On 07/16/2014 03:53 PM, Michal Privoznik wrote:
There's this question on the list that is asked over and over
again.
How do I get {cpu, memory, ...} usage in percentage? Or its modified
version: How do I plot nice graphs like virt-manager does?
It would be nice if we have an example to inspire people. And that's
what domtop should do. Yes, it could be written in different ways, but
I've chosen this one as I think it show explicitly what users need to
implement in order to imitate virt-manager's graphing.
Signed-off-by: Michal Privoznik <mprivozn(a)redhat.com>
---
.gitignore | 1 +
Makefile.am | 2 +-
cfg.mk | 2 +-
configure.ac | 1 +
examples/domtop/Makefile.am | 27 +++
examples/domtop/domtop.c | 388 ++++++++++++++++++++++++++++++++++++++++++++
libvirt.spec.in | 2 +-
7 files changed, 420 insertions(+), 3 deletions(-)
create mode 100644 examples/domtop/Makefile.am
create mode 100644 examples/domtop/domtop.c
+
+static void
+print_cpu_usage(const char *dom_name,
+ size_t cpu,
+ size_t ncpus,
+ unsigned long long then,
+ virTypedParameterPtr then_params,
+ size_t then_nparams,
+ unsigned long long now,
+ virTypedParameterPtr now_params,
+ size_t now_nparams)
+{
+ size_t i, j, k;
+ size_t nparams = now_nparams;
+
+ if (then_nparams != now_nparams) {
+ /* this should not happen (TM) */
+ ERROR("parameters counts don't match");
+ return;
+ }
+
+ for (i = 0; i < ncpus; i++) {
+ size_t pos;
+ double usage;
+
+ /* check if the vCPU is in the maps */
+ if (now_params[i * nparams].type == 0 ||
+ then_params[i * then_nparams].type == 0)
+ continue;
+
+ for (j = 0; j < nparams; j++) {
+ pos = i * nparams + j;
+ if (STREQ(then_params[pos].field, VIR_DOMAIN_CPU_STATS_CPUTIME) ||
+ STREQ(then_params[pos].field, VIR_DOMAIN_CPU_STATS_VCPUTIME))
+ break;
+ }
+
+ if (j == nparams) {
+ ERROR("unable to find %s", VIR_DOMAIN_CPU_STATS_CPUTIME);
+ return;
+ }
+
+ DEBUG("now_params=%llu then_params=%llu now=%llu then=%llu",
+ now_params[pos].value.ul, then_params[pos].value.ul, now, then);
+
+ /* @now_params and @then_params are in nanoseconds, @now and @then are
+ * in microseconds. In ideal world, we would translate them both into
+ * the same scale, divide one by another and multiply by factor of 100
+ * to get percentage. However, the count of floating point operations
+ * performed has a bad affect on the precision, so instead of dividing
s/affect/effect/
+ * @now_params and @then_params by 1000 and then multiplying
again by
+ * 100, we divide only once by 10 and get the same result. */
+ usage = (now_params[pos].value.ul - then_params[pos].value.ul) /
+ (now - then) / 10;
+
+ printf("CPU%zu: %.2lf\n", cpu + i, usage);
I think printing all the CPUs on one line would look nicer, and it would be
easier to see which numbers are changing.
ACK
Jan