[Gluster-devel] Questions on Gluster fault tolerance

[Cheng Shao]

Howdy!

I'm having some questions regarding how Gluster deals with various
failure cases:
- if a client obtained write lock on a file using fcntl, and later
crashed, can Gluster reclaim the lock so that other clients can access
the file? If so, what is the timeout to trigger the lock reclaim?
- if a client cannot reach one server of a replication pair for a read
operation, can the client automatically retry on the other server?
- if a client cannot reach either node of a replication pair, can the
client tries to send request to other server nodes?
- Is AFR being done among server nodes? What if some nodes in a
replication set is not reachable? Will gluster fail the write to the
client?

Also, my understanding is from 3.1.x on, gluster is using volume
command to configure the server volumes. Any change to the translators
should be set using volume command instead of changing the vol file
manually. Is my understanding correct?

thanks so much,
- cheng