[MEDIUM] introduce task->nice and boot access to statistics
The run queue scheduler now considers task->nice to queue a task and
to pick a task out of the queue. This makes it possible to boost the
access to statistics (both via HTTP and UNIX socket). The UNIX socket
receives twice as much a boost as the HTTP socket because it is more
sensible.
diff --git a/include/proto/task.h b/include/proto/task.h
index f99bbdb..de50cba 100644
--- a/include/proto/task.h
+++ b/include/proto/task.h
@@ -32,7 +32,8 @@
#include <types/task.h>
-extern unsigned int run_queue; /* run queue size */
+extern unsigned int run_queue; /* run queue size */
+extern unsigned int niced_tasks; /* number of niced tasks in the run queue */
extern struct pool_head *pool2_task;
/* perform minimal initializations, report 0 in case of error, 1 if OK. */
@@ -50,6 +51,8 @@
t->state = TASK_IDLE;
eb32_delete(&t->eb);
run_queue--;
+ if (likely(t->nice))
+ niced_tasks--;
}
return t;
}
@@ -65,9 +68,11 @@
if (t->eb.node.leaf_p)
eb32_delete(&t->eb);
- if (t->state == TASK_RUNNING)
+ if (t->state == TASK_RUNNING) {
run_queue--;
-
+ if (likely(t->nice))
+ niced_tasks--;
+ }
return t;
}
@@ -79,6 +84,7 @@
{
t->eb.node.leaf_p = NULL;
t->state = TASK_IDLE;
+ t->nice = 0;
return t;
}
diff --git a/include/types/task.h b/include/types/task.h
index b6c9c72..4e579c1 100644
--- a/include/types/task.h
+++ b/include/types/task.h
@@ -39,6 +39,7 @@
struct timeval expire; /* next expiration time for this task, use only for fast sorting */
void (*process)(struct task *t, struct timeval *next); /* the function which processes the task */
void *context; /* the task's context */
+ int nice; /* the task's current nice value from -1024 to +1024 */
};
#endif /* _TYPES_TASK_H */
diff --git a/src/proto_http.c b/src/proto_http.c
index 7b35ff7..308ecb0 100644
--- a/src/proto_http.c
+++ b/src/proto_http.c
@@ -5160,6 +5160,7 @@
t->logs.tv_request = now;
t->data_source = DATA_SRC_STATS;
t->data_state = DATA_ST_INIT;
+ t->task->nice = -32; /* small boost for HTTP statistics */
produce_content(t);
return 1;
}
diff --git a/src/proto_uxst.c b/src/proto_uxst.c
index c4c3d37..6968562 100644
--- a/src/proto_uxst.c
+++ b/src/proto_uxst.c
@@ -447,6 +447,7 @@
task_init(t);
t->process = l->handler;
t->context = s;
+ t->nice = -64; /* we want to boost priority for local stats */
s->task = t;
s->fe = NULL;
diff --git a/src/task.c b/src/task.c
index dd9638a..b28ae0b 100644
--- a/src/task.c
+++ b/src/task.c
@@ -24,6 +24,7 @@
struct pool_head *pool2_task;
unsigned int run_queue = 0;
+unsigned int niced_tasks = 0; /* number of niced tasks in the run queue */
/* Principle of the wait queue.
*
@@ -125,7 +126,11 @@
return pool2_task != NULL;
}
-/* puts the task <t> in run queue <q>, and returns <t> */
+/* Puts the task <t> in run queue at a position depending on t->nice.
+ * <t> is returned. The nice value assigns boosts in 32th of the run queue
+ * size. A nice value of -1024 sets the task to -run_queue*32, while a nice
+ * value of 1024 sets the task to run_queue*32.
+ */
struct task *task_wakeup(struct task *t)
{
if (t->state == TASK_RUNNING)
@@ -136,6 +141,18 @@
run_queue++;
t->eb.key = ++rqueue_ticks;
+
+ if (likely(t->nice)) {
+ int offset;
+
+ niced_tasks++;
+ if (likely(t->nice > 0))
+ offset = (unsigned)((run_queue * (unsigned int)t->nice) / 32U);
+ else
+ offset = -(unsigned)((run_queue * (unsigned int)-t->nice) / 32U);
+ t->eb.key += offset;
+ }
+
t->state = TASK_RUNNING;
eb32_insert(&rqueue[ticks_to_tree(t->eb.key)], &t->eb);
@@ -218,10 +235,10 @@
* used to assign a position to each task. This counter may be combined with
* other variables (eg: nice value) to set the final position in the tree. The
* counter may wrap without a problem, of course. We then limit the number of
- * tasks processed at once to 1/8 of the number of tasks in the queue, so that
- * general latency remains low and so that task positions have a chance to be
- * considered. It also reduces the number of trees to be evaluated when no task
- * remains.
+ * tasks processed at once to 1/4 of the number of tasks in the queue, and to
+ * 200 max in any case, so that general latency remains low and so that task
+ * positions have a chance to be considered. It also reduces the number of
+ * trees to be evaluated when no task remains.
*
* Just like with timers, we start with tree[(current - 1)], which holds past
* values, and stop when we reach the middle of the list. In practise, we visit
@@ -240,7 +257,12 @@
if (!run_queue)
return;
- max_processed = (run_queue + 7) / 8;
+ max_processed = run_queue;
+ if (max_processed > 200)
+ max_processed = 200;
+
+ if (likely(niced_tasks))
+ max_processed /= 4;
tree = ticks_to_tree(rqueue_ticks);
stop = (tree + TIMER_TREES / 2) & TIMER_TREE_MASK;
@@ -255,6 +277,8 @@
eb = eb32_next(eb);
run_queue--;
+ if (likely(t->nice))
+ niced_tasks--;
t->state = TASK_IDLE;
eb32_delete(&t->eb);