The shared storage must be redundant (for example, with RAID arrays) to avoid a single point
of failure.
SAN or NAS external storage is supported.
Note:
Be aware of the following limitations with SAN and NAS:
- Ensure that Orchestrator is not running on virtual machines if your storage is based on
SAN Fiber Channel technology, since virtual machines typically do not support SAN Fiber
Channel storage.
- A SAN storage requires a cluster file system (for example, GFS) for our HA solution to
work. A block level storage access will not work; only a file level storage access is
suitable. Files must be visible from both nodes concurrently. If a file is changed from
one node, the other node must be able to immediately incorporate the changes. For example,
a storage mount based on raw device mapping (RDM) is at the block level. RDM is a VMware
server virtualization environment that enables storage to be directly connected to a
virtual machine from the SAN. This will work, as file updates on one node are not visible
to the other node until the shared storage is remounted.
- A NAS storage based on NFS requires NFS Version 4. A solution based on NFS Version 3 may
cause problems, because NFS3 does not automatically release the locks held by a client.
This is critical for MySQL, where only one node at a time may write to the shared disk;
the Orchestrator MySQL version does not support locking on concurrent writes. With NFS3,
if the active node that is running MySQL goes down, the passive node is unlikely to be
able to start MySQL when it takes over, because some MySQL files may still be locked on
the shared storage. With NFS4 technology, all locks associated with a TCP session that
times out will be automatically released, making MySQL automatic failover possible. A NAS
storage based on CIFS (for example, Samba) will work, because there are no locking
issues.
MySQL data, ACM, and Aspera Orchestrator var data may be configured on separate
mount points, but this is not a requirement. Configure at least 20GB for the MySQL data, 1 GB
for ACM, and 20GB for the Orchestrator var data. Because ACM is not migrating the
mount points over failover, MySQL, Orchestrator var data ,and ACM data must be
mounted and visible on both nodes simultaneously.