Max cluster capacity

The capacity of a MooseFS cluster is the sum of the raw disk space that its Chunkservers report. MooseFS does not configure or check that sum. There is no parameter that caps total capacity. A Chunkserver or a disk is never refused because the cluster is already large.

The ceiling comes from the counter that holds the number. A Chunkserver reports used and total space as a 64-bit byte count. The Master Server adds those counts into a 64-bit total. The client receives the same 64-bit value. The client converts this value into what df reports for the mount. The largest capacity that MooseFS can represent along that path is 2⁶⁴ bytes: 16 EiB, less one byte.

Two things this figure is not.

This figure is not user space. A Chunkserver reports the raw space of its disks, so redundancy uses part of that space. With three copies, you write one byte of file for every three bytes of raw space. With Erasure Coding 8+2, you write eight bytes of file for every ten bytes of raw space. Base your plan on user space, not on this number.

A single mount does not always show this figure. A macOS client limits the value that it passes to statvfs to the largest value that fits, because the macOS kernel expects 32-bit block counts. df on a macOS mount therefore shows at most 512 TiB, however large the cluster is. The cluster is unaffected. Only that one reading is affected.