damccorm commented on code in PR #28263:
URL: https://github.com/apache/beam/pull/28263#discussion_r1311977323
##########
sdks/python/apache_beam/ml/inference/base.py:
##########
@@ -587,6 +588,14 @@ def run_inference(
keys,
self._unkeyed.run_inference(unkeyed_batch, model, inference_args))
+ # The first time a MultiProcessShared ModelManager is used for inference
+ # from this process, we should increment its max model count
+ if self._max_models_per_worker_hint is not None:
+ lock = threading.Lock()
+ if lock.acquire(blocking=False):
Review Comment:
The goal is to run this exactly once per process. So the idea is that in a
given process, a single thread will acquire the lock and increment the max
models. Then the remaining threads (or the original thread in the future) will
try to acquire the lock and fail, so they will not increment the value.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
