def _swift_spark_execution(self, master, key_path, sahara, connector,
job_binary_name, job_binary_url, user, password,
job_template_name, job_type, plugin,
cluster_size, args, main_class, cluster_id,
spark_applications_ids, workers_id, app_id,
expected_time, monitor_plugin, collect_period,
number_of_jobs, log_path, swift, container,
data, number_of_attempts):
# Preparing job
job_binary_id = self._get_job_binary_id(sahara, connector,
job_binary_name,
job_binary_url, user, password)
mains = [job_binary_id]
job_template_id = self._get_job_template_id(sahara, connector, mains,
job_template_name,
job_type)
self._log("%s | Starting job..." % (time.strftime("%H:%M:%S")))
# Running job
# What is os_utils?
# configs = os_utils.get_job_config(connector, plugin,
# cluster_size, user, password,
# args, main_class)
configs = None
job = connector.create_job_execution(sahara,
job_template_id,
cluster_id,
configs=configs)
self._log("%s | Created job" % (time.strftime("%H:%M:%S")))
spark_app_id = spark.get_running_app(master, spark_applications_ids,
number_of_attempts)
spark_applications_ids.append(spark_app_id)
self._log("%s | Spark app id" % (time.strftime("%H:%M:%S")))
job_exec_id = job.id
for worker_id in workers_id:
instances_log.log("%s|%s" % (app_id, worker_id))
job_status = connector.get_job_status(sahara, job_exec_id)
self._log("%s | Sahara job status: %s" %
(time.strftime("%H:%M:%S"), job_status))
info_plugin = {
"spark_submisson_url": "http://" + master,
"expected_time": expected_time,
"number_of_jobs": number_of_jobs
}
self._log("%s | Starting monitor" % (time.strftime("%H:%M:%S")))
monitor.start_monitor(api.monitor_url, spark_app_id, monitor_plugin,
info_plugin, collect_period)
self._log("%s | Starting controller" % (time.strftime("%H:%M:%S")))
controller.start_controller(api.controller_url, spark_app_id,
workers_id, data)
job_status = self._wait_on_job_finish(sahara, connector, job_exec_id,
app_id)
self._log("%s | Stopping monitor" % (time.strftime("%H:%M:%S")))
monitor.stop_monitor(api.monitor_url, spark_app_id)
self._log("%s | Stopping controller" % (time.strftime("%H:%M:%S")))
controller.stop_controller(api.controller_url, spark_app_id)
spark_applications_ids.remove(spark_app_id)
self._log("Finished application execution")
if connector.is_job_completed(job_status):
self.update_application_state("OK")
if connector.is_job_failed(job_status):
self.update_application_state("Error")
return job_status
def _hdfs_spark_execution(self, master, remote_hdfs, key_path, args,
job_bin_url, main_class, dependencies,
spark_applications_ids, expected_time,
monitor_plugin, collect_period, number_of_jobs,
workers_id, data, connector, swift, swift_logdir,
container, number_of_attempts):
job_exec_id = str(uuid.uuid4())[0:7]
self._log("%s | Job execution ID: %s" %
(time.strftime("%H:%M:%S"), job_exec_id))
# Defining params
local_path = '/tmp/spark-jobs/' + job_exec_id + '/'
# remote_path = 'ubuntu@' + master + ':' + local_path
job_input_paths, job_output_path, job_params = (hdfs.get_job_params(
key_path, remote_hdfs, args))
job_binary_path = hdfs.get_path(job_bin_url)
# Create temporary job directories
self._log("%s | Create temporary job directories" %
(time.strftime("%H:%M:%S")))
self._mkdir(local_path)
# Create cluster directories
self._log("%s | Creating cluster directories" %
(time.strftime("%H:%M:%S")))
remote.execute_command(master, key_path, 'mkdir -p %s' % local_path)
# Get job binary from hdfs
self._log("%s | Get job binary from hdfs" %
(time.strftime("%H:%M:%S")))
remote.copy_from_hdfs(master, key_path, remote_hdfs, job_binary_path,
local_path)
# Enabling event log on cluster
self._log("%s | Enabling event log on cluster" %
(time.strftime("%H:%M:%S")))
self._enable_event_log(master, key_path, local_path)
# Submit job
self._log("%s | Starting job" % (time.strftime("%H:%M:%S")))
local_binary_file = (
local_path + remote.list_directory(key_path, master, local_path))
spark_job = self._submit_job(master, key_path, main_class,
dependencies, local_binary_file, args)
spark_app_id = spark.get_running_app(master, spark_applications_ids,
number_of_attempts)
if spark_app_id is None:
self._log("%s | Error on submission of application, "
"please check the config file" %
(time.strftime("%H:%M:%S")))
(output, err) = spark_job.communicate()
self.stdout.log(output)
self.stderr.log(err)
raise ex.ConfigurationError()
spark_applications_ids.append(spark_app_id)
info_plugin = {
"spark_submisson_url": "http://" + master,
"expected_time": expected_time,
"number_of_jobs": number_of_jobs
}
self._log("%s | Starting monitor" % (time.strftime("%H:%M:%S")))
monitor.start_monitor(api.monitor_url, spark_app_id, monitor_plugin,
info_plugin, collect_period)
self._log("%s | Starting controller" % (time.strftime("%H:%M:%S")))
controller.start_controller(api.controller_url, spark_app_id,
workers_id, data)
(output, err) = spark_job.communicate()
self._log("%s | Stopping monitor" % (time.strftime("%H:%M:%S")))
monitor.stop_monitor(api.monitor_url, spark_app_id)
self._log("%s | Stopping controller" % (time.strftime("%H:%M:%S")))
controller.stop_controller(api.controller_url, spark_app_id)
self.stdout.log(output)
self.stderr.log(err)
self._log("%s | Copy log from cluster" % (time.strftime("%H:%M:%S")))
event_log_path = local_path + 'eventlog/'
self._mkdir(event_log_path)
remote_event_log_path = 'ubuntu@%s:%s%s' % (master, local_path,
spark_app_id)
remote.copy(key_path, remote_event_log_path, event_log_path)
self._log("%s | Upload log to Swift" % (time.strftime("%H:%M:%S")))
connector.upload_directory(swift, event_log_path, swift_logdir,
container)
spark_applications_ids.remove(spark_app_id)
self.update_application_state("OK")
return 'OK'
def start_application(self, data):
try:
self.update_application_state("Running")
plugin_log.log("%s | Starting application execution" %
(time.strftime("%H:%M:%S")))
binary_url = str(data['binary_url'])
execution_class = str(data['execution_class'])
execution_parameters = str(data['execution_parameters'])
expected_time = int(data['expected_time'])
number_of_jobs = int(data['number_of_jobs'])
starting_cap = int(data['starting_cap'])
# Optimizer integration
app_name = data['app_name']
days = 0
if app_name.lower() == 'bulma':
if 'days' in data.keys():
days = data['days']
else:
self._log("""%s | 'days' parameter missing""" %
(time.strftime("%H:%M:%S")))
raise ex.ConfigurationError()
cores, vms = optimizer.get_info(api.optimizer_url, expected_time,
app_name, days)
optimizer_command = ''
if cores >= 0:
optimizer_command = ' --total-executor-cores %d ' % cores
plugin_log.log("%s | Submission id: %s" %
(time.strftime("%H:%M:%S"), self.app_id))
plugin_log.log("%s | Connecting with Mesos cluster..." %
(time.strftime("%H:%M:%S")))
conn = ssh.get_connection(api.mesos_url, api.cluster_username,
api.cluster_password,
api.cluster_key_path)
plugin_log.log("%s | Connected with Mesos cluster" %
(time.strftime("%H:%M:%S")))
# Execute all the spark needed commands
# to run an spark job from command line
if execution_class != "" and execution_class is not None:
# If the class field is empty, it means that the
# job binary is python
binary_path = '~/exec_bin.jar'
spark_run = ('sudo %s --name %s ' + '--master mesos://%s:%s ' +
optimizer_command + '--class %s %s %s')
else:
binary_path = '~/exec_bin.py'
spark_run = ('sudo %s --name %s ' + '--master mesos://%s:%s ' +
optimizer_command + '%s %s %s')
plugin_log.log("%s | Download the binary to cluster" %
(time.strftime("%H:%M:%S")))
try:
stdin, stdout, stderr = conn.exec_command(
'wget %s -O %s' % (binary_url, binary_path))
plugin_log.log("%s | Waiting for download the binary..." %
(time.strftime("%H:%M:%S")))
# TODO: Fix possible wget error
stdout.read()
plugin_log.log("%s | Binary downloaded" %
(time.strftime("%H:%M:%S")))
except Exception as e:
plugin_log.log("%s | Error downloading binary" %
(time.strftime("%H:%M:%S")))
self.update_application_state("Error")
return "Error"
i, o, e = conn.exec_command(
spark_run %
(api.spark_path, self.app_id, api.mesos_url, api.mesos_port,
execution_class, binary_path, execution_parameters))
# Discovery ips of the executors from Mesos
# and discovery the ids on KVM using the ips
list_vms_one = (
'onevm list --user %s --password %s --endpoint %s' %
(api.one_username, api.one_password, api.one_url))
stdin, stdout, stderr = conn.exec_command(list_vms_one)
list_response = stdout.read()
vms_ips, master = mesos.get_executors_ip(conn, self.frameworks_url,
self.app_id)
plugin_log.log("%s | Master: %s" %
(time.strftime("%H:%M:%S"), master))
plugin_log.log("%s | Executors: %s" %
(time.strftime("%H:%M:%S"), vms_ips))
vms_ids = mesos.extract_vms_ids(list_response)
plugin_log.log("%s | Executors IDs: %s" %
(time.strftime("%H:%M:%S"), vms_ids))
executors_vms_ids = []
for ip in vms_ips:
for id in vms_ids:
vm_info_one = (
'onevm show %s '
'--user %s '
'--password %s '
'--endpoint %s' %
(id, api.one_username, api.one_password, api.one_url))
stdin, stdout, stderr = conn.exec_command(vm_info_one)
if ip in stdout.read():
executors_vms_ids.append(id)
break
plugin_log.log("%s | Executors IDs: %s" %
(time.strftime("%H:%M:%S"), executors_vms_ids))
# Set up the initial configuration of cpu cap
controller.setup_environment(api.controller_url, executors_vms_ids,
starting_cap, data)
info_plugin = {
"spark_submisson_url": master,
"expected_time": expected_time,
"number_of_jobs": number_of_jobs
}
plugin_log.log("%s | Starting monitor" %
(time.strftime("%H:%M:%S")))
monitor.start_monitor(api.monitor_url, self.app_id, 'spark-mesos',
info_plugin, 2)
plugin_log.log("%s | Starting controller" %
(time.strftime("%H:%M:%S")))
controller.start_controller(api.controller_url, self.app_id,
executors_vms_ids, data)
# This command locks the plugin execution
# until the execution be done
print o.read()
plugin_log.log("%s | Stopping monitor" %
(time.strftime("%H:%M:%S")))
monitor.stop_monitor(api.monitor_url, self.app_id)
plugin_log.log("%s | Stopping controller" %
(time.strftime("%H:%M:%S")))
controller.stop_controller(api.controller_url, self.app_id)
plugin_log.log("%s | Remove binaries" %
(time.strftime("%H:%M:%S")))
conn.exec_command('rm -rf ~/exec_bin.*')
plugin_log.log("%s | Finished application execution" %
(time.strftime("%H:%M:%S")))
self.update_application_state("OK")
return 'OK'
except Exception as e:
plugin_log.log(e.message)
print e.message
self.update_application_state("Error")
def start_application(self, data):
try:
self.update_application_state("Running")
user = api.user
password = api.password
project_id = api.project_id
auth_ip = api.auth_ip
domain = api.domain
public_key = api.public_key
connector = os_connector.OpenStackConnector(LOG)
nova = connector.get_nova_client(user, password, project_id,
auth_ip, domain)
monitor_plugin = data['monitor_plugin']
expected_time = data['expected_time']
log_path = data['log_path']
image_id = data['image_id']
flavor_id = data['flavor_id']
command = data['command']
cluster_size = data['cluster_size']
starting_cap = data['scaling_parameters']["starting_cap"]
app_start_time = 0
app_end_time = 0
LOG.log("Creating instance(s)")
print "Creating instance(s)..."
# Create a number of instances to run the application based on
# cluster_size, image_id, flavor_id and public_key
instances = self._create_instances(nova, connector, image_id,
flavor_id, public_key,
cluster_size)
LOG.log("Waiting until instance become active...")
print "Waiting until instance become active..."
# Retrive network information from all instances when they
# reach ACTIVE state
instances_nets = []
for instance_id in instances:
instance_status = connector.get_instance_status(
nova, instance_id)
while instance_status != 'ACTIVE':
instance_status = connector.get_instance_status(
nova, instance_id)
instance_ips = connector.get_instance_networks(
nova, instance_id)
instances_nets.append(instance_ips)
time.sleep(5)
time.sleep(30)
LOG.log("Checking if ssh is available")
print "Checking if ssh is available"
# Verify if ssh is available for any ip address for each instance
instances_ips = []
for instance_net in instances_nets:
for net_ip_list in instance_net.values():
for ip in net_ip_list:
attempts = 2
while attempts != -1:
try:
conn = self._get_ssh_connection(
ip, api.key_path)
instances_ips.append(ip)
attempts = -1
except Exception as e:
LOG.log("Fail to connect")
LOG.log(e.message)
print "Fail to connect"
print e.message
attempts -= 1
time.sleep(30)
LOG.log("Setting up environment")
print "Setting up environment"
# Set CPU cap in all instances
controller.setup_environment(api.controller_url, instances,
starting_cap, data)
# Execute application and start monitor and controller service.
applications = []
for ip in instances_ips:
LOG.log("Executing commands into the instance")
print "Executing commands into the instance"
# TODO Check if exec_command will work without blocking exec
conn = self._get_ssh_connection(ip, api.key_path)
conn.exec_command(command)
app_start_time = time.time()
app_id = "app-os-generic" + str(uuid.uuid4())[:8]
applications.append(app_id)
monitor_plugin = monitor_plugin
info_plugin = {
"host_ip": ip,
"log_path": log_path,
"expected_time": expected_time
}
collect_period = 1
try:
LOG.log("Starting monitoring")
print "Starting monitoring"
monitor.start_monitor(api.monitor_url, app_id,
monitor_plugin, info_plugin,
collect_period)
LOG.log("Starting scaling")
print "Starting scaling"
controller.start_controller(api.controller_url, app_id,
instances, data)
except Exception as e:
LOG.log(e.message)
print e.message
# Stop monitor and controller when each application stops
application_running = True
while application_running:
status_instances = []
for instance_id in instances:
status = connector.get_instance_status(nova, instance_id)
status_instances.append(status)
if self._instances_down(status_instances):
application_running = False
app_end_time = time.time()
LOG.log("Application finished")
print "Application finished"
for app_id in applications:
LOG.log("Stopping monitoring")
print "Stopping monitoring"
monitor.stop_monitor(api.monitor_url, app_id)
LOG.log("Stopping scaling")
print "Stopping scaling"
controller.stop_controller(api.controller_url, app_id)
else:
instance_status = []
time.sleep(2)
LOG.log("Removing instances...")
print "Removing instances..."
# Remove instances after the end of all applications
self._remove_instances(nova, connector, instances)
LOG.log("Finished application execution")
print "Finished application execution"
application_time = app_end_time - app_start_time
application_time_log.log(
"%s|%.0f|%.0f" % (app_id, app_start_time, application_time))
self.application_time = application_time
self.start_time = app_start_time
self.update_application_state("OK")
return str(application_time)
except Exception as e:
LOG.log(e.message)
print e.message
self.update_application_state("Error")
def start_monitoring(self, data, collect_period=1):
monitor.start_monitor(api.monitor_url, self.app_id,
data['monitor_plugin'], data['monitor_info'],
collect_period)
def start_application(self, data):
try:
# Download files that contains the items
jobs = requests.get(data['redis_workload']).text.\
split('\n')[:-1]
# Provision a redis database for the job. Die in case of error.
# TODO(clenimar): configure ``timeout`` via a request param,
# e.g. api.redis_creation_timeout.
redis_ip, redis_port = self.k8s.provision_redis_or_die(self.app_id)
#agent_port = k8s.create_cpu_agent(self.app_id)
# inject REDIS_HOST in the environment
data['env_vars']['REDIS_HOST'] = 'redis-%s' % self.app_id
# inject SCONE_CONFIG_ID in the environment
# FIXME: make SCONE_CONFIG_ID optional in submission
data['env_vars']['SCONE_CONFIG_ID'] = data['config_id']
# create a new Redis client and fill the work queue
if (self.rds == None):
self.rds = redis.StrictRedis(host=redis_ip, port=redis_port)
queue_size = len(jobs)
# Check if a visualizer will be created
self.enable_visualizer = data['enable_visualizer']
# Create all visualizer components
if self.enable_visualizer:
# Specify the datasource to be used in the visualization
datasource_type = data['visualizer_info']['datasource_type']
if datasource_type == "influxdb":
database_data = k8s.create_influxdb(self.app_id)
#TODO {javan} change name of redis_ip to node_ip in configuration file
database_data.update({"url": api.redis_ip})
data['monitor_info'].update(
{'database_data': database_data})
data['visualizer_info'].update(
{'database_data': database_data})
data['monitor_info'].update(
{'datasource_type': datasource_type})
print "Creating Visualization plataform"
data['visualizer_info'].update({
'enable_visualizer':
data['enable_visualizer'],
'plugin':
data['monitor_plugin'],
'visualizer_plugin':
data['visualizer_plugin'],
'username':
data['username'],
'password':
data['password']
})
visualizer.start_visualization(api.visualizer_url, self.app_id,
data['visualizer_info'])
self.visualizer_url = visualizer.get_visualizer_url(
api.visualizer_url, self.app_id)
print "Dashboard of the job created on: %s" % (
self.visualizer_url)
print "Creating Redis queue"
for job in jobs:
self.rds.rpush("job", job)
print "Creating Job"
self.k8s.create_job(self.app_id,
data['cmd'],
data['img'],
data['init_size'],
data['env_vars'],
config_id=data["config_id"])
starting_time = datetime.datetime.now().\
strftime('%Y-%m-%dT%H:%M:%S.%fGMT')
# Starting monitor
data['monitor_info'].update({
'count_jobs_url':
api.count_queue,
'number_of_jobs':
queue_size,
'submission_time':
starting_time,
'redis_ip':
redis_ip,
'redis_port':
redis_port,
'enable_visualizer':
self.enable_visualizer
}) #,
#'cpu_agent_port': agent_port})
monitor.start_monitor(api.monitor_url, self.app_id,
data['monitor_plugin'], data['monitor_info'],
2)
# Starting controller
data.update({'redis_ip': redis_ip, 'redis_port': redis_port})
controller.start_controller_k8s(api.controller_url, self.app_id,
data)
while not self.job_completed and not self.terminated:
self.update_application_state("ongoing")
self.job_completed = self.k8s.completed(self.app_id)
time.sleep(1)
# Stop monitor, controller and visualizer
if (self.get_application_state() == "ongoing"):
self.update_application_state("completed")
print "Job finished"
time.sleep(float(30))
if self.enable_visualizer:
visualizer.stop_visualization(api.visualizer_url, self.app_id,
data['visualizer_info'])
monitor.stop_monitor(api.monitor_url, self.app_id)
controller.stop_controller(api.controller_url, self.app_id)
print "Stoped services"
# delete redis resources
if not self.get_application_state() == 'terminated':
self.k8s.delete_redis_resources(self.app_id)
except Exception as ex:
self.update_application_state("error")
print "ERROR: %s" % ex
print "Application finished."
def start_application(self, data):
try:
# Download files that contains the items
jobs = requests.get(data['redis_workload']).text.\
split('\n')[:-1]
# Provision a redis database for the job. Die in case of error.
# TODO(clenimar): configure ``timeout`` via a request param,
# e.g. api.redis_creation_timeout.
redis_ip, redis_port = k8s.provision_redis_or_die(self.app_id)
self.rds = redis.StrictRedis(host=redis_ip, port=redis_port)
queue_size = len(jobs)
print "Creating Redis queue"
for job in jobs:
self.rds.rpush("job", job)
print "Creating Job"
k8s.create_job(self.app_id,
data['cmd'],
data['img'],
data['init_size'],
data['env_vars'],
config_id=data["config_id"])
starting_time = datetime.datetime.now().\
strftime('%Y-%m-%dT%H:%M:%S.%fGMT')
# Starting monitor
data['monitor_info'].update({
'count_jobs_url':
api.count_queue,
'number_of_jobs':
queue_size,
'submission_time':
starting_time,
'redis_ip':
redis_ip,
'redis_port':
redis_port,
'graphic_metrics':
data['graphic_metrics']
})
monitor.start_monitor(api.monitor_url, self.app_id,
data['monitor_plugin'], data['monitor_info'],
2)
# Starting controller
data.update({'redis_ip': redis_ip, 'redis_port': redis_port})
controller.start_controller_k8s(api.controller_url, self.app_id,
data)
while not self.job_completed and not self.terminated:
self.update_application_state("ongoing")
self.job_completed = k8s.completed(self.app_id)
time.sleep(1)
# Stop monitor and controller
if (self.get_application_state() == "ongoing"):
self.update_application_state("completed")
print "job finished"
monitor.stop_monitor(api.monitor_url, self.app_id)
controller.stop_controller(api.controller_url, self.app_id)
print "stoped services"
# delete redis resources
time.sleep(float(30))
if not self.get_application_state() == 'terminated':
k8s.delete_redis_resources(self.app_id)
except Exception as ex:
self.update_application_state("error")
print "ERROR: %s" % ex
print "Application finished."
