Merge pull request #285 from kronosnet/kill-latency

[links] kill redundant latency in link status and move it to stats
kronosnet · Mar 1, 2020 · f8b3f58 · f8b3f58
2 parents 8ac7c52 + cfd074d
commit f8b3f58
Show file tree

Hide file tree

Showing 5 changed files with 12 additions and 26 deletions.
diff --git a/libknet/internals.h b/libknet/internals.h
@@ -66,7 +66,6 @@ struct knet_link {
 	unsigned long long pong_timeout_adj;	/* timeout adjusted for latency */
 	uint8_t pong_timeout_backoff;		/* see link.h for definition */
 	unsigned int latency_max_samples;	/* precision */
-	unsigned int latency_cur_samples;
 	uint8_t pong_count;			/* how many ping/pong to send/receive before link is up */
 	uint64_t flags;
 	/* status */

diff --git a/libknet/libknet.h b/libknet/libknet.h
@@ -1962,9 +1962,6 @@ int knet_link_get_link_list(knet_handle_t knet_h, knet_node_id_t host_id,
  * dynconnected          the link has dynamic ip on the other end, and
  *                       we can see the other host is sending pings to us.
  *
- * latency               average latency of this link
- *                       see also knet_link_set/get_timeout.
- *
  * pong_last             if the link is down, this value tells us how long
  *                       ago this link was active. A value of 0 means that the link
  *                       has never been active.
@@ -2042,7 +2039,6 @@ struct knet_link_status {
 	uint8_t enabled;	        /* link is configured and admin enabled for traffic */
 	uint8_t connected;              /* link is connected for data (local view) */
 	uint8_t dynconnected;	        /* link has been activated by remote dynip */
-	unsigned long long latency;	/* average latency computed by fix/exp */
 	struct timespec pong_last;
 	unsigned int mtu;		/* current detected MTU on this link */
 	unsigned int proto_overhead;    /* contains the size of the IP protocol, knet headers and

diff --git a/libknet/links.c b/libknet/links.c
@@ -261,7 +261,7 @@ int knet_link_set_config(knet_handle_t knet_h, knet_node_id_t host_id, uint8_t l
 	link->pong_timeout_backoff = KNET_LINK_PONG_TIMEOUT_BACKOFF;
 	link->pong_timeout_adj = link->pong_timeout * link->pong_timeout_backoff; /* microseconds */
 	link->latency_max_samples = KNET_LINK_DEFAULT_PING_PRECISION;
-	link->latency_cur_samples = 0;
+	link->status.stats.latency_samples = 0;
 	link->flags = flags;
 
 	savederrno = pthread_mutex_init(&link->link_stats_mutex, NULL);

diff --git a/libknet/threads_heartbeat.c b/libknet/threads_heartbeat.c
@@ -190,7 +190,7 @@ static void _adjust_pong_timeouts(knet_handle_t knet_h)
 				dst_link->pong_timeout_backoff--;
 			}
 
-			dst_link->pong_timeout_adj = (dst_link->pong_timeout * dst_link->pong_timeout_backoff) + (dst_link->status.latency * KNET_LINK_PONG_TIMEOUT_LAT_MUL);
+			dst_link->pong_timeout_adj = (dst_link->pong_timeout * dst_link->pong_timeout_backoff) + (dst_link->status.stats.latency_ave * KNET_LINK_PONG_TIMEOUT_LAT_MUL);
 		}
 	}
 

diff --git a/libknet/threads_rx.c b/libknet/threads_rx.c
@@ -668,18 +668,18 @@ static void _parse_recv_from_links(knet_handle_t knet_h, int sockfd, const struc
 			 * in words : ('previous mean' * '(count -1)') + 'new value') / 'count'
 			 */
 
-			src_link->latency_cur_samples++;
+			src_link->status.stats.latency_samples++;
 
 			/*
 			 * limit to max_samples (precision)
 			 */
-			if (src_link->latency_cur_samples >= src_link->latency_max_samples) {
-				src_link->latency_cur_samples = src_link->latency_max_samples;
+			if (src_link->status.stats.latency_samples >= src_link->latency_max_samples) {
+				src_link->status.stats.latency_samples = src_link->latency_max_samples;
 			}
-			src_link->status.latency =
-				(((src_link->status.latency * (src_link->latency_cur_samples - 1)) + (latency_last / 1000llu)) / src_link->latency_cur_samples);
+			src_link->status.stats.latency_ave =
+				(((src_link->status.stats.latency_ave * (src_link->status.stats.latency_samples - 1)) + (latency_last / 1000llu)) / src_link->status.stats.latency_samples);
 
-			if (src_link->status.latency < src_link->pong_timeout_adj) {
+			if (src_link->status.stats.latency_ave < src_link->pong_timeout_adj) {
 				if (!src_link->status.connected) {
 					if (src_link->received_pong >= src_link->pong_count) {
 						log_info(knet_h, KNET_SUB_RX, "host: %u link: %u is up",
@@ -693,21 +693,12 @@ static void _parse_recv_from_links(knet_handle_t knet_h, int sockfd, const struc
 				}
 			}
 			/* Calculate latency stats */
-			if (src_link->status.latency > src_link->status.stats.latency_max) {
-				src_link->status.stats.latency_max = src_link->status.latency;
+			if (src_link->status.stats.latency_ave > src_link->status.stats.latency_max) {
+				src_link->status.stats.latency_max = src_link->status.stats.latency_ave;
 			}
-			if (src_link->status.latency < src_link->status.stats.latency_min) {
-				src_link->status.stats.latency_min = src_link->status.latency;
+			if (src_link->status.stats.latency_ave < src_link->status.stats.latency_min) {
+				src_link->status.stats.latency_min = src_link->status.stats.latency_ave;
 			}
-
-			/*
-			 * those 2 lines below make all latency average calculations consistent and capped to
-			 * link precision. In future we will kill the one above to keep only this one in
-			 * the stats structure, but for now we leave it around to avoid API/ABI
-			 * breakage as we backport the fixes to stable
-			 */
-			src_link->status.stats.latency_ave = src_link->status.latency;
-			src_link->status.stats.latency_samples = src_link->latency_cur_samples;
 		}
 		break;
 	case KNET_HEADER_TYPE_PMTUD: