lightningd: add slow_reconnect flag for transient failure.

We normally reconnect after 1 second: have a flag to say wait for
60.  This will be used in the next patch which handles "soft" errors.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>


Header from folded patch 'channel_fail_transient_slowretry.patch':

fixup! lightningd: add slow_reconnect flag for transient failure.

@ZmnSCPxj points out that function is unsafe, since omitting the bool
parameter still compiled.  Make it two separate functions, each
with a distinctive name so every caller has to be fixed.

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

lightningd/channel.c

@@ -420,17 +420,12 @@ void channel_set_billboard(struct channel *channel, bool perm, const char *str)
 	}
 }
 
-void channel_fail_transient(struct channel *channel, const char *fmt, ...)
+static void err_and_reconnect(struct channel *channel,
+			      const char *why,
+			      u32 seconds_before_reconnect)
 {
-	va_list ap;
-	const char *why;
-
-	va_start(ap, fmt);
-	why = tal_vfmt(channel, fmt, ap);
-	va_end(ap);
 	log_info(channel->log, "Peer transient failure in %s: %s",
 		 channel_state_name(channel), why);
-	tal_free(why);
 
 #if DEVELOPER
 	if (dev_disconnect_permanent(channel->peer->ld)) {
@@ -441,7 +436,24 @@ void channel_fail_transient(struct channel *channel, const char *fmt, ...)
 
 	channel_set_owner(channel, NULL);
 
-	/* Reconnect after 1 second: prevents some spurious reconnects
+	delay_then_reconnect(channel, seconds_before_reconnect,
-	 * during tests. */
+			     &channel->peer->addr);
-	delay_then_reconnect(channel, 1, &channel->peer->addr);
+}
+
+void channel_fail_reconnect_later(struct channel *channel, const char *fmt, ...)
+{
+	va_list ap;
+
+	va_start(ap, fmt);
+	err_and_reconnect(channel, tal_vfmt(tmpctx, fmt, ap), 60);
+	va_end(ap);
+}
+
+void channel_fail_reconnect(struct channel *channel, const char *fmt, ...)
+{
+	va_list ap;
+
+	va_start(ap, fmt);
+	err_and_reconnect(channel, tal_vfmt(tmpctx, fmt, ap), 1);
+	va_end(ap);
 }

lightningd/channel.h

@@ -175,8 +175,12 @@ const char *channel_state_str(enum channel_state state);
 void channel_set_owner(struct channel *channel, struct subd *owner);
 
 /* Channel has failed, but can try again. */
-PRINTF_FMT(2,3) void channel_fail_transient(struct channel *channel,
+PRINTF_FMT(2,3) void channel_fail_reconnect(struct channel *channel,
-					    const char *fmt,...);
+					      const char *fmt, ...);
+/* Channel has failed, but can try again after a minute. */
+PRINTF_FMT(2,3) void channel_fail_reconnect_later(struct channel *channel,
+						  const char *fmt,...);
+
 /* Channel has failed, give up on it. */
 void channel_fail_permanent(struct channel *channel, const char *fmt, ...);
 /* Permanent error, but due to internal problems, not peer. */

lightningd/channel_control.c

@@ -317,7 +317,8 @@ void peer_start_channeld(struct channel *channel,
 	if (!channel->owner) {
 		log_unusual(channel->log, "Could not subdaemon channel: %s",
 			    strerror(errno));
-		channel_fail_transient(channel, "Failed to subdaemon channel");
+		channel_fail_reconnect_later(channel,
+					     "Failed to subdaemon channel");
 		return;
 	}
 

lightningd/closing_control.c

@@ -191,7 +191,8 @@ void peer_start_closingd(struct channel *channel,
 	if (!channel->owner) {
 		log_unusual(channel->log, "Could not subdaemon closing: %s",
 			    strerror(errno));
-		channel_fail_transient(channel, "Failed to subdaemon closing");
+		channel_fail_reconnect_later(channel,
+					     "Failed to subdaemon closing");
 		return;
 	}
 

lightningd/connect_control.c

@@ -278,7 +278,7 @@ static void peer_please_disconnect(struct lightningd *ld, const u8 *msg)
 	if (uc)
 		kill_uncommitted_channel(uc, "Reconnected");
 	else if (c)
-		channel_fail_transient(c, "Reconnected");
+		channel_fail_reconnect(c, "Reconnected");
 }
 
 static unsigned connectd_msg(struct subd *connectd, const u8 *msg, const int *fds)

lightningd/peer_control.c

@@ -396,7 +396,7 @@ void channel_errmsg(struct channel *channel,
 
 	/* No per_peer_state means a subd crash or disconnection. */
 	if (!pps) {
-		channel_fail_transient(channel, "%s: %s",
+		channel_fail_reconnect(channel, "%s: %s",
 				       channel->owner->name, desc);
 		return;
 	}
@@ -1022,10 +1022,10 @@ static enum watch_result funding_depth_cb(struct lightningd *ld,
 		} else if (!short_channel_id_eq(channel->scid, &scid)) {
 			/* This normally restarts channeld, initialized with updated scid
 			 * and also adds it (at least our halve_chan) to rtable. */
-			channel_fail_transient(channel,
+			channel_fail_reconnect(channel,
-					        "short_channel_id changed to %s (was %s)",
+					       "short_channel_id changed to %s (was %s)",
-					        short_channel_id_to_str(tmpctx, &scid),
+					       short_channel_id_to_str(tmpctx, &scid),
-					        short_channel_id_to_str(tmpctx, channel->scid));
+					       short_channel_id_to_str(tmpctx, channel->scid));
 
 			*channel->scid = scid;
 			wallet_channel_save(ld->wallet, channel);
@@ -1373,7 +1373,7 @@ static struct command_result *json_disconnect(struct command *cmd,
 	channel = peer_active_channel(peer);
 	if (channel) {
 		if (*force) {
-			channel_fail_transient(channel,
+			channel_fail_reconnect(channel,
 					       "disconnect command force=true");
 			return command_success(cmd, json_stream_success(cmd));
 		}