Skip to main content
Version: 4.3.0

Storage device setup/removal

All storage devices used by chunkserver are defined in sfshdd.cfg (see man sfshdd.cfg for more details). They need to be mounted to a specific directory and defined in the file. The chunkserver will assume it's dedicated, but it's not strictly necessary for it to be (but highly recommended).

After any modification to the sfshdd.cfg files, the chunkserver needs to be reloaded, by:

systemctl reload saunafs-chunkserver

OR

sfschunkserver reload

OR sending a SIGHUP signal to the running process.

While any filesystem can be used, we recommend XFS for CMR drives with the following options:

/dev/sda /mnt/sfschunkservers/data/sda xfs rw,noexec,nofail,nodev,noatime,nodiratime,largeio,allocsize=16777216,inode64 0 0

allocsize can vary depending on the goal replication. For simple goals it should be 64 MiB, but for EC this will depend on the number of data parts. For example, with EC(4,3) (4 data parts and 3 parities), the alloc size should be (64 MiB / 4) 16 MiB (16777216). Or with EC(3,1), it should be (64 MiB / 3) 22.4 MiB (22369621)

Adding storage devices​

Adding a storage device is to simply point to the directory where it's mounted. For example, if you had /dev/sda mounted in /mnt/hdd1/, you need to simply add /mnt/hdd1 to the file.

Replacing storage devices (non-damaged)​

If you wish to replace the storage devices (provided they are still working), it may be best to mark them for removal with *. From the previous example, to replace /dev/sda, add * to the beginning:

*/mnt/hdd1

It will start copying files to other devices. Once the process finishes (all the chunks are copied to other drives), it can safely be removed by deleting or commenting the line.

Removing/replacing storage devices​

Removing the devices is achieved by simply deleting/commenting out the line. In this case the device will be treated as missing and, after some delay (defined by OPERATIONS_DELAY_INIT in sfsmaster.cfg), will start replicating from other devices to restore the chunks.

With bigger data, this is much faster than moving the data out of a device, but will put the entire system under strain as it will need to replicate data to elsewhere.