def __init__(self, name, type, description=None, metadata=None):
if not re.match("^[a-zA-Z0-9_\-.]+$", name):
raise ValueError(
'Artifact name may only contain alphanumeric characters, dashes, underscores, and dots. Invalid name: "%s"'
% name
)
# TODO: this shouldn't be a property of the artifact. It's a more like an
# argument to log_artifact.
storage_layout = StorageLayout.V2
if env.get_use_v1_artifacts():
storage_layout = StorageLayout.V1
self._storage_policy = WandbStoragePolicy(
config={
"storageLayout": storage_layout,
# TODO: storage region
}
)
self._api = InternalApi()
self._final = False
self._digest = None
self._file_entries = None
self._manifest = ArtifactManifestV1(self, self._storage_policy)
self._cache = get_artifacts_cache()
self._added_new = False
self._added_objs = {}
# You can write into this directory when creating artifact files
self._artifact_dir = compat_tempfile.TemporaryDirectory(
missing_ok_on_cleanup=True
)
self.type = type
self.name = name
self.description = description
self.metadata = metadata
def __init__(self, api):
self._api = api
self._tempdir = tempfile.TemporaryDirectory("wandb")
self._stats = stats.Stats()
self._incoming_queue = queue.Queue()
self._event_queue = queue.Queue()
self._step_checksum = step_checksum.StepChecksum(
self._api,
self._tempdir,
self._incoming_queue,
self._event_queue,
self._stats,
)
self._step_checksum.start()
self._step_upload = step_upload.StepUpload(self._api, self._stats,
self._event_queue,
self.MAX_UPLOAD_JOBS)
self._step_upload.start()
# Holds refs to tempfiles if users need to make a temporary file that
# stays around long enough for file pusher to sync
# TODO(artifacts): maybe don't do this
self._temp_file_refs = []
def __init__(self, name, type, description=None, metadata=None):
if not re.match('^[a-zA-Z0-9_\-.]+$', name):
raise ValueError(
'Artifact name may only contain alphanumeric characters, dashes, underscores, and dots. Invalid name: "%s"'
% name)
if type is None:
raise ValueError(
"type is required when logging artifacts, specify \"dataset\", \"model\", or a custom type"
)
# TODO: this shouldn't be a property of the artifact. It's a more like an
# argument to log_artifact.
self._storage_policy = WandbStoragePolicy()
self._file_specs = {}
self._api = InternalApi() # TODO: persist project in settings?
self._final = False
self._digest = None
self._file_entries = None
self._manifest = ArtifactManifestV1(self, self._storage_policy)
self._cache = artifacts_cache.get_artifacts_cache()
self._added_new = False
# You can write into this directory when creating artifact files
self._artifact_dir = compat_tempfile.TemporaryDirectory(
missing_ok_on_cleanup=True)
self.server_manifest = None
self.type = type
self.name = name
self.description = description
self.metadata = metadata
def main(argv):
args = parser.parse_args()
print('Load test starting')
project_name = args.project
if project_name is None:
project_name = 'artifacts-load-test-%s' % str(
datetime.now()).replace(' ', '-').replace(':', '-').replace('.', '-')
env_project = os.environ.get('WANDB_PROJECT')
sweep_id = os.environ.get('WANDB_SWEEP_ID')
if sweep_id:
del os.environ['WANDB_SWEEP_ID']
wandb_config_paths = os.environ.get('WANDB_CONFIG_PATHS')
if wandb_config_paths:
del os.environ['WANDB_CONFIG_PATHS']
wandb_run_id = os.environ.get('WANDB_RUN_ID')
if wandb_run_id:
del os.environ['WANDB_RUN_ID']
# set global entity and project before chdir'ing
from wandb.apis import InternalApi
api = InternalApi()
settings_entity = api.settings('entity')
settings_base_url = api.settings('base_url')
os.environ['WANDB_ENTITY'] = (os.environ.get('LOAD_TEST_ENTITY') or settings_entity)
os.environ['WANDB_PROJECT'] = project_name
os.environ['WANDB_BASE_URL'] = (os.environ.get('LOAD_TEST_BASE_URL') or settings_base_url)
# Change dir to avoid litering code directory
pwd = os.getcwd()
tempdir = tempfile.TemporaryDirectory()
os.chdir(tempdir.name)
artifact_name = 'load-artifact-' + ''.join(
random.choices(string.ascii_lowercase + string.digits, k=10))
print('Generating source data')
source_file_names = gen_files(
args.gen_n_files, args.gen_max_small_size, args.gen_max_large_size)
print('Done generating source data')
procs = []
stop_queue = multiprocessing.Queue()
stats_queue = multiprocessing.Queue()
# start all processes
# writers
for i in range(args.num_writers):
file_names = source_file_names
if args.non_overlapping_writers:
chunk_size = int(len(source_file_names) / args.num_writers)
file_names = source_file_names[i * chunk_size: (i+1) * chunk_size]
if args.distributed_fanout > 1:
p = multiprocessing.Process(
target=proc_version_writer_distributed,
args=(
stop_queue,
stats_queue,
project_name,
file_names,
artifact_name,
args.files_per_version_min,
args.files_per_version_max,
args.distributed_fanout,
args.blocking)
)
else:
p = multiprocessing.Process(
target=proc_version_writer,
args=(
stop_queue,
stats_queue,
project_name,
file_names,
artifact_name,
args.files_per_version_min,
args.files_per_version_max,
args.blocking)
)
p.start()
procs.append(p)
# readers
for i in range(args.num_readers):
p = multiprocessing.Process(
target=proc_version_reader,
args=(
stop_queue,
stats_queue,
project_name,
artifact_name,
i
)
)
p.start()
procs.append(p)
# deleters
for i in range(args.num_deleters):
p = multiprocessing.Process(
target=proc_version_deleter,
args=(
stop_queue,
stats_queue,
artifact_name,
args.min_versions_before_delete,
args.delete_period_max))
p.start()
procs.append(p)
# cache garbage collector
if args.cache_gc_period_max is None:
print('Testing cache GC process not enabled!')
else:
p = multiprocessing.Process(
target=proc_cache_garbage_collector,
args=(
stop_queue,
args.cache_gc_period_max))
p.start()
procs.append(p)
# reset environment
os.environ['WANDB_ENTITY'] = settings_entity
os.environ['WANDB_BASE_URL'] = settings_base_url
os.environ
if env_project is None:
del os.environ['WANDB_PROJECT']
else:
os.environ['WANDB_PROJECT'] = env_project
if sweep_id:
os.environ['WANDB_SWEEP_ID'] = sweep_id
if wandb_config_paths:
os.environ['WANDB_CONFIG_PATHS'] = wandb_config_paths
if wandb_run_id:
os.environ['WANDB_RUN_ID'] = wandb_run_id
# go back to original dir
os.chdir(pwd)
# test phase
start_time = time.time()
stats = defaultdict(int)
run = wandb.init(job_type='main-test-phase')
run.config.update(args)
while time.time() - start_time < args.test_phase_seconds:
stat_update = None
try:
stat_update = stats_queue.get(True, 5000)
except queue.Empty:
pass
print('** Test time: %s' % (time.time() - start_time))
if stat_update:
for k, v in stat_update.items():
stats[k] += v
wandb.log(stats)
print('Test phase time expired')
# stop all processes and wait til all are done
for i in range(len(procs)):
stop_queue.put(True)
print('Waiting for processes to stop')
fail = False
for proc in procs:
proc.join()
if proc.exitcode != 0:
print('FAIL! Test phase failed')
fail = True
sys.exit(1)
# drain remaining stats
while True:
try:
stat_update = stats_queue.get_nowait()
except queue.Empty:
break
for k, v in stat_update.items():
stats[k] += v
print('Stats')
import pprint
pprint.pprint(dict(stats))
if fail:
print('FAIL! Test phase failed')
sys.exit(1)
else:
print('Test phase successfully completed')
print('Starting verification phase')
os.environ['WANDB_ENTITY'] = (os.environ.get('LOAD_TEST_ENTITY') or settings_entity)
os.environ['WANDB_PROJECT'] = project_name
os.environ['WANDB_BASE_URL'] = (os.environ.get('LOAD_TEST_BASE_URL') or settings_base_url)
data_api = wandb.Api()
# we need list artifacts by walking runs, accessing via
# project.artifactType.artifacts only returns committed artifacts
for run in data_api.runs('%s/%s' % (api.settings('entity'), project_name)):
for v in run.logged_artifacts():
# TODO: allow deleted once we build deletion support
if v.state != 'COMMITTED' and v.state != 'DELETED':
print('FAIL! Artifact version not committed or deleted: %s' % v)
sys.exit(1)
print('Verification succeeded')
import time
from six.moves import queue
import warnings
import wandb
import wandb.util
from wandb.compat import tempfile
# Get rid of cleanup warnings in Python 2.7.
warnings.filterwarnings('ignore', 'Implicitly cleaning up', RuntimeWarning,
'wandb.compat.tempfile')
# Temporary directory for copies we make of some file types to
# reduce the probability that the file gets changed while we're
# uploading it.
TMP_DIR = tempfile.TemporaryDirectory('wandb')
EventFileChanged = collections.namedtuple('EventFileChanged',
('path', 'save_name', 'copy'))
EventJobDone = collections.namedtuple('EventJobDone', ('job'))
EventFinish = collections.namedtuple('EventFinish', ())
class UploadJob(threading.Thread):
def __init__(self, done_queue, push_function, save_name, path, copy=True):
"""A file upload thread.
Arguments:
done_queue: queue.Queue in which to put an EventJobDone event when
the upload finishes.
push_function: function(save_name, actual_path) which actually uploads
from pathlib import Path
return str(Path(path).resolve())
except:
# Pathlib isn't present for python versions earlier than 3.3
return os.path.realpath(path)
# Get rid of cleanup warnings in Python 2.7.
warnings.filterwarnings("ignore", "Implicitly cleaning up", RuntimeWarning,
"wandb.compat.tempfile")
# Temporary directory for copies we make of some file types to
# reduce the probability that the file gets changed while we're
# uploading it.
TMP_DIR = tempfile.TemporaryDirectory("wandb")
logger = logging.getLogger(__file__)
class FilePusher(object):
"""Parallel file upload class.
This manages uploading multiple files in parallel. It will restart a given file's
upload job if it receives a notification that that file has been modified.
The finish() method will block until all events have been processed and all
uploads are complete.
"""
MAX_UPLOAD_JOBS = 64
def __init__(self, api):
from wandb.sdk.internal import datastore
from wandb.sdk.internal import handler
from wandb.sdk.internal import sender
from wandb.sdk.internal import tb_watcher
from wandb.sdk.interface import interface
else:
from wandb.sdk_py27.internal import datastore
from wandb.sdk_py27.internal import handler
from wandb.sdk_py27.internal import sender
from wandb.sdk_py27.internal import tb_watcher
from wandb.sdk_py27.interface import interface
WANDB_SUFFIX = ".wandb"
SYNCED_SUFFIX = ".synced"
TFEVENT_SUBSTRING = ".tfevents."
TMPDIR = tempfile.TemporaryDirectory()
class _LocalRun(object):
def __init__(self, path, synced=None):
self.path = path
self.synced = synced
self.offline = os.path.basename(path).startswith("offline-")
self.datetime = datetime.datetime.strptime(
os.path.basename(path).split("run-")[1].split("-")[0],
"%Y%m%d_%H%M%S")
def __str__(self):
return self.path
from wandb.sdk import wandb_artifacts
else:
from wandb.sdk_py27 import wandb_run
from wandb.sdk_py27 import wandb_artifacts
return wandb_run, wandb_artifacts
# Get rid of cleanup warnings in Python 2.7.
warnings.filterwarnings(
"ignore", "Implicitly cleaning up", RuntimeWarning, "wandb.compat.tempfile"
)
# Staging directory so we can encode raw data into files, then hash them before
# we put them into the Run directory to be uploaded.
MEDIA_TMP = tempfile.TemporaryDirectory("wandb-media")
class Table(Media):
"""This is a table designed to display sets of records.
Arguments:
columns: ([str]) Names of the columns in the table.
Defaults to ["Input", "Output", "Expected"].
data: (array) 2D Array of values that will be displayed as strings.
dataframe: (pandas.DataFrame) DataFrame object used to create the table.
When set, the other arguments are ignored.
optional (Union[bool,List[bool]]): If None values are allowed. Singular bool
applies to all columns. A list of bool values applies to each respective column.
Default to True.
allow_mixed_types (bool): Determines if columns are allowed to have mixed types (disables type validation). Defaults to False
def main(argv):
print('Load test starting')
args = parser.parse_args()
# set global entity and project before chdir'ing
from wandb.apis import InternalApi
api = InternalApi()
os.environ['WANDB_ENTITY'] = api.settings('entity')
os.environ['WANDB_PROJECT'] = api.settings('project')
os.environ['WANDB_BASE_URL'] = api.settings('base_url')
# Change dir to avoid litering code directory
tempdir = tempfile.TemporaryDirectory()
os.chdir(tempdir.name)
artifact_name = 'load-artifact-' + ''.join(
random.choices(string.ascii_lowercase + string.digits, k=10))
print('Generating source data')
source_file_names = gen_files(args.gen_n_files, args.gen_max_small_size,
args.gen_max_large_size)
print('Done generating source data')
procs = []
stop_queue = multiprocessing.Queue()
# start all processes
# writers
for i in range(args.num_writers):
p = multiprocessing.Process(
target=proc_version_writer,
args=(stop_queue, source_file_names, artifact_name,
args.files_per_version_min, args.files_per_version_max))
p.start()
procs.append(p)
# readers
for i in range(args.num_readers):
p = multiprocessing.Process(target=proc_version_reader,
args=(stop_queue, artifact_name, i))
p.start()
procs.append(p)
# cache garbage collector
if args.cache_gc_period_max is None:
print('Testing cache GC process not enabled!')
else:
p = multiprocessing.Process(target=proc_cache_garbage_collector,
args=(stop_queue,
args.cache_gc_period_max))
p.start()
procs.append(p)
# test phase
time.sleep(args.test_phase_seconds)
print('Test phase time expired')
# stop all processes and wait til all are done
for i in range(len(procs)):
stop_queue.put(True)
print('Waiting for processes to stop')
for proc in procs:
proc.join()
if proc.exitcode != 0:
print('FAIL! Test phase failed')
sys.exit(1)
print('Test phase successfully completed')
print('Starting verification phase')
api = wandb.Api()
versions = api.artifact_versions('dataset', artifact_name)
for v in versions:
# TODO: allow deleted once we build deletion support
if v.state != 'COMMITTED':
print('FAIL! Artifact version not committed: %s' % v)
sys.exit(1)
print('Verification succeeded')
