mirror of
https://github.com/wassname/ray.git
synced 2026-07-08 09:02:59 +08:00
Change all instances of JobID to DriverID. (#4431)
This commit is contained in:
justinwyang
committed by
Robert Nishihara
Robert Nishihara
parent
06c768823c
commit
8dfc833a8b
@@ -840,19 +840,19 @@ class GlobalState(object):
|
||||
|
||||
return dict(total_available_resources)
|
||||
|
||||
def _error_messages(self, job_id):
|
||||
"""Get the error messages for a specific job.
|
||||
def _error_messages(self, driver_id):
|
||||
"""Get the error messages for a specific driver.
|
||||
|
||||
Args:
|
||||
job_id: The ID of the job to get the errors for.
|
||||
driver_id: The ID of the driver to get the errors for.
|
||||
|
||||
Returns:
|
||||
A list of the error messages for this job.
|
||||
A list of the error messages for this driver.
|
||||
"""
|
||||
assert isinstance(job_id, ray.DriverID)
|
||||
assert isinstance(driver_id, ray.DriverID)
|
||||
message = self.redis_client.execute_command(
|
||||
"RAY.TABLE_LOOKUP", ray.gcs_utils.TablePrefix.ERROR_INFO, "",
|
||||
job_id.binary())
|
||||
driver_id.binary())
|
||||
|
||||
# If there are no errors, return early.
|
||||
if message is None:
|
||||
@@ -864,7 +864,7 @@ class GlobalState(object):
|
||||
for i in range(gcs_entries.EntriesLength()):
|
||||
error_data = ray.gcs_utils.ErrorTableData.GetRootAsErrorTableData(
|
||||
gcs_entries.Entries(i), 0)
|
||||
assert job_id.binary() == error_data.JobId()
|
||||
assert driver_id.binary() == error_data.DriverId()
|
||||
error_message = {
|
||||
"type": decode(error_data.Type()),
|
||||
"message": decode(error_data.ErrorMessage()),
|
||||
@@ -873,31 +873,32 @@ class GlobalState(object):
|
||||
error_messages.append(error_message)
|
||||
return error_messages
|
||||
|
||||
def error_messages(self, job_id=None):
|
||||
"""Get the error messages for all jobs or a specific job.
|
||||
def error_messages(self, driver_id=None):
|
||||
"""Get the error messages for all drivers or a specific driver.
|
||||
|
||||
Args:
|
||||
job_id: The specific job to get the errors for. If this is None,
|
||||
then this method retrieves the errors for all jobs.
|
||||
driver_id: The specific driver to get the errors for. If this is
|
||||
None, then this method retrieves the errors for all drivers.
|
||||
|
||||
Returns:
|
||||
A dictionary mapping job ID to a list of the error messages for
|
||||
that job.
|
||||
A dictionary mapping driver ID to a list of the error messages for
|
||||
that driver.
|
||||
"""
|
||||
if job_id is not None:
|
||||
assert isinstance(job_id, ray.DriverID)
|
||||
return self._error_messages(job_id)
|
||||
if driver_id is not None:
|
||||
assert isinstance(driver_id, ray.DriverID)
|
||||
return self._error_messages(driver_id)
|
||||
|
||||
error_table_keys = self.redis_client.keys(
|
||||
ray.gcs_utils.TablePrefix_ERROR_INFO_string + "*")
|
||||
job_ids = [
|
||||
driver_ids = [
|
||||
key[len(ray.gcs_utils.TablePrefix_ERROR_INFO_string):]
|
||||
for key in error_table_keys
|
||||
]
|
||||
|
||||
return {
|
||||
binary_to_hex(job_id): self._error_messages(ray.DriverID(job_id))
|
||||
for job_id in job_ids
|
||||
binary_to_hex(driver_id): self._error_messages(
|
||||
ray.DriverID(driver_id))
|
||||
for driver_id in driver_ids
|
||||
}
|
||||
|
||||
def actor_checkpoint_info(self, actor_id):
|
||||
|
||||
@@ -76,7 +76,7 @@ def construct_error_message(driver_id, error_type, message, timestamp):
|
||||
message_offset = builder.CreateString(message)
|
||||
|
||||
ray.core.generated.ErrorTableData.ErrorTableDataStart(builder)
|
||||
ray.core.generated.ErrorTableData.ErrorTableDataAddJobId(
|
||||
ray.core.generated.ErrorTableData.ErrorTableDataAddDriverId(
|
||||
builder, driver_offset)
|
||||
ray.core.generated.ErrorTableData.ErrorTableDataAddType(
|
||||
builder, error_type_offset)
|
||||
|
||||
@@ -32,67 +32,51 @@ cdef extern from "ray/id.h" namespace "ray" nogil:
|
||||
@staticmethod
|
||||
CActorCheckpointID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CActorClassID "ray::ActorClassID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CActorClassID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CActorID "ray::ActorID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CActorID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CActorHandleID "ray::ActorHandleID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CActorHandleID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CClientID "ray::ClientID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CClientID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CConfigID "ray::ConfigID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CConfigID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CFunctionID "ray::FunctionID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CFunctionID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CDriverID "ray::DriverID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CDriverID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CJobID "ray::JobID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CJobID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CTaskID "ray::TaskID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CTaskID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CObjectID" ray::ObjectID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
CObjectID from_binary(const c_string &binary)
|
||||
|
||||
|
||||
cdef cppclass CWorkerID "ray::WorkerID"(CUniqueID):
|
||||
|
||||
@staticmethod
|
||||
|
||||
@@ -19,7 +19,6 @@ from ray.includes.unique_ids cimport (
|
||||
CConfigID,
|
||||
CDriverID,
|
||||
CFunctionID,
|
||||
CJobID,
|
||||
CObjectID,
|
||||
CTaskID,
|
||||
CUniqueID,
|
||||
|
||||
@@ -1142,8 +1142,8 @@ def error_info():
|
||||
"""Return information about failed tasks."""
|
||||
worker = global_worker
|
||||
worker.check_connected()
|
||||
return (global_state.error_messages(job_id=worker.task_driver_id) +
|
||||
global_state.error_messages(job_id=DriverID.nil()))
|
||||
return (global_state.error_messages(driver_id=worker.task_driver_id) +
|
||||
global_state.error_messages(driver_id=DriverID.nil()))
|
||||
|
||||
|
||||
def _initialize_serialization(driver_id, worker=global_worker):
|
||||
@@ -1288,8 +1288,9 @@ def init(redis_address=None,
|
||||
node_ip_address (str): The IP address of the node that we are on.
|
||||
object_id_seed (int): Used to seed the deterministic generation of
|
||||
object IDs. The same value can be used across multiple runs of the
|
||||
same job in order to generate the object IDs in a consistent
|
||||
manner. However, the same ID should not be used for different jobs.
|
||||
same driver in order to generate the object IDs in a consistent
|
||||
manner. However, the same ID should not be used for different
|
||||
drivers.
|
||||
local_mode (bool): True if the code should be executed serially
|
||||
without Ray. This is useful for debugging.
|
||||
ignore_reinit_error: True if we should suppress errors from calling
|
||||
@@ -1663,8 +1664,8 @@ def listen_error_messages_raylet(worker, task_error_queue, threads_stopped):
|
||||
assert gcs_entry.EntriesLength() == 1
|
||||
error_data = ray.gcs_utils.ErrorTableData.GetRootAsErrorTableData(
|
||||
gcs_entry.Entries(0), 0)
|
||||
job_id = error_data.JobId()
|
||||
if job_id not in [
|
||||
driver_id = error_data.DriverId()
|
||||
if driver_id not in [
|
||||
worker.task_driver_id.binary(),
|
||||
DriverID.nil().binary()
|
||||
]:
|
||||
|
||||
Reference in New Issue
Block a user