training hang when remove/add instances about elastic

[kiukchung]

Thanks for the bug report! I've seen this issue happen due to the dataloader processes that get spawned in the background - the exact root cause is still TBD, but the symptom is exactly like you describe (trainers get stuck). Do make sure that is the root cause can you try running the classy vision example with --num_workers 0 parameter? This prevents the dataloader processes from being spawned (e.g. the dataloader works in-proces).

[kiukchung]

closing as I've not heard back from the OP, feel free to reopen if this continues being an issue.

[VincentLeeMax]

closing as I've not heard back from the OP, feel free to reopen if this continues being an issue.

Hi, @kiukchung，I encountered this problem, setting --num_workers 0 works for me, but the training become very slow, any idea of this problem?