BUG/MINOR: activity: take care of late wakeups in "show tasks"

During the call to thread_isolate(), some other threads might have
performed some task_wakeup() which will have a call date past the
one we retrieved. It could be avoided by taking the current date
once we're alone but this would significantly affect the latency
measurements by adding the isolation time. Instead we're now only
accounting positive times, so that late wakeups normally appear
with a zero latency.

No backport is needed, this is 2.4.

diff --git a/src/activity.c b/src/activity.c
index 96f5008f32a9f9bdd483954197871a1c975f44be..0d945ac96446e4c66d67b6107fa7e3e1c5cc3eaa 100644 (file)
--- a/src/activity.c
+++ b/src/activity.c
@@ -209,7 +209,10 @@ static int cli_io_handler_show_tasks(struct appctx *appctx)
         * possible, using a sched_activity array to collect metrics with
         * limited collision, then we'll report statistics only. The tasks'
         * #calls will reflect the number of occurrences, and the lat_time will
-        * reflect the latency when set.
+        * reflect the latency when set. We prefer to take the time before
+        * calling thread_isolate() so that the wait time doesn't impact the
+        * measurement accuracy. However this requires to take care of negative
+        * times since tasks might be queued after we retrieve it.
         */
 
        now_ns = now_mono_time();
@@ -226,7 +229,8 @@ static int cli_io_handler_show_tasks(struct appctx *appctx)
                entry = sched_activity_entry(tmp_activity, t->process);
                if (t->call_date) {
                        lat = now_ns - t->call_date;
-                       entry->lat_time += lat;
+                       if ((int64_t)lat > 0)
+                               entry->lat_time += lat;
                }
                entry->calls++;
                rqnode = eb32sc_next(rqnode, ~0UL);
@@ -241,7 +245,8 @@ static int cli_io_handler_show_tasks(struct appctx *appctx)
                        entry = sched_activity_entry(tmp_activity, t->process);
                        if (t->call_date) {
                                lat = now_ns - t->call_date;
-                               entry->lat_time += lat;
+                               if ((int64_t)lat > 0)
+                                       entry->lat_time += lat;
                        }
                        entry->calls++;
                        rqnode = eb32sc_next(rqnode, ~0UL);
@@ -253,7 +258,8 @@ static int cli_io_handler_show_tasks(struct appctx *appctx)
                        entry = sched_activity_entry(tmp_activity, t->process);
                        if (!TASK_IS_TASKLET(t) && t->call_date) {
                                lat = now_ns - t->call_date;
-                               entry->lat_time += lat;
+                               if ((int64_t)lat > 0)
+                                       entry->lat_time += lat;
                        }
                        entry->calls++;
                }
@@ -265,7 +271,8 @@ static int cli_io_handler_show_tasks(struct appctx *appctx)
                                entry = sched_activity_entry(tmp_activity, t->process);
                                if (!TASK_IS_TASKLET(t) && t->call_date) {
                                        lat = now_ns - t->call_date;
-                                       entry->lat_time += lat;
+                                       if ((int64_t)lat > 0)
+                                               entry->lat_time += lat;
                                }
                                entry->calls++;
                        }