More efficient IO loop event execution to avoid long loops

If there are lots of loops in a single thread and only some of the loops
are actually active, the other loops are now kept aside and not checked
until they actually get some timers, events or active sockets.

This should help with extreme loads like 100k tables and protocols.

Also ping and loop pickup mechanism was allowing subtle race
conditions. Now properly handling collisions between loop ping and pickup.

1 files changed, 7 insertions, 1 deletions

diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index cca1283c..122b0c22 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -217,6 +217,8 @@ bgp_open(struct bgp_proto *p)
   req->port = p->cf->local_port;
   req->flags = p->cf->free_bind ? SKF_FREEBIND : 0;
 
+  BGP_TRACE(D_EVENTS, "Requesting listen socket at %I%J port %u", req->addr, req->iface, req->port);
+
   add_tail(&bgp_listen_pending, &req->n);
   ev_schedule(&bgp_listen_event);
 }
@@ -243,7 +245,9 @@ bgp_listen_create(void *_ UNUSED)
 	break;
 
     /* Not found any */
-    if (!NODE_VALID(bs))
+    if (NODE_VALID(bs))
+      BGP_TRACE(D_EVENTS, "Found a listening socket: %p", bs);
+    else
     {
       sock *sk = sk_new(proto_pool);
       sk->type = SK_TCP_PASSIVE;
@@ -275,6 +279,8 @@ bgp_listen_create(void *_ UNUSED)
 
       init_list(&bs->requests);
       add_tail(&bgp_sockets, &bs->n);
+
+      BGP_TRACE(D_EVENTS, "Created new listening socket: %p", bs);
     }
 
     add_tail(&bs->requests, &req->n);