Re: glusterfs

[brook%biology.nmsu.edu@localhost (Brook Milligan)]

Manuel Bouyer writes:
 > On Mon, Dec 05, 2011 at 01:06:07PM -0700, Brook Milligan wrote:
 > I don't think defaults is a recognised keyword for NetBSD's mount,
 > and _netdev is probably not. 'rw' should be enough.

Thanks.  That works.

So I have an understanding of how to set up a server running glusterd
and set up volumes with gluster.  Here is a summary of my
understanding:

- Server: run glusterd; configure/start volumes with gluster
- Client: mount volumes with fstab

Is that really all there is to it, even for complex setups with
multiple servers and volumes?

How is the performance relative to, for example, traditional nfs
mounts?

How do glusterfs and glusterfsd fit into the picture?  Are they an
alternative means of accomplishing the same?

Finally, I have a few observations as a result of making several
mistakes setting this up.  Perhaps they indicate bugs; alternatively
they may just be my mistakes.

- First, it seems that the root filesystem cannot have extended
  attributes.  If it does, I get the following trap as soon as it is
  mounted (copied by hand):

  root on wd0a dumps on wd0b
  fatal integer divide fault in supervisor mode
  trap type 0 code 0 eip c0470014 cs8 eflags 10246 cr2 cd222000 ilevel 0
  kernel: supervisor trap interger divide fault, code=0
  Stopped in pid 0.1 (system) at netbsd:ufs_bmaparray+0x64: divl 0x50(%exc),%eax

  The backtrace clearly indicates that it is coming from somewhere in
  the extended attribute handling.

- Second, if for some reason perfused does not successfully mount a
  filesystem, it hangs the system during shutdown.  Shouldn't this be
  successfully killed off during the shutdown process?

Thanks for your help getting this going.

Cheers,
Brook