I would like to know in an (2+1) Erasure coding storage setup, can it detect corruption when one node is lost, and there is a silent data corruption from HDD on one of the remaining node.
I am experimenting GlusterFS dispersed volume which use erasure coding.
I created a 3 node (2+1) dispersed volume (erasure coding). Put a file in it. Then I poweroff one of the node. On one of the remaining node, I use vi to modify a single character of that file in the underlying storage (bricks) of that volume. Then restore the modification timestamp of that file.
Then, I use sha256sum to read that file from the volume, Gluster return Input/Output error (EIO).
Is this the effect of Erasure coding, or the effect GlusterFS's metadata to detect this kind of corruption.
