On Wed, Nov 23, 2011 at 10:41:32AM -0700, Eric Blake wrote:
Hmm - passing nparams==1 to set just read_bytes_sec has the side
effect
of wiping out any existing total_iops_sec, even though those two
parameters seem somewhat orthogonal, all because we zero-initialized the
struct that we pass on to the monitor command. Is that intended? I can
live with it (but it probably ought to be documented), but we may want
to consider being more flexible, by using '0' to clear a previous limit,
but initializing to '-1' to imply the limit does not change. Then the
qemu_monitor_json code should only emit the arguments that are actually
being changed, rather than blindly always outputting 6 parameters even
when the user only passed in one parameter. But I'm okay delaying that
to a separate patch based on whether others think it would be a good
improvement.
+1. I believe I had pointed this out previously as well (albeit not as
concisely as this).
--
Adam Litke <agl(a)us.ibm.com>
IBM Linux Technology Center