join(outputs)
ibaddrs = []
for output in outputs:
host = output.host
port = output.client.port
if output.exit_code != 0:
print(list(output.stdout))
print(list(output.stderr))
raise RuntimeError("Failed to get IB ip for {}:{}".format(
host, port))
try:
ibaddr = list(output.stdout)[0].split()[0]
except IndexError:
raise RuntimeError("Failed to get IB ip for {}:{} - "
"No ib interface found!".format(host, port))
print("Mapping {}:{} -> {}".format(host, port, ibaddr))
if port == self._master_scp.port:
cprint("IB Master: {}".format(ibaddr), "green")
ibaddrs = [ibaddr] + ibaddrs
else:
ibaddrs.append(ibaddr)
with NamedTemporaryFile(delete=False, mode="wt") as nfh:
self.nodefile = nfh.name
for addr in ibaddrs:
nfh.write("{}\n".format(addr))
self.ibaddrs = ibaddrs
self.copy_to_all_nodes(self.nodefile, "./nodefile")
def _create_cluster_ssh_conns(self):
hostips = [n["publicIpAddress"] for n in self.cluster_nodes]
hostconfigs = [HostConfig(port=n["port"]) for n in self.cluster_nodes]
self._all_ssh = ParallelSSHClient(hostips,
host_config=hostconfigs,
user=self.admin_username)
self._master_ssh = ParallelSSHClient(hostips[:1],
host_config=hostconfigs[:1],
user=self.admin_username)
self._master_scp = SSHClient(hostips[0],
port=hostconfigs[0].port,
user=self.admin_username)
def copy_to_all_nodes(self, source, dest):
copy_jobs = self._all_ssh.copy_file(source, dest)
joinall(copy_jobs, raise_error=True)
def copy_to_master_node(self, source, dest):
self._master_scp.copy_file(source, dest)
def copy_from_master_node(self, source, dest):
self._master_scp.copy_remote_file(source, dest)
def run_on_all_nodes(self, command):
outputs = self._all_ssh.run_command(command, shell="bash -c")
self._all_ssh.join(outputs, consume_output=True)
for output in outputs:
if int(output.exit_code) != 0:
host = output.host
port = output.client.port
raise RuntimeError(self._check_logs_emessage(host, port))
def run_on_master_node(self, command):
outputs = self._master_ssh.run_command(command, shell="bash -c")
self._master_ssh.join(outputs)
for output in outputs:
if int(output.exit_code) != 0:
host = output.host
port = output.client.port
raise RuntimeError(self._check_logs_emessage(host, port))
def attempt_termination(self):
try:
self.terminate()
except RuntimeError as err:
print(colored("ERROR: {}\n\n", "red", attrs=["bold"]).format(err))
self.warn_unterminated()
def warn_unterminated(self):
print(
colored("WARNING: {}", "red", attrs=["bold"]).format(
colored(
"Cluster {} is still running - terminate manually to avoid "
"additional compute costs".format(
colored(self.cluster_name, "green")),
"red",
)))
# Check settings and redeploy if required settings have changed
print("Found existing cluster")
if cluster.vm_size.lower() != aml_settings["vm_size"].lower(
) or cluster.vm_priority.lower() != aml_settings["vm_priority"].lower():
cluster.delete()
cluster.wait_for_completion(show_output=True)
raise ComputeTargetException(
"Cluster is of incorrect size or has incorrect priority. Deleting cluster and provisioning a new one."
)
# Update AMLCompute
#if cluster.provisioning_configuration.min_nodes != aml_settings["min_nodes"] or cluster.provisioning_configuration.max_nodes != aml_settings["max_nodes"] or cluster.provisioning_configuration.idle_seconds_before_scaledown != aml_settings["idle_seconds_before_scaledown"]:
print("Updating settings of Cluster")
cluster.update(min_nodes=aml_settings["min_nodes"],
max_nodes=aml_settings["max_nodes"],
idle_seconds_before_scaledown=aml_settings[
"idle_seconds_before_scaledown"])
# Wait until the operation has completed
cluster.wait_for_completion(show_output=True)
print("Successfully updated Cluster definition")
except ComputeTargetException:
print("Loading failed")
print("Creating new AML Compute resource")
compute_config = AmlCompute.provisioning_configuration(
vm_size=aml_settings["vm_size"],
vm_priority=aml_settings["vm_priority"],
min_nodes=aml_settings["min_nodes"],
max_nodes=aml_settings["max_nodes"],
idle_seconds_before_scaledown=aml_settings[
def main():
# Loading input values
print("::debug::Loading input values")
parameters_file = os.environ.get("INPUT_PARAMETERSFILE",
default="workspace.json")
azure_credentials = os.environ.get("INPUT_AZURECREDENTIALS", default="{}")
azure_credentials = json.loads(azure_credentials)
# Loading parameters file
print("::debug::Loading parameters file")
parameters_file_path = os.path.join(".aml", parameters_file)
try:
with open(parameters_file_path) as f:
parameters = json.load(f)
except FileNotFoundError:
print(
f"::error::Could not find parameter file in {parameters_file_path}. Please provide a parameter file in your repository (e.g. .aml/workspace.json)."
)
return
# Loading Workspace
sp_auth = ServicePrincipalAuthentication(
tenant_id=azure_credentials.get("tenantId", ""),
service_principal_id=azure_credentials.get("clientId", ""),
service_principal_password=azure_credentials.get("clientSecret", ""))
try:
print("::debug::Loading existing Workspace")
ws = Workspace.get(
name=parameters.get("name", None),
subscription_id=azure_credentials.get("subscriptionId", ""),
resource_group=parameters.get("resourceGroup", None),
auth=sp_auth)
print("::debug::Successfully loaded existing Workspace")
except AuthenticationException as exception:
print(
f"::error::Could not retrieve user token. Please paste output of `az ad sp create-for-rbac --name <your-sp-name> --role contributor --scopes /subscriptions/<your-subscriptionId>/resourceGroups/<your-rg> --sdk-auth` as value of secret variable: AZURE_CREDENTIALS: {exception}"
)
return
except AuthenticationError as exception:
print(f"::error::Microsoft REST Authentication Error: {exception}")
return
except AdalError as exception:
print(
f"::error::Active Directory Authentication Library Error: {exception}"
)
return
except ProjectSystemException as exception:
print(f"::error::Workspace authorizationfailed: {exception}")
return
# TODO: Create compute if not existing.
try:
# Loading AMLCompute
print("::debug::Loading existing AML Compute")
cluster = AmlCompute(workspace=ws, name=parameters["name"])
# Check settings and redeploy if required settings have changed
print("::debug::Found existing cluster")
if cluster.vm_size.lower() != parameters["vm_size"].lower(
) or cluster.vm_priority.lower() != parameters["vm_priority"].lower():
cluster.delete()
cluster.wait_for_completion(show_output=True)
raise ComputeTargetException(
"Cluster is of incorrect size or has incorrect priority. Deleting cluster and provisioning a new one."
)
# Update AMLCompute
#if cluster.provisioning_configuration.min_nodes != aml_settings["min_nodes"] or cluster.provisioning_configuration.max_nodes != aml_settings["max_nodes"] or cluster.provisioning_configuration.idle_seconds_before_scaledown != aml_settings["idle_seconds_before_scaledown"]:
print("::debug::Updating settings of Cluster")
cluster.update(min_nodes=parameters["min_nodes"],
max_nodes=parameters["max_nodes"],
idle_seconds_before_scaledown=parameters[
"idle_seconds_before_scaledown"])
# Wait until the operation has completed
cluster.wait_for_completion(show_output=True)
print("::debug::Successfully updated Cluster definition")
except ComputeTargetException:
print("::debug::Loading failed")
print("::debug::Creating new AML Compute resource")
compute_config = AmlCompute.provisioning_configuration(
vm_size=parameters["vm_size"],
vm_priority=parameters["vm_priority"],
min_nodes=parameters["min_nodes"],
max_nodes=parameters["max_nodes"],
idle_seconds_before_scaledown=parameters[
"idle_seconds_before_scaledown"],
tags=parameters["tags"],
description=parameters["description"])
# Deploy to VNET if provided
if parameters["vnet_resource_group_name"] and parameters[
"vnet_name"] and parameters["subnet_name"]:
compute_config.vnet_resourcegroup_name = parameters[
"vnet_resource_group_name"]
compute_config.vnet_name = parameters["vnet_name"]
compute_config.subnet_name = parameters["subnet_name"]
# Set Credentials if provided
if parameters["admin_username"] and parameters["admin_user_password"]:
compute_config.admin_username = parameters["admin_username"]
compute_config.admin_user_password = parameters[
"admin_user_password"]
elif parameters["admin_username"] and parameters["admin_user_ssh_key"]:
compute_config.admin_username = parameters["admin_username"]
compute_config.admin_user_ssh_key = parameters[
"admin_user_ssh_key"]
# Create Compute Target
cluster = ComputeTarget.create(
workspace=ws,
name=parameters["name"],
provisioning_configuration=compute_config)
# Wait until the cluster is attached
cluster.wait_for_completion(show_output=True)
# Checking status of AMLCompute Cluster
print("::debug::Checking status of AMLCompute Cluster")
if cluster.provisioning_state == "Failed":
cluster.delete()
raise Exception(
"::debug::Deployment of AMLCompute Cluster failed with the following status: {} and logs: \n{}"
.format(cluster.provisioning_state, cluster.provisioning_errors))
print(parameters)
print(
"::debug::Successfully finished Azure Machine Learning Compute Action")
