doh, yea, this is dup of
HDFS-2378. Somehow when Eli and I were discussing it this morning I forgot that I'd already filed a JIRA for it. Since we already have the patch here, we can just commit this one and mark the other as dup, I guess?
In trunk the block recovery algorithm is quite different so I don't think it applies – the block recovery is triggered as part of the pipeline setup, rather than an RPC. But maybe we have a similar type of bug in trunk with timeouts in the pipeline setup process? Mind taking a look, Uma?