In a HA setup one DDP Head is active. When a failover occurs the other Head takes over. This process on average takes a few seconds and in 99,99 % of all situations will not interrupt any client on reading or writing. Exceptions are for file commands which are not repeatable.
For its metadata and file system both Heads share a volume on one of the storage devices attached. For data high availability redundant storage array’s with write cache synchronisation are connected using multi path. When there is a failure in the data path a failover transparent for clients writing and reading occurs. When a few HPC clients need to be connected, AVFS and a cluster of nonredundant NVME-oF/RoCE or InfiniBand storage arrays is also a possibility. These can be with software defined raid and up to 200Gb/s cards.