it would be nice to have a test accompanying this patch.

@jecluis I'll have a try, thanks.

@jecluis done, but, if the global size is too big, running cost of a long time.

@yangruifeng couldn't you just put a file worth a few bytes instead?

@jecluis if the file is too small, %USED is 0.

Showing the user bytes stored here is a deliberate choice (as is showing hte "max avail" that indicates how many more user bytes can be stored). If we want to show the raw storage consumed for the pool we should put that in a different column...

@liewegas
yes, i found that the %USED is not correct when i want to show a pie chart of USED.
it not notional or raw.
i think that the %USED should be notional-used/notional-pool-total,
and add a different column %RAW USED to show raw-used/raw-total.
what do you think? thanks.

@liewegas done.

So I've only seen it once, but in http://gitbuilder.sepia.ceph.com/gitbuilder-ceph-tarball-trusty-amd64-basic/log.cgi?log=e5585bc2c93199552e3263394e7c11d33e8f3c40 the test this adds is one of two causing make check to fail.

../qa/workunits/cephtool/test.sh:31: expect_false: test 0 '!=' 8192

It looks like it can race if stuff is veeeery slow which might be what's going on and I presume there are other solutions (one I see from looking quickly is checking several subsequent map epochs).

Although I haven't dug into this so it's vaguely possible the prior rbd test failure somehow mucked things up for this test? Seems unlikely though if I understand what this one is doing.

@gregsfortytwo maybe 'sleep 10' is not enough, change to

  #wait for update
  cal_raw_used_size=`ceph df detail | grep cephdf_for_test | awk -F ' ' '{printf "%d\n", 2 * $4}'`
  while [ $cal_raw_used_size -eq 0 ]
  do
    cal_raw_used_size=`ceph df detail | grep cephdf_for_test | awk -F ' ' '{printf "%d\n", 2 * $4}'`
  done