Implement reconnect timer including backoff.

This allows us to enable the last test case for compat_reqrep.

3 files changed, 29 insertions, 17 deletions

diff --git a/src/core/endpt.c b/src/core/endpt.c
index 5a642a2f..9aae89eb 100644
--- a/src/core/endpt.c
+++ b/src/core/endpt.c
@@ -161,11 +161,31 @@ nni_dialer(void *arg)
 	nni_ep *ep = arg;
 	int rv;
 	nni_time cooldown;
+	nni_duration maxrtime;
+	nni_duration defrtime;
+	nni_duration rtime;
 	nni_mtx *mx = &ep->ep_sock->s_mx;
 
+	nni_mtx_lock(mx);
+	defrtime = ep->ep_sock->s_reconn;
+	if ((maxrtime = ep->ep_sock->s_reconnmax) == 0) {
+		maxrtime = defrtime;
+	}
+	nni_mtx_unlock(mx);
+
 	for (;;) {
 		nni_mtx_lock(mx);
+		if ((defrtime != ep->ep_sock->s_reconn) ||
+		    (maxrtime != ep->ep_sock->s_reconnmax)) {
+			// Times changed, so reset them.
+			defrtime = ep->ep_sock->s_reconn;
+			if ((maxrtime = ep->ep_sock->s_reconnmax) == 0) {
+				maxrtime = defrtime;
+			}
+			rtime = defrtime;
+		}
 		while ((!ep->ep_close) && (ep->ep_pipe != NULL)) {
+			rtime = defrtime;
 			nni_cv_wait(&ep->ep_cv);
 		}
 		if (ep->ep_close) {
@@ -179,22 +199,19 @@ nni_dialer(void *arg)
 		case 0:
 			// good connection
 			continue;
-		case NNG_ENOMEM:
-			cooldown = 1000000;
-			break;
 		case NNG_ECLOSED:
 			return;
 
 		default:
-			// XXX: THIS NEEDS TO BE A PROPER BACKOFF.
-			cooldown = 1000000;
+			cooldown = nni_clock() + rtime;
+			rtime *= 2;
+			if ((maxrtime >= defrtime) && (rtime > maxrtime)) {
+				rtime = maxrtime;
+			}
 			break;
 		}
 		// we inject a delay so we don't just spin hard on
-		// errors like connection refused.  For NNG_ENOMEM, we
-		// wait even longer, since the system needs time to
-		// release resources.
-		cooldown += nni_clock();
+		// errors like connection refused.
 		nni_mtx_lock(mx);
 		while (!ep->ep_close) {
 			// We need a different condvar...
diff --git a/src/core/socket.c b/src/core/socket.c
index 24078941..b0ce172f 100644
--- a/src/core/socket.c
+++ b/src/core/socket.c
@@ -274,7 +274,7 @@ nni_sock_open(nni_sock **sockp, uint16_t pnum)
 	sock->s_rcvtimeo = -1;
 	sock->s_closing = 0;
 	sock->s_reconn = NNI_SECOND;
-	sock->s_reconnmax = NNI_SECOND;
+	sock->s_reconnmax = 0;
 	sock->s_reapexit = 0;
 	NNI_LIST_INIT(&sock->s_pipes, nni_pipe, p_node);
 	NNI_LIST_INIT(&sock->s_reaps, nni_pipe, p_node);
diff --git a/tests/compat_reqrep.c b/tests/compat_reqrep.c
index a418b8d9..51ff4fb8 100644
--- a/tests/compat_reqrep.c
+++ b/tests/compat_reqrep.c
@@ -121,29 +121,24 @@ int main ()
     test_close (rep1);
 
 
-#if 0 // The default reconnect interval is waaay to large for this to pass.
     /*  Check sending a request when the peer is not available. (It should
         be sent immediatelly when the peer comes online rather than relying
         on the resend algorithm. */
     req1 = test_socket (AF_SP, NN_REQ);
     timeo = 10;
-    rc = nn_setsockopt (req1, NN_SOL_SOCKET, NN_RECONNECT_IVL,
-        &timeo, sizeof (timeo));
+    test_setsockopt (req1, NN_SOL_SOCKET, NN_RECONNECT_IVL, &timeo, sizeof (timeo));
     test_send (req1, "ABC");
     test_connect (req1, SOCKET_ADDRESS);
 
     rep1 = test_socket (AF_SP, NN_REP);
     test_bind (rep1, SOCKET_ADDRESS);
     timeo = 500;
-    rc = nn_setsockopt (rep1, NN_SOL_SOCKET, NN_RCVTIMEO,
-       &timeo, sizeof (timeo));
-    printf("RC = %d errno %d (%s)\n", rc, errno, strerror(errno));
+    test_setsockopt (rep1, NN_SOL_SOCKET, NN_RCVTIMEO, &timeo, sizeof (timeo));
     errno_assert (rc == 0);
     test_recv (rep1, "ABC");
 
     test_close (req1);
     test_close (rep1);
-#endif
 
     /*  Check removing socket request sent to (It should
         be sent immediatelly to other peer rather than relying