Using balanced
or similar preferred content expressions makes files be
spread amoung repositories in a group, but does not take numcopies into
account. Could this be made to support numcopies?
The number of nodes to store each file on can be specified
(balanced=foo:N
). That is a separate number than numcopies. When
different files have different numcopies, a single number there is not
always useful. (Although it may be that someone always wants each file in
the cluster to have say, 3 copies, regardless of numcopies.)
Could balanced=foo:lackingcopies
just make enough nodes want copies to
satisfy numcopies? (Assuming there are enough nodes.)
That would not be the same as using balanced=foo or lackingcopies=1
,
because that makes every node want every file until numcopies is satisfied.
So that does not evenly balance files amoung nodes. Eg, if git-annex
sync
is syncing with 3 nodes and happens to always try sending to A first,
A would get every file, and B and C would not get any files with that
preferred content expression.