Yes, Milind is currently looking at this. We're hoping to get a design
nailed over the next couple of weeks.
It might make sense to do this in two phases, with the first phase
simply writing the image and edits to multiple volumes. The second part
of the proposal will likely need a lot more discussion on Jira.
Backup namenodes are hard to do. We'd need to replicate not just the
image and edits but also the blocks map. Keeping backups in sync and
dealing with failures can quickly become hairy. An alternative might be
to replicate the image and edits to a subset of the datanodes.
Let's get a preliminary design posted and continue to discuss from there.
Doug Cutting wrote:
Torsten Curdt wrote:
http://mail-archives.apache.org/mod_mbox/lucene-hadoop-dev/200609.mbox/[EMAIL PROTECTED]
Someone started yet?
Good question. Milind, are you currently working on this? Should we
start a Jira issue for at least the first part of the proposal (write
namenode data to multiple volumes)?
Doug
