def _file_exists(storage_name, filename):
storage = api_util.get_storage(storage_name)
return storage.file_exists(filename)
def run_garbage_collector(api_name,
server_name,
redis_host="localhost",
redis_port=6379,
data_db=0,
pubsub_db=1):
signal.signal(signal.SIGTERM, raise_keyboard_interrupt)
signal.signal(signal.SIGINT, raise_keyboard_interrupt)
core.api_util.DatabaseContext.load_conf()
core.api_util.RedisContext.set_params(api_name, server_name, redis_host,
redis_port, data_db, pubsub_db)
# Loading providers and storages
conf = api_util.get_conf()
conf.read(os.path.join(API_PATH, 'config.conf'))
allowed_providers = json.loads(conf.get("general", "allowed_providers"))
providers = []
for provider_name in allowed_providers:
providers.append(api_util.get_provider(provider_name))
allowed_storages = json.loads(conf.get("general", "allowed_storages"))
storages = []
for storage_name in allowed_storages:
storages.append(api_util.get_storage(storage_name))
running_jobs = RunningJobs()
thread_list = []
for provider in providers:
running_workers = RunningWorkers()
worker_collector = WorkerCollector(provider, api_name, server_name,
running_jobs, running_workers)
worker_collector.start()
thread_list.append(worker_collector)
provider_artefact_collector = ProviderArtefactCollector(
provider, running_jobs, running_workers)
provider_artefact_collector.start()
thread_list.append(provider_artefact_collector)
job_collector = JobCollector(running_jobs)
job_collector.start()
thread_list.append(job_collector)
model_collector = ModelCollector()
model_collector.start()
thread_list.append(model_collector)
# FIXME: Disable for now
# file_collector = FileCollector(api_name)
# file_collector.start()
# thread_list.append(file_collector)
#
# for storage in storages:
# storage_collector = StorageCollector(storage)
# storage_collector.start()
# thread_list.append(storage_collector)
try:
while True:
time.sleep(0.1)
for proc in thread_list:
if not proc.is_alive():
proc.reraise()
except error_util.all_errors as e:
with error_util.before_raising():
if error_util.is_abort(e):
log.info("Signal received, exiting...")
else:
error_util.log_error(log, e)
log.info("Garbage collection cleaning...")
stop_and_join(thread_list)
log.info("Garbage collection is cleaned")
log.info("Garbage collection cleaning...")
stop_and_join(thread_list)
log.info("Garbage collection is cleaned")
def _delete_file(storage_name, filename):
storage = api_util.get_storage(storage_name)
storage.delete_file(filename)
def _get_file_creation_date(storage_name, filename):
storage = api_util.get_storage(storage_name)
return storage.get_file_creation_date(filename)
def _list_files(storage_name):
storage = api_util.get_storage(storage_name)
return storage.list_files()
def calculate(api_name, server_name, job, project, mesh, calculation,
calc_param_file, provider_name, machine, nbr_machines,
split_results, client_login, client_ip, api_version):
"""
Launch the machine(s), send the files, start the worker script, wait for progress and results and saving results
:param api_name: The name of the api
:type api_name: str
:param server_name: The name of the server (ex: apidev.zephycloud.com)
:type server_name: str
:param job: The job information
:type job: dict[str, any]
:param project: The main project
:type project: dict[str, any]
:param mesh: The mesh used for this calculation
:type mesh: dict[str, any]
:param calculation: The calculation to launch
:type calculation: dict[str, any]
:param calc_param_file: The main job parameter file
:type calc_param_file: str
:param provider_name: The name of the provider
:type provider_name: str
:param machine: The type of machine to launch
:type machine: str
:param nbr_machines: The number of machines to run
:type nbr_machines: int
:param split_results: Do you want the result file to be splitted ?
:type split_results: bool
:param client_login: The login of the job owner
:type client_login: str
:param client_ip: The client ip address of the http request string this job
:type client_ip: str
:param api_version: The version of the http api where the user ask to launch this job
:type api_version: str
"""
job_id = int(job['id'])
nbr_machines = int(nbr_machines)
models.jobs.set_job_status(job_id, models.jobs.JOB_STATUS_RUNNING)
project_codename = project['uid']
user_id = project["user_id"]
calc_id = calculation['id']
tmp_folder = api_util.get_conf().get("general", "tmp_folder")
provider = api_util.get_provider(provider_name)
storage = api_util.get_storage(project['storage'])
tags = {
'operation': "calc",
'job_id': str(job_id),
'server': server_name,
'api': api_name,
'api_version': api_version,
'client': client_login,
'client_ip': client_ip,
'debug': DO_NOT_KILL_INSTANCES,
'trusted': IS_TOOLCHAIN_SECURED
}
models.users.charge_user_fix_price(user_id, job_id,
"Calculation storage cost")
result_name = project_codename + "-calc-" + str(job_id)
result_file = cmd_util.ResultFile(project_codename, result_name + ".zip")
internal_file = cmd_util.ResultFile(project_codename,
result_name + "_workfiles.zip")
if split_results:
iterations_file = cmd_util.ResultFile(project_codename,
result_name + "_iterations.zip")
reduce_file = cmd_util.ResultFile(project_codename,
result_name + "_reduce.zip")
# Uploading file on cloud storage
log.info("Uploading param file to storage")
models.calc.save_calc_param_file(user_id, project_codename,
calculation['name'], calc_param_file)
try:
# Creating worker
with cmd_util.using_workers(
api_name,
provider,
job_id,
machine,
nbr_machines,
tags,
debug_keep_instances_alive=DO_NOT_KILL_INSTANCES) as workers:
# Launch main script
with cmd_util.TaskProcess(job_id, job["project_uid"], "calc",
workers, [split_results]) as task_proc:
conn = workers.ssh_connection
# Charge user
end_time = models.users.charge_user_computing(
user_id, job_id, "Cloud computation cost")
if models.users.get_credit(user_id) <= 0:
raise api_util.NoMoreCredits()
log.info("Sending project files on worker")
worker_in_storage = storages.SshStorage(
conn, api_util.WORKER_INPUT_PATH, IS_TOOLCHAIN_SECURED)
cmd_util.copy_project_file(
user_id,
project_codename,
storage,
worker_in_storage,
"project_file.zip",
tmp_folder,
key=models.projects.PROJECT_FILE_RAW)
cmd_util.copy_project_file(
user_id,
project_codename,
storage,
worker_in_storage,
"anal.zip",
tmp_folder,
key=models.projects.PROJECT_FILE_ANALYSED)
cmd_util.copy_project_file(user_id,
project_codename,
storage,
worker_in_storage,
"mesh.zip",
tmp_folder,
file_id=mesh['result_file_id'])
worker_in_storage.upload_file(calc_param_file,
"calc_params.zip")
os.remove(calc_param_file)
log.info("Project files sent to the worker")
# Tell the script to start
log.info("Starting the computation")
task_proc.start()
last_fetched_progress_time = datetime.datetime.utcfromtimestamp(
0)
is_stopped = False
while True:
task_status = task_proc.check_status()
# Charge if we need
if datetime.datetime.utcnow() > end_time:
end_time = models.users.charge_user_computing(
user_id, job_id, "Cloud computation cost")
if models.users.get_credit(user_id) <= 0:
models.jobs.save_job_text(job_id, "No more credit")
raise api_util.NoMoreCredits()
if task_status != models.jobs.JOB_STATUS_RUNNING:
log.info("Computation finished with status: " +
models.jobs.job_status_to_str(task_status))
break
if (datetime.datetime.utcnow() - last_fetched_progress_time
).seconds > STATUS_FETCHING_DELAY:
fetch_progress(conn, user_id, project_codename,
calculation['name'], calculation['id'],
storage, tmp_folder)
last_fetched_progress_time = datetime.datetime.utcnow()
if not is_stopped:
calculation = models.calc.get_calc(
user_id, project['uid'], calculation['id'])
if not calculation:
raise api_util.ToolchainError("Calculation " +
str(calc_id) +
" disappeared")
if calculation['status'] == models.calc.STATUS_STOPPED:
log.info("Stopping computation")
stop_calc(conn, project_codename)
is_stopped = True
time.sleep(1)
# Checking if the machine is still here
if not conn.ping():
models.jobs.save_job_text(job_id,
"Worker instance disappeared")
raise api_util.ToolchainError(
"Worker instance disappeared")
# Fetching computed data
log.info("Saving results")
worker_out_storage = storages.SshStorage(
conn, api_util.WORKER_OUTPUT_PATH, IS_TOOLCHAIN_SECURED)
log_file = util.path_join(api_util.WORKER_OUTPUT_PATH,
"worker.log")
if conn.file_exists(log_file):
with file_util.temp_filename(dir=tmp_folder) as tmp:
conn.get_file(log_file, tmp)
models.jobs.save_job_log(job_id, tmp)
else:
log.warning("No worker log file")
if not result_file.exists(worker_out_storage):
log.error("Unable to find file " + str(result_file) +
" on worker")
raise api_util.ToolchainError(
"Task failed, no result file")
result_file.save_on_storage(worker_out_storage, storage,
tmp_folder)
if split_results:
if not iterations_file.exists(worker_out_storage):
log.error("Unable to find file " +
str(iterations_file) + " on worker")
raise api_util.ToolchainError(
"Task failed, no result file")
iterations_file.save_on_storage(worker_out_storage,
storage, tmp_folder)
if not reduce_file.exists(worker_out_storage):
log.error("Unable to find file " + str(reduce_file) +
" on worker")
raise api_util.ToolchainError(
"Task failed, no result file")
reduce_file.save_on_storage(worker_out_storage, storage,
tmp_folder)
if internal_file.exists(worker_out_storage):
internal_file.save_on_storage(worker_out_storage, storage,
tmp_folder)
else:
log.warning("No internal files found on server")
fetch_progress(conn, user_id, project_codename,
calculation['name'], calculation['id'], storage,
tmp_folder)
log.info("Computation result fetched")
# Signaling all output was fetched
task_proc.stop_and_wait()
# Charge again if required
if datetime.datetime.utcnow() > end_time:
models.users.charge_user_computing(project["user_id"], job_id,
"Cloud computation cost")
# Uploading file on cloud storage
result_file.save_in_database(user_id)
internal_file_id = internal_file.save_in_database(
user_id) if internal_file.saved else None
if split_results:
iterations_file.save_in_database(user_id)
reduce_file.save_in_database(user_id)
models.calc.save_result_files(user_id, project_codename,
calculation['name'],
result_file.file_id,
iterations_file.file_id,
reduce_file.file_id,
internal_file_id)
else:
models.calc.save_result_files(user_id, project_codename,
calculation['name'],
result_file.file_id, None, None,
internal_file_id)
except error_util.all_errors:
with error_util.before_raising():
if REMOVE_RESULTS_ON_ERROR:
internal_file.delete_from_distant(storage)
result_file.delete_from_distant(storage)
if split_results:
iterations_file.delete_from_distant(storage)
reduce_file.delete_from_distant(storage)
log.info("Results saved")
def analyse(api_name, server_name, job, project, storage_name, project_file,
provider_name, machine, nbr_machines, client_login, client_ip,
api_version):
"""
Launch the machine(s), send the files, start the worker script, wait for progress and results and saving results
:param api_name: The name of the api
:type api_name: str
:param server_name: The name of the server (ex: apidev.zephycloud.com)
:type server_name: str
:param job: The job information
:type job: dict[str, any]
:param project: The main project
:type project: dict[str, any]
:param storage_name: The name of the storage where the project will be located
:type storage_name: str
:param project_file: The raw project file to analyse
:type project_file: str
:param provider_name: The name of the provider
:type provider_name: str
:param machine: The type of machine to launch
:type machine: str
:param nbr_machines: The number of machines to run
:type nbr_machines: int
:param client_login: The login of the job owner
:type client_login: str
:param client_ip: The client ip address of the http request string this job
:type client_ip: str
:param api_version: The version of the http api where the user ask to launch this job
:type api_version: str
"""
job_id = int(job['id'])
project_codename = job["project_uid"]
analyzed_filename = job["project_uid"] + "-anal-" + str(job_id) + ".zip"
user_id = project["user_id"]
provider = api_util.get_provider(provider_name)
storage = api_util.get_storage(storage_name)
tmp_folder = api_util.get_conf().get("general", "tmp_folder")
tags = {
'operation': "anal",
'job_id': str(job_id),
'server': server_name,
'api': api_name,
'api_version': api_version,
'client': client_login,
'client_ip': client_ip,
'trusted': IS_TOOLCHAIN_SECURED
}
# Uploading file on cloud storage
log.info("Uploading project file to storage")
models.projects.append_file_to_project(
user_id,
job["project_uid"],
project_file,
"project_" + job["project_uid"] + ".zip",
key=models.projects.PROJECT_FILE_RAW,
overwrite=True)
log.info("Project file uploaded")
models.users.charge_user_fix_price(user_id, job_id, "Project storage cost")
analyzed_file = cmd_util.ResultFile(project_codename, analyzed_filename)
try:
# Creating worker
with cmd_util.using_workers(
api_name,
provider,
job_id,
machine,
nbr_machines,
tags,
debug_keep_instances_alive=DO_NOT_KILL_INSTANCES) as workers:
with cmd_util.TaskProcess(job_id, job["project_uid"], "anal",
workers) as task_proc:
conn = workers.ssh_connection
# Charge user
end_time = models.users.charge_user_computing(
user_id, job_id, "Cloud computation cost")
if models.users.get_credit(user_id) <= 0:
raise api_util.NoMoreCredits()
log.info("Sending project files on worker")
conn.send_file(
project_file,
util.path_join(api_util.WORKER_INPUT_PATH,
"project_file.zip"))
os.remove(project_file)
log.info("Project files sent to the worker")
# Tell the script to start
log.info("Starting the computation")
task_proc.start()
while True:
task_status = task_proc.check_status()
# Charge if we need
if datetime.datetime.utcnow() > end_time:
end_time = models.users.charge_user_computing(
user_id, job_id, "Cloud computation cost")
if models.users.get_credit(user_id) <= 0:
models.jobs.save_job_text(job_id, "No more credit")
raise api_util.NoMoreCredits()
if task_status != models.jobs.JOB_STATUS_RUNNING:
log.info("Computation finished with status: " +
models.jobs.job_status_to_str(task_status))
break
time.sleep(5)
# Checking if the machine is still here
if not conn.ping():
models.jobs.save_job_text(job_id,
"Worker instance disappeared")
raise api_util.ToolchainError(
"Worker instance disappeared")
# Fetching computed data
log.info("Fetching results")
worker_out_storage = storages.SshStorage(
conn, api_util.WORKER_OUTPUT_PATH, IS_TOOLCHAIN_SECURED)
log_file = util.path_join(api_util.WORKER_OUTPUT_PATH,
"worker.log")
if conn.file_exists(log_file):
with file_util.temp_filename(dir=tmp_folder) as tmp:
conn.get_file(log_file, tmp)
models.jobs.save_job_log(job_id, tmp)
else:
log.warning("No worker log file")
if not analyzed_file.exists(worker_out_storage):
log.error("Unable to find file " + str(analyzed_file) +
" on worker")
raise api_util.ToolchainError(
"Task failed, no result file")
analyzed_file.save_on_storage(worker_out_storage, storage,
tmp_folder)
log.info("Computation result fetched")
# Signaling all output was fetched
task_proc.stop_and_wait()
# Charge again if required
if datetime.datetime.utcnow() > end_time:
models.users.charge_user_computing(project["user_id"], job_id,
"Cloud computation cost")
# Uploading file on cloud storage
analyzed_file.save_in_database(
user_id, key=models.projects.PROJECT_FILE_ANALYSED)
except error_util.all_errors:
with error_util.before_raising():
if REMOVE_RESULTS_ON_ERROR:
analyzed_file.delete_from_distant(storage)
log.info("Result saved")