The failure is caused by the strong-global metadata consistency requirement, not by the Strict ILM placement requirement. StorageGRID stores object metadata in Cassandra, normally maintaining three metadata replicas per site. With three Storage Nodes at Site-2, taking one node down for maintenance and then losing another leaves only one metadata replica available at that site. The required metadata quorum can no longer be achieved for strong-global operations. NetApp documents that strong-global guarantees read-after-write consistency across all sites and requires sufficient metadata replicas to complete the operation.

The distinction from Strict ingest is important. Strict controls object-data placement: all copies specified by the ILM rule must be created before StorageGRID returns success. Here, the ILM rule requires only one copy at each site. Because one Storage Node remains operational at Site-2, StorageGRID can potentially place the Site-2 object copy there. Therefore, the data-placement requirement itself is not necessarily violated.

However, with two of the three metadata-bearing Storage Nodes unavailable at Site-2, Cassandra cannot satisfy the strong-global metadata consistency requirement, causing client PUT operations to fail.

Reference topics: S3 Consistency → Strong-global → Cassandra Metadata Replicas → Metadata Quorum → Strict ILM Ingest Behavior