Re: [Libguestfs] [PATCH nbdkit v2 3/3] filters: stats: Add flush stats
On Mon, Dec 2, 2019 at 12:28 PM Richard W.M. Jones <rjones(a)redhat.com> wrote:
I have pushed some parts of these patches in order to reduce the delta
between your patches and upstream. However still some problems with
the series:
Patch 1: Same problem with scale as discussed before.
I addressed this here:
https://www.redhat.com/archives/libguestfs/2019-December/msg00011.html
Patch 2: At least the documentation needs to be updated since it no
longer matches what is printed. The idea of collecting the time taken
in each operation is good on its own, so I pushed that part of it
along with small const-correctness and whitespace fixes:
https://github.com/libguestfs/nbdkit/commit/f280530d7d042d5e8f100125ab061...
Thanks for pushing it.
Why don't we show the total time and time / operation on each
line of
output (ie. per operation), instead of synthesizing the total by
adding up reads and writes?
I think that synthesizing totals give better view on total application
throughput, and add
information that was not available before, like total number of ops.
Showing two rate
values per operation looks confusing to me.
But how about this:
----------------------------------------------
total: 2299 ops, 2.172 s, 6.00 GiB, 2.76 GiB/s
520.73 MiB/s write, 2.23 GiB/s zero
-----------------------------------------------
write: 1271 ops, 0.356 s, 1.13 GiB, 3.19 GiB/s
zero: 1027 ops, 0.012 s, 4.86 GiB, 405.00 GiB/s
extents: 1 ops, 0.000 s, 2.00 GiB, 485.29 GiB/s
flush: 2 ops, 1.252 s
Patch 3: The idea of collecting flush stats is good so I pushed that
part, as well as updating the documentation:
https://github.com/libguestfs/nbdkit/commit/8adf601835aee3779e278e13cae04...
Nice, I missed the pod file.
Nir