If we just wait for the old vnode to die, we are not guaranteed that the
new one will have yet been started and registered with the vnode
manager, so it's possible we will end up trying to do a call into the
old dead vnode in the subsequent test code. We saw a couple of test
failures in giddyup recently which I believe were caused by this race
condition.
To fix, we can wait for the vnode manager to return a new pid instead of
just waiting for the old pid to die.
All of the replication tests assume that num_nodes will be an integer,
not a list. Fix partition repair to do similar. Also, change
ho_concurrency to be a integer as well.
Changing from infinity (that's too long) to configure max waiting time.
This does not resolve other issues with the test that are still being
investigated: lots of data not found on first riak_kv vnode repaired in
some testers. But does prevent it from completely blocking riak_test
which happened on many machines.
This should resolve issue #113.
Added function to load code modules, clean directories within data and
using new put dir function where needed.
Changing die-die-die! approach to use cast in case remote call to node
to shoot itself in the head hangs.
