Addresses missed comments from multichunk object transfer PR. #1908
Conversation
object manager config bug fix. addresses other comments from ray-project#1827.
|
Test PASSed. |
src/common/state/ray_config.h
Outdated
| @@ -196,6 +219,21 @@ class RayConfig { | |||
| /// corresponding driver. Since spillback currently occurs on a 100ms timer, | |||
| /// a value of 100 corresponds to a warning every 10 seconds. | |||
| int64_t actor_creation_num_spillbacks_warning_; | |||
|
|
|||
| /// Time out, in milliseconds, to wait before retrying a failed pull in the | |||
There was a problem hiding this comment.
timeout should be one word
src/common/state/ray_config.h
Outdated
| actor_creation_num_spillbacks_warning_(100) {} | ||
| actor_creation_num_spillbacks_warning_(100), | ||
| // TODO: Setting this to large values results in latency, which needs to | ||
| // be addressed. This timeout is often on the critical path to object |
There was a problem hiding this comment.
critical path "for"
| store_notification_(main_service, config.store_socket_name), | ||
| // release_delay of 2 * config.max_sends is to ensure the pool does not release | ||
| store_notification_(main_service, config_.store_socket_name), | ||
| // release_delay of 2 * config_.max_sends is to ensure the pool does not release |
There was a problem hiding this comment.
why not just use plasma_default_release_delay_? Coupling it with max_sends seems pretty strange to me
There was a problem hiding this comment.
Here's the scenario we had in mind when we implemented this. If we have max_sends objects split into 2 chunks, and the first chunk of each is gotten and released before each second chunk is gotten, then any release delay <max_sends (or <=max_sends if the bound is inclusive) will result in the first object actually being released. When we get the second chunk of the first object that was gotten, the associated get call to the object store client will incur the usual cost of a get. Perhaps a release delay of max_sends+1 is sufficient.
|
Test PASSed. |
|
Test FAILed. |
|
Test FAILed. |
|
retest this please |
|
Test PASSed. |
* master: Handle interrupts correctly for ASIO synchronous reads and writes. (ray-project#1929) [DataFrame] Adding read methods and tests (ray-project#1712) Allow task_table_update to fail when tasks are finished. (ray-project#1927) [rllib] Contribute DDPG to RLlib (ray-project#1877) [xray] Workers blocked in a `ray.get` release their resources (ray-project#1920) Raylet task dispatch and throttling worker startup (ray-project#1912) [DataFrame] Eval fix (ray-project#1903) [tune] Polishing docs (ray-project#1846) [tune] [rllib] Automatically determine RLlib resources and add queueing mechanism for autoscaling (ray-project#1848) Preemptively push local arguments for actor tasks (ray-project#1901) [tune] Allow fetching pinned objects from trainable functions (ray-project#1895) Multithreading refactor for ObjectManager. (ray-project#1911) Add slice functionality (ray-project#1832) [DataFrame] Pass read_csv kwargs to _infer_column (ray-project#1894) Addresses missed comments from multichunk object transfer PR. (ray-project#1908) Allow numpy arrays to be passed by value into tasks (and inlined in the task spec). (ray-project#1816) [xray] Lineage cache requests notifications from the GCS about remote tasks (ray-project#1834) Fix UI issue for non-json-serializable task arguments. (ray-project#1892) Remove unnecessary calls to .hex() for object IDs. (ray-project#1910) Allow multiple raylets to be started on a single machine. (ray-project#1904) # Conflicts: # python/ray/rllib/__init__.py # python/ray/rllib/dqn/dqn.py
Addresses missed comments from #1827.
This also fixes a bug when passing in the object manager configuration.