Github user yingdachen commented on the issue:
https://github.com/apache/tez/pull/33
Will be addressing the review comments in the next few days.
In addition, some more details will be provided for fault tolerance in
terms of AM recovery. I am thinking that to begin with, we will at least
guarantee that the introduction of concurrent edge would not interfere with
normal AM recovery.
Design for fully functional recovery (for example, making the entire
sub-graph connected by concurrent edge a a failure region) will likely to
follow after that, since there are quite some details to polish.
---
