Details
Description
StorageContainerManager's block protocol creates a new container for each allocate block call instead of using existing open containers. This behavior is not seen once the cluster is restarted.
When createContainer flag is set, the container state is changed from ALLOCATED to CREATING. But in refreshContainers call only ALLOCATED state is handled.
During restart loadAllocatedContainers loads the containers map properly, which fixes the issue after restart. But we will face the same issue later when we allocate new containers.
Attachments
Attachments
Issue Links
- is duplicated by
-
HDFS-12367 Ozone: Too many open files error while running corona
- Resolved