Table cork: Stop creating updates when there are too many pending.

The corked procedure gets a callback when uncorked. Supported by table
maintenance routines and also BGP.

3 files changed, 29 insertions, 1 deletions

diff --git a/nest/route.h b/nest/route.h
index 88a56073..310cea92 100644
--- a/nest/route.h
+++ b/nest/route.h
@@ -157,6 +157,7 @@ struct rtable_config {
   btime min_settle_time;		/* Minimum settle time for notifications */
   btime max_settle_time;		/* Maximum settle time for notifications */
   btime export_settle_time;		/* Delay before exports are announced */
+  uint cork_limit;			/* Amount of routes to be pending on export to cork imports */
 };
 
 typedef struct rtable {
@@ -187,6 +188,9 @@ typedef struct rtable {
   int gc_counter;			/* Number of operations since last GC */
   byte prune_state;			/* Table prune state, 1 -> scheduled, 2-> running */
   byte nhu_state;			/* Next Hop Update state */
+
+  byte cork_active;			/* Congestion control activated */
+
   struct fib_iterator prune_fit;	/* Rtable prune FIB iterator */
   struct fib_iterator nhu_fit;		/* Next Hop Update FIB iterator */
   struct tbf rl_pipe;			/* Rate limiting token buffer for pipe collisions */
@@ -378,6 +382,8 @@ struct rt_export_hook {
   void (*stopped)(struct rt_export_request *);	/* Stored callback when export is stopped */
 };
 
+extern struct event_cork rt_cork;
+
 #define TIS_DOWN	0
 #define TIS_UP		1
 #define TIS_STOP	2
diff --git a/nest/rt-table.c b/nest/rt-table.c
index 6dd6587e..b4cd0448 100644
--- a/nest/rt-table.c
+++ b/nest/rt-table.c
@@ -110,6 +110,7 @@ const char *rt_export_state_name(u8 state)
     return rt_export_state_name_array[state];
 }
 
+struct event_cork rt_cork;
 
 /* Like fib_route(), but skips empty net entries */
 static inline void *
@@ -1061,7 +1062,13 @@ rte_announce(rtable *tab, net *net, struct rte_storage *new, struct rte_storage
   if (tab->first_export == NULL)
     tab->first_export = rpe;
 
-  if (!tm_active(tab->export_timer))
+  if ((tab->first_export->seq + tab->config->cork_limit <= tab->next_export_seq) && !tab->cork_active)
+  {
+    ev_cork(&rt_cork);
+    tab->cork_active = 1;
+    tm_start(tab->export_timer, 0);
+  }
+  else if (!tm_active(tab->export_timer))
     tm_start(tab->export_timer, tab->config->export_settle_time);
 }
 
@@ -2042,6 +2049,7 @@ rt_free(resource *_r)
   DBG("Deleting routing table %s\n", r->name);
   ASSERT_DIE(r->use_count == 0);
   ASSERT_DIE(r->rt_count == 0);
+  ASSERT_DIE(!r->cork_active);
   ASSERT_DIE(EMPTY_LIST(r->imports));
   ASSERT_DIE(EMPTY_LIST(r->exports));
 
@@ -2106,6 +2114,9 @@ rt_setup(pool *pp, struct rtable_config *cf)
   t->hcu_event = ev_new_init(p, rt_update_hostcache, t);
   t->nhu_event = ev_new_init(p, rt_next_hop_update, t);
 
+  t->nhu_event->cork = &rt_cork;
+  t->prune_event->cork = &rt_cork;
+
   t->export_timer = tm_new_init(p, rt_announce_exports, t, 0, 0);
   t->last_rt_change = t->gc_time = current_time();
   t->next_export_seq = 1;
@@ -2128,6 +2139,7 @@ rt_init(void)
   rt_table_pool = rp_new(&root_pool, "Routing tables");
   rte_update_pool = lp_new_default(rt_table_pool);
   init_list(&routing_tables);
+  ev_init_cork(&rt_cork, "Route Table Cork");
 }
 
 /**
@@ -2420,6 +2432,14 @@ done:;
 
   if (EMPTY_LIST(tab->pending_exports) && tm_active(tab->export_timer))
     tm_stop(tab->export_timer);
+
+  /* If reduced to at most one export block pending */
+  if (tab->cork_active &&
+      ((!tab->first_export) || (tab->first_export->seq + 128 > tab->next_export_seq)))
+  {
+    tab->cork_active = 0;
+    ev_uncork(&rt_cork);
+  }
 }
 
 void
@@ -2699,6 +2719,7 @@ rt_new_table(struct symbol *s, uint addr_type)
   c->gc_min_time = 5;
   c->min_settle_time = 1 S;
   c->max_settle_time = 20 S;
+  c->cork_limit = 4 * page_size / sizeof(struct rt_pending_export);
   c->config = new_config;
 
   add_tail(&new_config->tables, &c->n);
diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index 96df671f..5f2e7dfd 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -584,6 +584,7 @@ bgp_conn_enter_established_state(struct bgp_conn *conn)
     p->link_addr = p->neigh->iface->llv6->ip;
 
   conn->sk->fast_rx = 0;
+  conn->sk->cork = &rt_cork;
 
   p->conn = conn;
   p->last_error_class = 0;