[Gluster-users] Gluster crashes when cascading AFR

Anand Avati avati at zresearch.com
Wed Dec 17 11:01:22 UTC 2008


Rainer,
  Please have a look at the configuration below. This configuration is
only for experimental purposes. What you should be interested in are
the options inside the afr volumes. You'll need to split this 2 level
graph into different vol files as well --  afrABCD on the client,
afrAB and afrCD in the distributor, posixA through posixD one per
machine (with protocol translators slipped in appropriately). Do use
write-behind at both the client and distributor levels, otherwise I'm
figuring you will see a lot of call bails during the write because of
the boosted expense of each write at the distributor level (from the
client's p.o.v).

  We're really interested to know how well/bad this configuration
works when you actually have 40x50 instead of 2x2 copies :-)

--snip--
volume posixA
  type storage/posix
  option directory /tmp/posixA
end-volume

volume posixB
  type storage/posix
  option directory /tmp/posixB
end-volume

volume posixC
  type storage/posix
  option directory /tmp/posixC
end-volume

volume posixD
  type storage/posix
  option directory /tmp/posixD
end-volume

volume afrAB
  type cluster/afr
  option data-self-heal off
  option metadata-self-heal off
  option entry-self-heal off
  option data-change-log off
  option metadata-change-log off
  option entry-change-log off
  option data-lock-server-count 0
  option metadata-lock-server-count 0
  option entry-lock-server-count 0
  subvolumes posixA posixB
end-volume


volume afrCD
  type cluster/afr
  option data-self-heal off
  option metadata-self-heal off
  option entry-self-heal off
  option data-change-log off
  option metadata-change-log off
  option entry-change-log off
  option data-lock-server-count 0
  option metadata-lock-server-count 0
  option entry-lock-server-count 0
  subvolumes posixC posixD
end-volume


volume afrABCD
  type cluster/afr
  option data-self-heal off
  option metadata-self-heal off
  option entry-self-heal off
  option data-change-log off
  option metadata-change-log off
  option entry-change-log off
  option data-lock-server-count 0
  option metadata-lock-server-count 0
  option entry-lock-server-count 0
  subvolumes afrAB afrCD
end-volume

--snip--

Avati




More information about the Gluster-users mailing list