def _handle_failure(self, error):
logger.exception("Error in monitor loop")
if self.autoscaler is not None and \
os.environ.get("RAY_AUTOSCALER_FATESHARE_WORKERS", "") == "1":
self.autoscaler.kill_workers()
# Take down autoscaler workers if necessary.
self.destroy_autoscaler_workers()
# Something went wrong, so push an error to all current and future
# drivers.
message = f"The autoscaler failed with the following error:\n{error}"
if _internal_kv_initialized():
_internal_kv_put(DEBUG_AUTOSCALING_ERROR, message, overwrite=True)
redis_client = ray._private.services.create_redis_client(
self.redis_address, password=self.redis_password)
gcs_publisher = None
if args.gcs_address:
gcs_publisher = GcsPublisher(address=args.gcs_address)
elif gcs_pubsub_enabled():
gcs_publisher = GcsPublisher(
address=get_gcs_address_from_redis(redis_client))
from ray._private.utils import publish_error_to_driver
publish_error_to_driver(
ray_constants.MONITOR_DIED_ERROR,
message,
redis_client=redis_client,
gcs_publisher=gcs_publisher)
def _handle_failure(self, error):
logger.exception("Error in monitor loop")
if (
self.autoscaler is not None
and os.environ.get("RAY_AUTOSCALER_FATESHARE_WORKERS", "") == "1"
):
self.autoscaler.kill_workers()
# Take down autoscaler workers if necessary.
self.destroy_autoscaler_workers()
# Something went wrong, so push an error to all current and future
# drivers.
message = f"The autoscaler failed with the following error:\n{error}"
if _internal_kv_initialized():
_internal_kv_put(
ray_constants.DEBUG_AUTOSCALING_ERROR, message, overwrite=True
)
gcs_publisher = GcsPublisher(address=args.gcs_address)
from ray._private.utils import publish_error_to_driver
publish_error_to_driver(
ray_constants.MONITOR_DIED_ERROR,
message,
gcs_publisher=gcs_publisher,
)