def heartbeat(self):
while True:
try:
url = '%s/node/%s' % (self.grid_url, str(self.node_id))
request = JSONHTTPRequest('POST', url, {
'cpu': self.mon.cpu(),
}, self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(
e,
"Heatbeat Failed: Unable to establish a connection to the grid"
)
if self.retry_attempts < self.RETRY_MAX_ATTEMPTS:
try:
self.node_id = self.register_node()
except ServerUnavailableException as e:
self.retry_attempts += 1
else:
print "Unable to connect to The Grid after %d unsuccessful attempts" % (
self.RETRY_MAX_ATTEMPTS)
# Reset the node state to clean up any running tasks
self.reset_node_state()
thread.interrupt_main()
time.sleep(self.HEARTBEAT_INTERVAL)
def heartbeat(self):
while True:
try:
url = '%s/node/%s' % (self.grid_url, str(self.node_id))
request = JSONHTTPRequest( 'POST', url, {
'cpu': self.mon.cpu(),
}, self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(e,
"Heatbeat Failed: Unable to establish a connection to the grid")
if self.retry_attempts < self.RETRY_MAX_ATTEMPTS:
try:
self.node_id = self.register_node()
except ServerUnavailableException as e:
self.retry_attempts += 1
else:
print "Unable to connect to The Grid after %d unsuccessful attempts" % (self.RETRY_MAX_ATTEMPTS)
# Reset the node state to clean up any running tasks
self.reset_node_state()
thread.interrupt_main()
time.sleep(self.HEARTBEAT_INTERVAL)
def get_task_file(self, task): try: url = "%s/job/%s/files/%s" % (self.grid_url, task.job_id, task.filename) request = HTTPRequest( 'GET', url, "", self.auth_header) except (HTTPException, URLError) as e: node_utils.request_error_cli(e, "Unable to establish a connection to The Grid") sys.exit(1) # This will be horrible with large files fp = open(task.input_path, 'w+') fp.write(request.response) fp.close()
def get_task_file(self, task):
try:
url = "%s/job/%s/files/%s" % (self.grid_url, task.job_id,
task.filename)
request = HTTPRequest('GET', url, "", self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(
e, "Unable to establish a connection to The Grid")
sys.exit(1)
# This will be horrible with large files
fp = open(task.input_path, 'w+')
fp.write(request.response)
fp.close()
def send_task_output(self, task): # Send the results of stdout try: url = '%s/job/%s/output/%s' % (self.grid_url, str(task.job_id), task.output_name + ".o") request = FileHTTPRequest( 'PUT', url, task.output_path, self.auth_header ) except (HTTPException, URLError) as e: node_utils.request_error_cli(e, "Unable to establish a connection to the grid") # Send the results of stderr try: url = '%s/job/%s/output/%s' % (self.grid_url, str(task.job_id), task.output_name + ".e") request = FileHTTPRequest( 'PUT', url, task.error_path, self.auth_header ) except (HTTPException, URLError) as e: node_utils.request_error_cli(e, "Unable to establish a connection to the grid")
def register_node(self):
try:
request = JSONHTTPRequest( 'POST', self.grid_url + '/node', {
'host': self.host,
'port': self.port,
'cores': self.cores,
'programs': self.programs,
'cost': self.cost,
}, self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(e, "Unable to establish a connection to The Grid")
raise ServerUnavailableException("The Grid is currently unavailable.")
# Reset the node state in the event of Server failure
self.reset_node_state()
return request.response['node_id']
def finish_task(self, task, kill_msg = None):
self.send_task_output(task)
self.cleanup_task_files(task)
# Inform the server the task is complete
try:
url = '%s/job/%s/workunit' % (self.grid_url, str(task.job_id))
request = JSONHTTPRequest( 'POST', url, {
'work_unit_id': task.work_unit_id,
'kill_msg': kill_msg,
}, self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(e, "Unable to establish a connection to the grid")
# Update the task internally to reflect that the server has
# received all files and the complete status.
task.finish()
def finish_task(self, task, kill_msg=None):
self.send_task_output(task)
self.cleanup_task_files(task)
# Inform the server the task is complete
try:
url = '%s/job/%s/workunit' % (self.grid_url, str(task.job_id))
request = JSONHTTPRequest('POST', url, {
'work_unit_id': task.work_unit_id,
'kill_msg': kill_msg,
}, self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(
e, "Unable to establish a connection to the grid")
# Update the task internally to reflect that the server has
# received all files and the complete status.
task.finish()
def register_node(self):
try:
request = JSONHTTPRequest(
'POST', self.grid_url + '/node', {
'host': self.host,
'port': self.port,
'cores': self.cores,
'programs': self.programs,
'cost': self.cost,
}, self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(
e, "Unable to establish a connection to The Grid")
raise ServerUnavailableException(
"The Grid is currently unavailable.")
# Reset the node state in the event of Server failure
self.reset_node_state()
return request.response['node_id']
def send_task_output(self, task):
# Send the results of stdout
try:
url = '%s/job/%s/output/%s' % (self.grid_url, str(
task.job_id), task.output_name + ".o")
request = FileHTTPRequest('PUT', url, task.output_path,
self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(
e, "Unable to establish a connection to the grid")
# Send the results of stderr
try:
url = '%s/job/%s/output/%s' % (self.grid_url, str(
task.job_id), task.output_name + ".e")
request = FileHTTPRequest('PUT', url, task.error_path,
self.auth_header)
except (HTTPException, URLError) as e:
node_utils.request_error_cli(
e, "Unable to establish a connection to the grid")
