mirror of
https://github.com/wassname/ray.git
synced 2026-07-06 05:16:30 +08:00
[tune] Change log handling for Tune (#3661)
Also provides a small retry mechanism for a transient error as reported by #3340. Closes #3653.
This commit is contained in:
@@ -234,10 +234,16 @@ class RayTrialExecutor(TrialExecutor):
|
||||
assert self._committed_resources.cpu >= 0
|
||||
assert self._committed_resources.gpu >= 0
|
||||
|
||||
def _update_avail_resources(self):
|
||||
resources = ray.global_state.cluster_resources()
|
||||
def _update_avail_resources(self, num_retries=5):
|
||||
for i in range(num_retries):
|
||||
resources = ray.global_state.cluster_resources()
|
||||
if not resources:
|
||||
logger.warning("Cluster resources not detected. Retrying...")
|
||||
time.sleep(0.5)
|
||||
|
||||
num_cpus = resources["CPU"]
|
||||
num_gpus = resources["GPU"]
|
||||
|
||||
self._avail_resources = Resources(int(num_cpus), int(num_gpus))
|
||||
self._resources_initialized = True
|
||||
|
||||
|
||||
Reference in New Issue
Block a user