stat-segment: scaling improvement

don't walk the entire list of entries each time a new one is added

approximate/indicative numbers recorded on a VM:
after:
  50000 tunnels in 1.124443 secs, 44466.45 tunnels/sec
before:
  50000 tunnels in 5.202779 secs, 9610.25 tunnels/sec

Change-Id: Ie1155416be76f55f2a0a88360ce53e15aeace785
Signed-off-by: Neale Ranns <nranns@cisco.com>

3 files changed, 41 insertions, 18 deletions

diff --git a/src/vlib/counter.c b/src/vlib/counter.c
index 5c3350fceec..5b73d5fa21a 100644
--- a/src/vlib/counter.c
+++ b/src/vlib/counter.c
@@ -74,16 +74,16 @@ vlib_clear_combined_counters (vlib_combined_counter_main_t * cm)
     }
 }
 
-void *vlib_stats_push_heap (void) __attribute__ ((weak));
+void *vlib_stats_push_heap (void *) __attribute__ ((weak));
 void *
-vlib_stats_push_heap (void)
+vlib_stats_push_heap (void *unused)
 {
   return 0;
 };
 
-void vlib_stats_pop_heap (void *, void *, int) __attribute__ ((weak));
+void vlib_stats_pop_heap (void *, void *, u32, int) __attribute__ ((weak));
 void
-vlib_stats_pop_heap (void *notused, void *notused2, int type)
+vlib_stats_pop_heap (void *notused, void *notused2, u32 i, int type)
 {
 };
 
@@ -92,13 +92,13 @@ vlib_validate_simple_counter (vlib_simple_counter_main_t * cm, u32 index)
 {
   vlib_thread_main_t *tm = vlib_get_thread_main ();
   int i;
-  void *oldheap = vlib_stats_push_heap ();
+  void *oldheap = vlib_stats_push_heap (cm->counters);
 
   vec_validate (cm->counters, tm->n_vlib_mains - 1);
   for (i = 0; i < tm->n_vlib_mains; i++)
     vec_validate_aligned (cm->counters[i], index, CLIB_CACHE_LINE_BYTES);
 
-  vlib_stats_pop_heap (cm, oldheap,
+  vlib_stats_pop_heap (cm, oldheap, index,
 		       2 /* STAT_DIR_TYPE_COUNTER_VECTOR_SIMPLE */ );
 }
 
@@ -107,13 +107,13 @@ vlib_validate_combined_counter (vlib_combined_counter_main_t * cm, u32 index)
 {
   vlib_thread_main_t *tm = vlib_get_thread_main ();
   int i;
-  void *oldheap = vlib_stats_push_heap ();
+  void *oldheap = vlib_stats_push_heap (cm->counters);
 
   vec_validate (cm->counters, tm->n_vlib_mains - 1);
   for (i = 0; i < tm->n_vlib_mains; i++)
     vec_validate_aligned (cm->counters[i], index, CLIB_CACHE_LINE_BYTES);
 
-  vlib_stats_pop_heap (cm, oldheap,
+  vlib_stats_pop_heap (cm, oldheap, index,
 		       3 /*STAT_DIR_TYPE_COUNTER_VECTOR_COMBINED */ );
 }
 
diff --git a/src/vpp/stats/stat_segment.c b/src/vpp/stats/stat_segment.c
index 8496f840aaf..d2a410cb3a5 100644
--- a/src/vpp/stats/stat_segment.c
+++ b/src/vpp/stats/stat_segment.c
@@ -48,10 +48,11 @@ vlib_stat_segment_unlock (void)
  * Change heap to the stats shared memory segment
  */
 void *
-vlib_stats_push_heap (void)
+vlib_stats_push_heap (void *old)
 {
   stat_segment_main_t *sm = &stat_segment_main;
 
+  sm->last = old;
   ASSERT (sm && sm->shared_header);
   return clib_mem_set_heap (sm->heap);
 }
@@ -79,7 +80,8 @@ lookup_or_create_hash_index (void *oldheap, char *name, u32 next_vector_index)
 }
 
 void
-vlib_stats_pop_heap (void *cm_arg, void *oldheap, stat_directory_type_t type)
+vlib_stats_pop_heap (void *cm_arg, void *oldheap, u32 cindex,
+		     stat_directory_type_t type)
 {
   vlib_simple_counter_main_t *cm = (vlib_simple_counter_main_t *) cm_arg;
   stat_segment_main_t *sm = &stat_segment_main;
@@ -126,8 +128,17 @@ vlib_stats_pop_heap (void *cm_arg, void *oldheap, stat_directory_type_t type)
   /* Update the 2nd dimension offset vector */
   int i;
   vec_validate (offset_vector, vec_len (cm->counters) - 1);
-  for (i = 0; i < vec_len (cm->counters); i++)
-    offset_vector[i] = stat_segment_offset (shared_header, cm->counters[i]);
+
+  if (sm->last != offset_vector)
+    {
+      for (i = 0; i < vec_len (cm->counters); i++)
+	offset_vector[i] =
+	  stat_segment_offset (shared_header, cm->counters[i]);
+    }
+  else
+    offset_vector[cindex] =
+      stat_segment_offset (shared_header, cm->counters[cindex]);
+
   ep->offset_vector = stat_segment_offset (shared_header, offset_vector);
   sm->directory_vector[vector_index].offset =
     stat_segment_offset (shared_header, cm->counters);
@@ -659,7 +670,7 @@ stat_segment_register_gauge (u8 * name, stat_segment_update_fn update_fn,
 
   ASSERT (shared_header);
 
-  oldheap = vlib_stats_push_heap ();
+  oldheap = vlib_stats_push_heap (NULL);
   vlib_stat_segment_lock ();
 
   memset (&e, 0, sizeof (e));
@@ -719,7 +730,7 @@ statseg_sw_interface_add_del (vnet_main_t * vnm, u32 sw_if_index, u32 is_add)
   stat_segment_main_t *sm = &stat_segment_main;
   stat_segment_shared_header_t *shared_header = sm->shared_header;
 
-  void *oldheap = vlib_stats_push_heap ();
+  void *oldheap = vlib_stats_push_heap (sm->interfaces);
   vlib_stat_segment_lock ();
 
   vec_validate (sm->interfaces, sw_if_index);
@@ -756,11 +767,22 @@ statseg_sw_interface_add_del (vnet_main_t * vnm, u32 sw_if_index, u32 is_add)
 					      ep->offset_vector) : 0;
 
   vec_validate (offset_vector, vec_len (sm->interfaces) - 1);
-  for (i = 0; i < vec_len (sm->interfaces); i++)
+
+  if (sm->last != sm->interfaces)
+    {
+      /* the interface vector moved, so need to recalulate the offset array */
+      for (i = 0; i < vec_len (sm->interfaces); i++)
+	{
+	  offset_vector[i] =
+	    sm->interfaces[i] ? stat_segment_offset (shared_header,
+						     sm->interfaces[i]) : 0;
+	}
+    }
+  else
     {
-      offset_vector[i] =
-	sm->interfaces[i] ? stat_segment_offset (shared_header,
-						 sm->interfaces[i]) : 0;
+      offset_vector[sw_if_index] =
+	sm->interfaces[sw_if_index] ?
+	stat_segment_offset (shared_header, sm->interfaces[sw_if_index]) : 0;
     }
   ep->offset_vector = stat_segment_offset (shared_header, offset_vector);
 
diff --git a/src/vpp/stats/stat_segment.h b/src/vpp/stats/stat_segment.h
index 6e75e6338d4..4167695a94c 100644
--- a/src/vpp/stats/stat_segment.h
+++ b/src/vpp/stats/stat_segment.h
@@ -113,6 +113,7 @@ typedef struct
   u8 *socket_name;
   ssize_t memory_size;
   u8 node_counters_enabled;
+  void *last;
   void *heap;
   stat_segment_shared_header_t *shared_header;	/* pointer to shared memory segment */
   int memfd;