Details
-
Improvement
-
Resolution: Unresolved
-
Major
-
5.5.0
Description
Inspired by MB-27914, currently on a dataset rebalance, we wait for all readers before we are able to hold a write lock on a dataset to do the metadata switch to the dataset's new node group. This wait could be a long period if there are many queued jobs to be processed. We should consider alternatives to this behavior such as recompiling queued jobs.