Hmm..almost looks like i thought this mostly because of a ui bug - i think perhaps its showing green for a moment when it should not. When i try and check the same thing through the zk tree, it looks right.

I did tighten things so that the leader for sure sees a down state before the replica registers its live node though.

I committed a bit to perhaps toughen this - but I'm not sure what is going on with the UI yet - though I've tried a bit of debugging. I've got to check out the zk servlet next - have not looked at that before.

rmuir20120906-bulk-40-change

moving all 4.0 issues not touched in a month to 4.1

note: i still see this

So I think it's the result of a node getting the updated live nodes and having stale cluster state...we should address it somehow.

To elaborate - we get the live nodes and cluster state independently - and live nodes should be much faster to get.

I've gotten to the bottom of this - patch attached.

[trunk commit] Mark Robert Miller
http://svn.apache.org/viewvc?view=revision&revision=1444239

SOLR-3655: A restarted node can briefly appear live and active before it really is in some cases.

[branch_4x commit] Mark Robert Miller
http://svn.apache.org/viewvc?view=revision&revision=1444241

SOLR-3655: A restarted node can briefly appear live and active before it really is in some cases.

Closed after release.