Enable fractional resources and resource IDs for xray. (#2187)

* Implement GPU IDs and fractional resources.

* Add documentation and python exceptions.

* Fix signed/unsigned comparison.

* Fix linting.

* Fixes from rebase.

* Re-enable tests that use ray.wait.

* Don't kill the raylet if an infeasible task is submitted.

* Ignore tests that require better load balancing.

* Linting

* Ignore array test.

* Ignore stress test reconstructions tests.

* Don't kill node manager if remote node manager disconnects.

* Ignore more stress tests.

* Naming changes

* Remove outdated todo

* Small fix

* Re-enable test.

* Linting

* Fix resource bookkeeping for blocked tasks.

* Fix linting

* Fix Java client.

* Ignore test

* Ignore put error tests

python/ray/ray_constants.py

@@ -12,6 +12,15 @@ def env_integer(key, default):
     return default
 
 
+# If a remote function or actor (or some other export) has serialized size
+# greater than this quantity, print an warning.
+PICKLE_OBJECT_WARNING_SIZE = 10**7
+
+# The maximum resource quantity that is allowed. TODO(rkn): This could be
+# relaxed, but the current implementation of the node manager will be slower
+# for large resource quantities due to bookkeeping of specific resource IDs.
+MAX_RESOURCE_QUANTITY = 512
+
 # Different types of Ray errors that can be pushed to the driver.
 # TODO(rkn): These should be defined in flatbuffers and must be synced with
 # the existing C++ definitions.
@@ -29,10 +38,6 @@ WORKER_DIED_PUSH_ERROR = "worker_died"
 PUT_RECONSTRUCTION_PUSH_ERROR = "put_reconstruction"
 HASH_MISMATCH_PUSH_ERROR = "object_hash_mismatch"
 
-# If a remote function or actor (or some other export) has serialized size
-# greater than this quantity, print an warning.
-PICKLE_OBJECT_WARNING_SIZE = 10**7
-
 # Abort autoscaling if more than this number of errors are encountered. This
 # is a safety feature to prevent e.g. runaway node launches.
 AUTOSCALER_MAX_NUM_FAILURES = env_integer("AUTOSCALER_MAX_NUM_FAILURES", 5)