Requirements, limitations, and support for clustered watch folder

Use this information to understand the requirements, limitations, and support for installing clustered watch folder.

RPM and package requirements
In order for watch folder events to be generated from a node, the node must have the following packages installed:
  • For RHEL and SLES, the librdkafka package requires the openssl-devel and cyrus-sasl-devel packages.
  • For Ubuntu, the librdkafka package requires the libssl-dev and libsasl2-dev packages.
  • librdkafka (gpfs.librdkafka rpm/package).
OS and hardware requirements
Note: Not all of these requirements apply to the support of remotely mounted file systems.

Start of changeWatch folder is supported on all Linux® OS versions supported by IBM Storage Scale. For more information, refer to IBM Storage Scale FAQ in IBM® Documentation.End of change

Security requirements and limitations
  • Root authority is required to run mmwatch.
Restrictions imposed by mixed environments and protocols
  • Events will not be generated on non-Linux® nodes.
  • IBM Storage Scale clustered watch folder has full support for the following protocols (support for all other protocols must be considered limited):
    • NFS-Ganesha
    • SMB
    • Native UNIX file access
Clustered watch folder attributes
For more information about the availability and limitations of the clustered watch folder attributes, see JSON attributes in clustered watch folder.
GPFS file system requirements and limitations
  • Clustered watch folder can be enabled only for file systems that are created or upgraded to IBM Storage Scale 5.0.3 or later.
  • When clustered watch folder is enabled on a file system, changing the file system name is not allowed. Clustered watch folder must first be disabled to change the file system name.
Miscellaneous requirements, limitations, recommendations, and support statements
  • Clustered watch folder is available in IBM Storage Scale Advanced Edition, IBM Storage Scale Data Management Edition, IBM Storage Scale Developer Edition, or IBM Storage Scale Erasure Code Edition.
  • Only 25 cluster watches are supported per cluster.
    Note:
    • For more information about setting up a basic external sink to push events to, see https://kafka.apache.org/quickstart.
    • If you want to use a specific network interface for the brokers in the external Kafka sink, specify the advertised.listeners configuration parameter for the brokers with the IP address on the specified network interface card.
  • Clustered watch folder is supported in SELinux enforcing, permissive, and disabled modes. When clustered watch folder is run in enforcing mode, there is an extra event generated that is related to attributes due to the SELinux labeling of files.
  • Event generation occurs in the I/O path. So, to avoid delaying I/O operations while authentication and configuration occurs, the I/O operations that occur during that time span are not audited. On a per node basis, authentication and configuration occurs on the first I/O path after clustered watch is enabled, during initialization of the mmfsd server, and every time the file system is mounted.
  • Events of clustered watch are sent in batches to the Kafka brokers. This means that when a file system operation generates an event, the event is then appended to a batch rather than sent directly to Kafka. Therefore, if a file system operation returns successfully, it is not guaranteed that the operation is sent to the external Kafka queue. For example, if the node goes down after an operation completes but before the event batch is sent, that operation is never be received by Kafka. Rather than throttle the file system while waiting for Kafka to receive events, the file system attempts (for a short period of time) to send the events before dropping them.