AFM to cloud object storage limitations

• Hard links are supported only in the Local-Update (LU) mode. The creation of hard links fails with permission denied or E_PERM error on other modes.
• The primary resources of object servers are buckets and objects.
• AFM can synchronize empty directories to the cloud object server only if an AFM to cloud object fileset is created by using the mmafmcosconfig --directory-object command.
• The ChangeTimes operation is supported on the cache filesets. However, this operation does not replicate to the target cloud object server.
• To replicate the chmod and chown metadata operations to a cloud object storage, you need to use the mmafmcosconfig command with the --xattr option.
• Renaming a non-empty directory is only supported for filesets in LU mode. When the fileset is not in LU mode and you attempt to rename a non-empty directory, the operation fails with the E_NOTEMPTY error. Empty directories and local directories can be renamed.
• Parallel data transfer for write operations is not supported on an AFM to cloud object storage enabled fileset. Objects are not synchronized by splitting chunks across gateway nodes but objects are queued on different gateway nodes based on the mapping.
• Deletion of objects on the cloud object storage synchronizes the deletion of files on the cache. But empty directories might remain on the cache.
• Support of file and directory naming convention is in accordance with the cloud object storage server supported guidelines. Some cloud object storage servers do not support special characters. The file or directory name must not contain special characters.
• File paths that have more characters than maximum characters limitation are not supported.
• You can run AFM to cloud object storage commands such as mmafmcoskeys, mmafmcosconfig, mmafmcosctl, and mmafmcosaccess only from a Linux node.
• Immutability and appendOnly features are not supported on AFM to cloud object storage filesets.
• The --iam-mode option is not supported on AFM to cloud object storage filesets.
• Reading AFM uncached files from the snapshot results in the error 5 (EIO). To avoid the error, set the afmSnapUncachedRead option on the fileset. This configuration returns zeros for the evicted blocks instead of error 5.
• A dependent fileset linking is not supported in an AFM to cloud object storage fileset.
• In some cases, an AFM cache has pending changes such as delete and rename to replicate, and the fileset recovery is triggered. In this case, the cloud object storage might have old data and renamed files or directories are replicated. Because of the files or directories replication, some extra files might exist on the cloud object storage. The AFM cache and the cloud object storage have the latest copy of the files.
• AFM to cloud object storage does not maintain sparseness of files when uploading the files to cloud object storage.
• Based on the user defined policy, reconcile command does not upload a directory with objects that do not meet the policy criteria to cloud objects storage.
• Renaming of an uncached file is not supported on an AFM to cloud object storage MU-mode fileset. To rename a file, you need to download this file data first and then rename it.
• While creating samba exports of an AFM to cloud object storage enabled fileset, Samba Oplocks needs to be disabled.
• AFM to cloud object storage does not delete empty directories on the cloud bucket.
• Adding a new node to the cluster might not synchronize the AFM to cloud object storage keys on the node. To synchronize the AFM to cloud object storage keys on the newly added node, you can run the following command on the new node:

  /usr/lpp/mmfs/bin/mmccrnv _afmcos refresh

• AFM does not support storing an object list file in the AFM fileset that is used for uploading or downloading. To run upload or download by using the object list file, you can create an object list file outside the AFM fileset linked-path.

Upload or write and download or read operations are replicated to a target object server. The operations such as delete and rename are not synchronized between a cache and an object server.

• If data on a cache is deleted or renamed, the data is not synchronized to an object server.
• If an object on an object storage server is deleted or renamed, and the object is already replicated to a cache, the object is not synchronized to the cache.