Rate limit IO-based queue wakeups to linger.ms (#2509)

Author: edenhill

src/rdkafka_partition.c

@@ -666,7 +666,7 @@ void rd_kafka_toppar_enq_msg (rd_kafka_toppar_t *rktp, rd_kafka_msg_t *rkm) {
         rd_kafka_toppar_unlock(rktp);
 
         if (wakeup_q) {
-                rd_kafka_q_yield(wakeup_q);
+                rd_kafka_q_yield(wakeup_q, rd_true/*rate-limit*/);
                 rd_kafka_q_destroy(wakeup_q);
         }
 }

src/rdkafka_queue.c

@@ -256,7 +256,7 @@ int rd_kafka_q_move_cnt (rd_kafka_q_t *dstq, rd_kafka_q_t *srcq,
 
 	if (!dstq->rkq_fwdq && !srcq->rkq_fwdq) {
 		if (cnt > 0 && dstq->rkq_qlen == 0)
-			rd_kafka_q_io_event(dstq);
+			rd_kafka_q_io_event(dstq, rd_false/*no rate-limiting*/);
 
 		/* Optimization, if 'cnt' is equal/larger than all
 		 * items of 'srcq' we can move the entire queue. */
@@ -730,6 +730,8 @@ void rd_kafka_q_io_event_enable (rd_kafka_q_t *rkq, int fd,
                 qio->fd = fd;
                 qio->size = size;
                 qio->payload = (void *)(qio+1);
+                qio->ts_rate = rkq->rkq_rk->rk_conf.buffering_max_us;
+                qio->ts_last = 0;
                 qio->event_cb = NULL;
                 qio->event_cb_opaque = NULL;
                 memcpy(qio->payload, payload, size);

src/rdkafka_queue.h

@@ -90,6 +90,8 @@ struct rd_kafka_q_io {
 	int    fd;
 	void  *payload;
 	size_t size;
+        rd_ts_t ts_rate;  /**< How often the IO wakeup may be performed (us) */
+        rd_ts_t ts_last;  /**< Last IO wakeup */
         /* For callback-based signalling */
         void (*event_cb) (rd_kafka_t *rk, void *opaque);
         void *event_cb_opaque;
@@ -284,10 +286,12 @@ static RD_INLINE RD_UNUSED int rd_kafka_q_is_fwded (rd_kafka_q_t *rkq) {
 /**
  * @brief Trigger an IO event for this queue.
  *
+ * @param rate_limit if true, rate limit IO-based wakeups.
+ *
  * @remark Queue MUST be locked
  */
 static RD_INLINE RD_UNUSED
-void rd_kafka_q_io_event (rd_kafka_q_t *rkq) {
+void rd_kafka_q_io_event (rd_kafka_q_t *rkq, rd_bool_t rate_limit) {
 
 	if (likely(!rkq->rkq_qio))
 		return;
@@ -297,6 +301,15 @@ void rd_kafka_q_io_event (rd_kafka_q_t *rkq) {
                 return;
         }
 
+
+        if (rate_limit) {
+                rd_ts_t now = rd_clock();
+                if (likely(rkq->rkq_qio->ts_last + rkq->rkq_qio->ts_rate > now))
+                        return;
+
+                rkq->rkq_qio->ts_last = now;
+        }
+
         /* Ignore errors, not much to do anyway. */
         if (rd_write(rkq->rkq_qio->fd, rkq->rkq_qio->payload,
                      (int)rkq->rkq_qio->size) == -1)
@@ -320,7 +333,7 @@ int rd_kafka_op_cmp_prio (const void *_a, const void *_b) {
  * @brief Wake up waiters without enqueuing an op.
  */
 static RD_INLINE RD_UNUSED void
-rd_kafka_q_yield (rd_kafka_q_t *rkq) {
+rd_kafka_q_yield (rd_kafka_q_t *rkq, rd_bool_t rate_limit) {
         rd_kafka_q_t *fwdq;
 
         mtx_lock(&rkq->rkq_lock);
@@ -337,12 +350,12 @@ rd_kafka_q_yield (rd_kafka_q_t *rkq) {
                 rkq->rkq_flags |= RD_KAFKA_Q_F_YIELD;
                 cnd_signal(&rkq->rkq_cond);
                 if (rkq->rkq_qlen == 0)
-                        rd_kafka_q_io_event(rkq);
+                        rd_kafka_q_io_event(rkq, rate_limit);
 
                 mtx_unlock(&rkq->rkq_lock);
         } else {
                 mtx_unlock(&rkq->rkq_lock);
-                rd_kafka_q_yield(fwdq);
+                rd_kafka_q_yield(fwdq, rate_limit);
                 rd_kafka_q_destroy(fwdq);
         }
 
@@ -413,7 +426,7 @@ int rd_kafka_q_enq1 (rd_kafka_q_t *rkq, rd_kafka_op_t *rko,
                 rd_kafka_q_enq0(rkq, rko, at_head);
                 cnd_signal(&rkq->rkq_cond);
                 if (rkq->rkq_qlen == 1)
-                        rd_kafka_q_io_event(rkq);
+                        rd_kafka_q_io_event(rkq, rd_false/*no rate-limiting*/);
 
                 if (do_lock)
                         mtx_unlock(&rkq->rkq_lock);
@@ -518,7 +531,7 @@ int rd_kafka_q_concat0 (rd_kafka_q_t *rkq, rd_kafka_q_t *srcq, int do_lock) {
 
 		TAILQ_CONCAT(&rkq->rkq_q, &srcq->rkq_q, rko_link);
 		if (rkq->rkq_qlen == 0)
-			rd_kafka_q_io_event(rkq);
+			rd_kafka_q_io_event(rkq, rd_false/*no rate-limiting*/);
                 rkq->rkq_qlen += srcq->rkq_qlen;
                 rkq->rkq_qsize += srcq->rkq_qsize;
 		cnd_signal(&rkq->rkq_cond);
@@ -559,7 +572,7 @@ void rd_kafka_q_prepend0 (rd_kafka_q_t *rkq, rd_kafka_q_t *srcq,
                 /* Move srcq to rkq */
                 TAILQ_MOVE(&rkq->rkq_q, &srcq->rkq_q, rko_link);
 		if (rkq->rkq_qlen == 0 && srcq->rkq_qlen > 0)
-			rd_kafka_q_io_event(rkq);
+			rd_kafka_q_io_event(rkq, rd_false/*no rate-limiting*/);
                 rkq->rkq_qlen += srcq->rkq_qlen;
                 rkq->rkq_qsize += srcq->rkq_qsize;