Workaround for iowait being decremented

The iowait value reported by the kernel on NO_HZ systems can decrement
as a result of inaccurate iowait tracking. Waiting on IO can be first
accounted as iowait but then instead as idle.

Function get_per_cpu_interval() considers iowait going backwards between
two readings as a CPU coming back online and resets the iowait value of
the first reading to 0. If iowait is decremented only because of
inaccurate tracking, this causes that almost all time between the two
readings is incorrectly recognized by sar as being spent in iowait.

The patch updates the code in get_per_cpu_interval() to recognize this
situation. If the iowait value between two readings decremented but the
idle value did not then the code now considers it as a problem with the
iowait reporting and corrects the first value according to the second
reading. Otherwise, the code remains treating decremented iowait as a
CPU coming back online.

Fixes #14.

Signed-off-by: Sebastien GODARD <sysstat@users.noreply.github.com>

diff --git a/rd_stats.c b/rd_stats.c
index 56d42d006797d5a6f6534aa5d4806a24d711954f..fb93f23fed12bcef96f8e731ea915a749f34c773 100644 (file)
--- a/rd_stats.c
+++ b/rd_stats.c
@@ -440,12 +440,26 @@ unsigned long long get_per_cpu_interval(struct stats_cpu *scc,
         * value was greater than ULLONG_MAX - 0x7ffff (the counter probably
         * overflew).
         */
+       if ((scc->cpu_iowait < scp->cpu_iowait) && (scp->cpu_iowait < (ULLONG_MAX - 0x7ffff))) {
+               /*
+                * The iowait value reported by the kernel can also decrement as
+                * a result of inaccurate iowait tracking. Waiting on IO can be
+                * first accounted as iowait but then instead as idle.
+                * Therefore if the idle value during the same period did not
+                * decrease then consider this is a problem with the iowait
+                * reporting and correct the previous value according to the new
+                * reading. Otherwise, treat this as CPU coming back online.
+                */
+               if ((scc->cpu_idle > scp->cpu_idle) || (scp->cpu_idle >= (ULLONG_MAX - 0x7ffff))) {
+                       scp->cpu_iowait = scc->cpu_iowait;
+               }
+               else {
+                       scp->cpu_iowait = 0;
+               }
+       }
        if ((scc->cpu_idle < scp->cpu_idle) && (scp->cpu_idle < (ULLONG_MAX - 0x7ffff))) {
                scp->cpu_idle = 0;
        }
-       if ((scc->cpu_iowait < scp->cpu_iowait) && (scp->cpu_iowait < (ULLONG_MAX - 0x7ffff))) {
-               scp->cpu_iowait = 0;
-       }
 
        /*
         * Don't take cpu_guest and cpu_guest_nice into account