hadoop job progress going back

[Doopah Shaf]

Running jobs sometimes displays progress as follows:

map 100% reduce 10%

and then

map 100% reduce 0%

What causes this? Is this speculative execution restarting tasks?

Cheers,
DS