def to_air_checkpoint(self) -> Optional[Checkpoint]:
from ray.tune.trainable.util import TrainableUtil
checkpoint_data = self.dir_or_data
if not checkpoint_data:
return None
if isinstance(checkpoint_data, ray.ObjectRef):
checkpoint_data = ray.get(checkpoint_data)
if isinstance(checkpoint_data, str):
checkpoint_dir = TrainableUtil.find_checkpoint_dir(checkpoint_data)
checkpoint = Checkpoint.from_directory(checkpoint_dir)
elif isinstance(checkpoint_data, bytes):
with tempfile.TemporaryDirectory() as tmpdir:
TrainableUtil.create_from_pickle(checkpoint_data, tmpdir)
# Double wrap in checkpoint so we hold the data in memory and
# can remove the temp directory
checkpoint = Checkpoint.from_dict(
Checkpoint.from_directory(tmpdir).to_dict())
elif isinstance(checkpoint_data, dict):
checkpoint = Checkpoint.from_dict(checkpoint_data)
else:
raise RuntimeError(
f"Unknown checkpoint data type: {type(checkpoint_data)}")
return checkpoint
def save(self, checkpoint_dir: Optional[str] = None) -> str:
"""Saves the current model state to a checkpoint.
Subclasses should override ``save_checkpoint()`` instead to save state.
This method dumps additional metadata alongside the saved path.
If a remote checkpoint dir is given, this will also sync up to remote
storage.
Args:
checkpoint_dir: Optional dir to place the checkpoint.
Returns:
str: path that points to xxx.pkl file.
Note the return path should match up with what is expected of
`restore()`.
"""
checkpoint_dir = TrainableUtil.make_checkpoint_dir(
checkpoint_dir or self.logdir, index=self.iteration)
checkpoint_dict_or_path = self.save_checkpoint(checkpoint_dir)
trainable_state = self.get_state()
checkpoint_path = TrainableUtil.process_checkpoint(
checkpoint_dict_or_path,
parent_dir=checkpoint_dir,
trainable_state=trainable_state,
)
# Maybe sync to cloud
self._maybe_save_to_cloud(checkpoint_dir)
return checkpoint_path
def testFindCheckpointDir(self):
checkpoint_path = os.path.join(self.checkpoint_dir, "0/my/nested/chkpt")
os.makedirs(checkpoint_path)
found_dir = TrainableUtil.find_checkpoint_dir(checkpoint_path)
self.assertEqual(self.checkpoint_dir, found_dir)
with self.assertRaises(FileNotFoundError):
parent = os.path.dirname(found_dir)
TrainableUtil.find_checkpoint_dir(parent)
def __call__(self, checkpoint: _TrackedCheckpoint):
"""Requests checkpoint deletion asynchronously.
Args:
checkpoint: Checkpoint to delete.
"""
if not self.runner:
return
if (checkpoint.storage_mode == CheckpointStorage.PERSISTENT
and checkpoint.dir_or_data):
checkpoint_path = checkpoint.dir_or_data
logger.debug("Trial %s: Deleting checkpoint %s", self.trial_id,
checkpoint_path)
# TODO(ujvl): Batch remote deletes.
# We first delete the remote checkpoint. If it is on the same
# node as the driver, it will also remove the local copy.
ray.get(self.runner.delete_checkpoint.remote(checkpoint_path))
# Delete local copy, if any exists.
if os.path.exists(checkpoint_path):
try:
checkpoint_dir = TrainableUtil.find_checkpoint_dir(
checkpoint_path)
shutil.rmtree(checkpoint_dir)
except FileNotFoundError:
logger.debug(
"Local checkpoint dir not found during deletion.")
def _create_checkpoint_dir(self,
checkpoint_dir: Optional[str] = None
) -> Optional[str]:
# Create checkpoint_xxxxx directory and drop checkpoint marker
checkpoint_dir = TrainableUtil.make_checkpoint_dir(
checkpoint_dir or self.logdir, index=self.iteration)
return checkpoint_dir
def _maybe_load_from_cloud(self, checkpoint_path: str) -> bool:
if not self.uses_cloud_checkpointing:
return False
rel_checkpoint_dir = TrainableUtil.find_rel_checkpoint_dir(
self.logdir, checkpoint_path)
external_uri = os.path.join(self.remote_checkpoint_dir,
rel_checkpoint_dir)
local_dir = os.path.join(self.logdir, rel_checkpoint_dir)
if self.custom_syncer:
# Only keep for backwards compatibility
self.custom_syncer.sync_down(remote_dir=external_uri,
local_dir=local_dir)
self.custom_syncer.wait_or_retry()
return True
checkpoint = Checkpoint.from_uri(external_uri)
retry_fn(
lambda: checkpoint.to_directory(local_dir),
subprocess.CalledProcessError,
num_retries=3,
sleep_time=1,
)
return True
def setUp(self):
self.checkpoint_dir = os.path.join(
ray._private.utils.get_user_temp_dir(), "tune", "MyTrainable123"
)
self.checkpoint_dir = TrainableUtil.make_checkpoint_dir(
self.checkpoint_dir, "0"
)
def save_checkpoint(self, tmp_checkpoint_dir: str = ""):
checkpoint_path = super().save_checkpoint()
parent_dir = TrainableUtil.find_checkpoint_dir(checkpoint_path)
preprocessor = self._merged_config.get("preprocessor", None)
if parent_dir and preprocessor:
save_preprocessor_to_dir(preprocessor, parent_dir)
return checkpoint_path
def restore_from_object(self, obj):
"""Restores training state from a checkpoint object.
These checkpoints are returned from calls to save_to_object().
"""
tmpdir = tempfile.mkdtemp("restore_from_object", dir=self.logdir)
checkpoint_path = TrainableUtil.create_from_pickle(obj, tmpdir)
self.restore(checkpoint_path)
shutil.rmtree(tmpdir)
def restore(self, trial: Trial) -> None:
"""Restores training state from a given model checkpoint.
Args:
trial: The trial to be restored.
Raises:
RuntimeError: This error is raised if no runner is found.
AbortTrialExecution: This error is raised if the trial is
ineligible for restoration, given the Tune input arguments.
"""
checkpoint = trial.checkpoint
if checkpoint.dir_or_data is None:
return
if trial.runner is None:
raise RuntimeError(
"Trial {}: Unable to restore - no runner found.".format(trial))
checkpoint_dir = checkpoint.dir_or_data
node_ip = checkpoint.node_ip
if checkpoint.storage_mode == CheckpointStorage.MEMORY:
logger.debug("Trial %s: Attempting restore from object", trial)
# Note that we don't store the remote since in-memory checkpoints
# don't guarantee fault tolerance and don't need to be waited on.
with self._change_working_directory(trial):
trial.runner.restore_from_object.remote(checkpoint_dir)
else:
logger.debug("Trial %s: Attempting restore from %s", trial,
checkpoint_dir)
if (trial.uses_cloud_checkpointing or not trial.sync_on_checkpoint
or not os.path.exists(checkpoint_dir)):
# If using cloud checkpointing, trial will get cp from cloud.
# If not syncing to driver, assume it has access to the cp
# on the local fs.
with self._change_working_directory(trial):
remote = trial.runner.restore.remote(
checkpoint_dir, node_ip)
elif trial.sync_on_checkpoint:
# This provides FT backwards compatibility in the
# case where no cloud checkpoints are provided.
logger.debug("Trial %s: Reading checkpoint into memory", trial)
checkpoint_path = TrainableUtil.find_checkpoint_dir(
checkpoint_dir)
obj = Checkpoint.from_directory(checkpoint_path).to_bytes()
with self._change_working_directory(trial):
remote = trial.runner.restore_from_object.remote(obj)
else:
raise _AbortTrialExecution(
"Pass in `sync_on_checkpoint=True` for driver-based trial"
"restoration. Pass in an `upload_dir` for remote "
"storage-based restoration")
self._futures[remote] = (_ExecutorEventType.RESTORING_RESULT,
trial)
trial.restoring_from = checkpoint
def testConvertTempToPermanent(self):
checkpoint_dir = FuncCheckpointUtil.mk_temp_checkpoint_dir(self.logdir)
new_checkpoint_dir = FuncCheckpointUtil.create_perm_checkpoint(
checkpoint_dir, self.logdir, step=4)
assert new_checkpoint_dir == TrainableUtil.find_checkpoint_dir(
new_checkpoint_dir)
assert os.path.exists(new_checkpoint_dir)
assert not FuncCheckpointUtil.is_temp_checkpoint_dir(
new_checkpoint_dir)
tmp_checkpoint_dir = FuncCheckpointUtil.mk_temp_checkpoint_dir(
self.logdir)
assert tmp_checkpoint_dir != new_checkpoint_dir
def save_to_object(self):
"""Saves the current model state to a Python object.
It also saves to disk but does not return the checkpoint path.
Returns:
Object holding checkpoint data.
"""
tmpdir = tempfile.mkdtemp("save_to_object", dir=self.logdir)
checkpoint_path = self.save(tmpdir)
# Save all files in subtree and delete the tmpdir.
obj = TrainableUtil.checkpoint_to_object(checkpoint_path)
shutil.rmtree(tmpdir)
return obj
def write_checkpoint(trial: Trial, index: int):
checkpoint_dir = TrainableUtil.make_checkpoint_dir(trial.logdir,
index=index)
result = {"training_iteration": index}
with open(os.path.join(checkpoint_dir, "cp.json"), "w") as f:
json.dump(result, f)
tune_cp = _TrackedCheckpoint(
dir_or_data=checkpoint_dir,
storage_mode=CheckpointStorage.PERSISTENT,
metrics=result,
)
trial.saving_to = tune_cp
return checkpoint_dir
def create_checkpoint(preprocessor: Optional[Preprocessor] = None,
config: Optional[dict] = None) -> Checkpoint:
rl_trainer = RLTrainer(
algorithm=_DummyAlgo,
config=config or {},
preprocessor=preprocessor,
)
rl_trainable_cls = rl_trainer.as_trainable()
rl_trainable = rl_trainable_cls()
with tempfile.TemporaryDirectory() as checkpoint_dir:
checkpoint_file = rl_trainable.save(checkpoint_dir)
checkpoint_path = TrainableUtil.find_checkpoint_dir(checkpoint_file)
checkpoint_data = Checkpoint.from_directory(checkpoint_path).to_dict()
return Checkpoint.from_dict(checkpoint_data)
def testPickleCheckpoint(self):
for i in range(5):
path = os.path.join(self.checkpoint_dir, str(i))
with open(path, "w") as f:
f.write(str(i))
checkpoint_path = os.path.join(self.checkpoint_dir, "0")
data_dict = TrainableUtil.pickle_checkpoint(checkpoint_path)
loaded = cloudpickle.loads(data_dict)
checkpoint_name = os.path.basename(checkpoint_path)
self.assertEqual(loaded["checkpoint_name"], checkpoint_name)
for i in range(5):
path = os.path.join(self.checkpoint_dir, str(i))
self.assertEqual(loaded["data"][str(i)], open(path, "rb").read())
def delete_checkpoint(self, checkpoint_path: Union[str, Checkpoint]):
"""Deletes local copy of checkpoint.
Args:
checkpoint_path: Path to checkpoint.
"""
# Ensure Checkpoints are converted
if isinstance(checkpoint_path,
Checkpoint) and checkpoint_path._local_path:
checkpoint_path = checkpoint_path._local_path
try:
checkpoint_dir = TrainableUtil.find_checkpoint_dir(checkpoint_path)
except FileNotFoundError:
# The checkpoint won't exist locally if the
# trial was rescheduled to another worker.
logger.debug(
f"Local checkpoint not found during garbage collection: "
f"{self.trial_id} - {checkpoint_path}")
return
else:
if self.uses_cloud_checkpointing:
if self.custom_syncer:
# Keep for backwards compatibility
self.custom_syncer.delete(
self._storage_path(checkpoint_dir))
self.custom_syncer.wait_or_retry()
else:
checkpoint_uri = self._storage_path(checkpoint_dir)
retry_fn(
lambda: delete_external_checkpoint(checkpoint_uri),
subprocess.CalledProcessError,
num_retries=3,
sleep_time=1,
)
if os.path.exists(checkpoint_dir):
shutil.rmtree(checkpoint_dir)
def to_air_checkpoint(self) -> Optional[Checkpoint]:
from ray.tune.trainable.util import TrainableUtil
checkpoint_data = self.dir_or_data
if not checkpoint_data:
return None
if isinstance(checkpoint_data, ray.ObjectRef):
checkpoint_data = ray.get(checkpoint_data)
if isinstance(checkpoint_data, str):
try:
checkpoint_dir = TrainableUtil.find_checkpoint_dir(
checkpoint_data)
except FileNotFoundError:
if log_once("checkpoint_not_available"):
logger.error(
f"The requested checkpoint is not available on this node, "
f"most likely because you are using Ray client or disabled "
f"checkpoint synchronization. To avoid this, enable checkpoint "
f"synchronization to cloud storage by specifying a "
f"`SyncConfig`. The checkpoint may be available on a different "
f"node - please check this location on worker nodes: "
f"{checkpoint_data}")
return None
checkpoint = Checkpoint.from_directory(checkpoint_dir)
elif isinstance(checkpoint_data, bytes):
checkpoint = Checkpoint.from_bytes(checkpoint_data)
elif isinstance(checkpoint_data, dict):
checkpoint = Checkpoint.from_dict(checkpoint_data)
else:
raise RuntimeError(
f"Unknown checkpoint data type: {type(checkpoint_data)}")
return checkpoint
def get_trial_checkpoints_paths(
self,
trial: Trial,
metric: Optional[str] = None) -> List[Tuple[str, Number]]:
"""Gets paths and metrics of all persistent checkpoints of a trial.
Args:
trial: The log directory of a trial, or a trial instance.
metric: key for trial info to return, e.g. "mean_accuracy".
"training_iteration" is used by default if no value was
passed to ``self.default_metric``.
Returns:
List of [path, metric] for all persistent checkpoints of the trial.
"""
metric = metric or self.default_metric or TRAINING_ITERATION
if isinstance(trial, str):
trial_dir = os.path.expanduser(trial)
# Get checkpoints from logdir.
chkpt_df = TrainableUtil.get_checkpoints_paths(trial_dir)
# Join with trial dataframe to get metrics.
trial_df = self.trial_dataframes[trial_dir]
path_metric_df = chkpt_df.merge(trial_df,
on="training_iteration",
how="inner")
return path_metric_df[["chkpt_path", metric]].values.tolist()
elif isinstance(trial, Trial):
checkpoints = trial.get_trial_checkpoints()
# Support metrics given as paths, e.g.
# "info/learner/default_policy/policy_loss".
return [(c.dir_or_data, unflattened_lookup(metric, c.metrics))
for c in checkpoints]
else:
raise ValueError("trial should be a string or a Trial instance.")
def save(self, checkpoint_dir: Optional[str] = None) -> str:
"""Saves the current model state to a checkpoint.
Subclasses should override ``save_checkpoint()`` instead to save state.
This method dumps additional metadata alongside the saved path.
If a remote checkpoint dir is given, this will also sync up to remote
storage.
Args:
checkpoint_dir: Optional dir to place the checkpoint.
Returns:
The given or created checkpoint directory.
Note the return path should match up with what is expected of
`restore()`.
"""
checkpoint_dir = self._create_checkpoint_dir(
checkpoint_dir=checkpoint_dir)
# User saves checkpoint
checkpoint_dict_or_path = self.save_checkpoint(checkpoint_dir)
if checkpoint_dict_or_path is None:
# checkpoint_dict_or_path can only be None in class trainables.
# In that case the default is to use the root checkpoint directory.
assert checkpoint_dir
checkpoint_dict_or_path = checkpoint_dir
elif checkpoint_dir is None:
# checkpoint_dir is only None in function trainables. In that case,
# checkpoint_dict_or_path points to the already saved checkpoint dir.
# This will be considered the root dir.
assert isinstance(checkpoint_dict_or_path, str)
checkpoint_dir = checkpoint_dict_or_path
# Get trainable metadata
metadata = self.get_state()
if isinstance(checkpoint_dict_or_path, dict):
metadata["relative_checkpoint_path"] = ""
metadata["saved_as_dict"] = True
Checkpoint.from_dict(checkpoint_dict_or_path).to_directory(
checkpoint_dir)
# Re-drop marker
TrainableUtil.mark_as_checkpoint_dir(checkpoint_dir)
else:
# Make sure the checkpoint dir is contained
if not checkpoint_dict_or_path.startswith(checkpoint_dir):
raise ValueError(
f"The returned checkpoint path must be within the given "
f"checkpoint dir ({checkpoint_dir}): {checkpoint_dict_or_path}"
)
# Get relative path to returned checkpoint
relative_checkpoint_path = os.path.relpath(checkpoint_dict_or_path,
checkpoint_dir)
metadata["relative_checkpoint_path"] = relative_checkpoint_path
metadata["saved_as_dict"] = False
TrainableUtil.write_metadata(checkpoint_dir, metadata)
# Maybe sync to cloud
self._maybe_save_to_cloud(checkpoint_dir)
return checkpoint_dir
def restore(
self,
checkpoint_path: Union[str, Checkpoint],
checkpoint_node_ip: Optional[str] = None,
):
"""Restores training state from a given model checkpoint.
These checkpoints are returned from calls to save().
Subclasses should override ``load_checkpoint()`` instead to
restore state.
This method restores additional metadata saved with the checkpoint.
`checkpoint_path` should match with the return from ``save()``.
`checkpoint_path` can be
`~/ray_results/exp/MyTrainable_abc/
checkpoint_00000/checkpoint`. Or,
`~/ray_results/exp/MyTrainable_abc/checkpoint_00000`.
`self.logdir` should generally be corresponding to `checkpoint_path`,
for example, `~/ray_results/exp/MyTrainable_abc`.
`self.remote_checkpoint_dir` in this case, is something like,
`REMOTE_CHECKPOINT_BUCKET/exp/MyTrainable_abc`
Args:
checkpoint_path: Path to restore checkpoint from. If this
path does not exist on the local node, it will be fetched
from external (cloud) storage if available, or restored
from a remote node.
checkpoint_node_ip: If given, try to restore
checkpoint from this node if it doesn't exist locally or
on cloud storage.
"""
# Ensure Checkpoints are converted
if isinstance(checkpoint_path, Checkpoint):
return self._restore_from_checkpoint_obj(checkpoint_path)
if not self._maybe_load_from_cloud(checkpoint_path) and (
# If a checkpoint source IP is given
checkpoint_node_ip
# And the checkpoint does not currently exist on the local node
and not os.path.exists(checkpoint_node_ip)
# And the source IP is different to the current IP
and checkpoint_node_ip != ray.util.get_node_ip_address()):
checkpoint = get_checkpoint_from_remote_node(
checkpoint_path, checkpoint_node_ip)
if checkpoint:
checkpoint.to_directory(checkpoint_path)
if not os.path.exists(checkpoint_path):
raise ValueError(
f"Could not recover from checkpoint as it does not exist on local "
f"disk and was not available on cloud storage or another Ray node. "
f"Got checkpoint path: {checkpoint_path} and IP {checkpoint_node_ip}"
)
checkpoint_dir = TrainableUtil.find_checkpoint_dir(checkpoint_path)
metadata = TrainableUtil.load_metadata(checkpoint_dir)
if metadata["saved_as_dict"]:
# If data was saved as a dict (e.g. from a class trainable),
# also pass the dict to `load_checkpoint()`.
checkpoint_dict = Checkpoint.from_directory(
checkpoint_dir).to_dict()
# If other files were added to the directory after converting from the
# original dict (e.g. marker files), clean these up
checkpoint_dict.pop(_DICT_CHECKPOINT_ADDITIONAL_FILE_KEY, None)
to_load = checkpoint_dict
else:
# Otherwise, pass the relative checkpoint path
relative_checkpoint_path = metadata["relative_checkpoint_path"]
to_load = os.path.join(checkpoint_dir, relative_checkpoint_path)
# Set metadata
self._experiment_id = metadata["experiment_id"]
self._iteration = metadata["iteration"]
self._timesteps_total = metadata["timesteps_total"]
self._time_total = metadata["time_total"]
self._episodes_total = metadata["episodes_total"]
# Actually load checkpoint
self.load_checkpoint(to_load)
self._time_since_restore = 0.0
self._timesteps_since_restore = 0
self._iterations_since_restore = 0
self._restored = True
logger.info("Restored on %s from checkpoint: %s",
self.get_current_ip(), checkpoint_dir)
state = {
"_iteration": self._iteration,
"_timesteps_total": self._timesteps_total,
"_time_total": self._time_total,
"_episodes_total": self._episodes_total,
}
logger.info("Current state after restoring: %s", state)
def test_find_rel_checkpoint_dir(checkpoint_path, logdir):
assert (
TrainableUtil.find_rel_checkpoint_dir(logdir, checkpoint_path) == "checkpoint0"
)
