Make rados/rest bench work for multiple write instances without metadata conflict.

[guangyy]

rados bench is useful to benchmark the cluster, however, there is one user case it does not support: read after write if there are multiple write instance, this is because the current implementation hard code the metadata object which will make the last update win, as a result, the metadata of the previous run is lost and user is unable to run a read (seq) based on that round of write.

This patch provides user the ability to specify a run-name where to store the metadata (a default one is used if user does not provide one).

Signed-off-by: Guang Yang (yguang@yahoo-inc.com)

[gregsfortytwo]

This looks good to me. Can you describe how you tested it before I merge? :)

[guangyy]

Hi @gregsfortytwo,

1. I noticed one issue with rest-bench in terms of argument parsing and made a fix.
2. I tested the below scenarios:
   2.1. launch a couple of instances to write (one with default and others with run-name)
   2.2. run seq read (either with default or run-name), make sure they can success
   2.3. run cleanup for above meta (either with default, run-name or prefix)

All those test passed.

However, I did not end to end test rest-bench as I don't have an radosgw environment, but I suppose that is low risk (I used gdb to check those run-name and prefix parsing are correct for rest bench).

[gregsfortytwo]

Thanks!