BUG/MEDIUM: tasks: make sure we pick all tasks in the run queue
Commit 09eeb76 ("BUG/MEDIUM: tasks: Don't forget to increase/decrease
tasks_run_queue.") addressed a count issue in the run queue and uncovered
another issue with the way the tasks are dequeued from the global run
queue. The number of tasks to pick is computed using an integral divide,
which results in up to nbthread-1 tasks never being run. The fix simply
consists in getting rid of the divide and checking the task count in the
loop.
No backport is needed, this is 1.9-specific.
diff --git a/src/task.c b/src/task.c
index b0f9b49..86833cf 100644
--- a/src/task.c
+++ b/src/task.c
@@ -251,9 +251,6 @@
{
struct task *t;
int max_processed;
-#ifdef USE_THREAD
- uint64_t average = 0;
-#endif
tasks_run_queue_cur = tasks_run_queue; /* keep a copy for reporting */
nb_tasks_cur = nb_tasks;
@@ -268,15 +265,12 @@
}
#ifdef USE_THREAD
- average = tasks_run_queue / global.nbthread;
-
/* Get some elements from the global run queue and put it in the
* local run queue. To try to keep a bit of fairness, just get as
* much elements from the global list as to have a bigger local queue
* than the average.
*/
- while ((task_list_size[tid] + rqueue_size[tid]) <= average) {
-
+ while ((task_list_size[tid] + rqueue_size[tid]) * global.nbthread <= tasks_run_queue) {
/* we have to restart looking up after every batch */
rq_next = eb32sc_lookup_ge(&rqueue, rqueue_ticks - TIMER_LOOK_BACK, tid_bit);
if (unlikely(!rq_next)) {