diff --git a/lightningd/connect_control.c b/lightningd/connect_control.c
index ceb41d302..761e6d307 100644
--- a/lightningd/connect_control.c
+++ b/lightningd/connect_control.c
@@ -236,6 +236,12 @@ static struct command_result *json_connect(struct command *cmd,
 					   &peer->addr);
 	}
 
+	/* When a peer disconnects, we give subds time to clean themselves up
+	 * (this lets connectd ensure they've seen the final messages).  But
+	 * now it's going to try to reconnect, we've gotta force them out. */
+	if (peer)
+		peer_channels_cleanup(peer);
+
 	subd_send_msg(cmd->ld->connectd,
 		      take(towire_connectd_connect_to_peer(NULL, &id_addr.id, addr, true)));
 
diff --git a/lightningd/peer_control.c b/lightningd/peer_control.c
index 97f50b8b9..56ea80828 100644
--- a/lightningd/peer_control.c
+++ b/lightningd/peer_control.c
@@ -177,16 +177,10 @@ void maybe_delete_peer(struct peer *peer)
 	delete_peer(peer);
 }
 
-static void peer_channels_cleanup(struct lightningd *ld,
-				  const struct node_id *id)
+void peer_channels_cleanup(struct peer *peer)
 {
-	struct peer *peer;
 	struct channel *c, **channels;
 
-	peer = peer_by_id(ld, id);
-	if (!peer)
-		return;
-
 	/* Freeing channels can free peer, so gather first. */
 	channels = tal_arr(tmpctx, struct channel *, 0);
 	list_for_each(&peer->channels, c, list)
@@ -1726,11 +1720,6 @@ void peer_connected(struct lightningd *ld, const u8 *msg)
 		fatal("Connectd gave bad CONNECT_PEER_CONNECTED message %s",
 		      tal_hex(msg, msg));
 
-	/* When a peer disconnects, we give subds time to clean themselves up
-	 * (this lets connectd ensure they've seen the final messages).  But
-	 * now it's reconnected, we've gotta force them out. */
-	peer_channels_cleanup(ld, &id);
-
 	/* If we connected, and it's a normal address */
 	if (!hook_payload->incoming
 	    && hook_payload->addr.itype == ADDR_INTERNAL_WIREADDR
@@ -1743,6 +1732,15 @@ void peer_connected(struct lightningd *ld, const u8 *msg)
 	/* If we're already dealing with this peer, hand off to correct
 	 * subdaemon.  Otherwise, we'll hand to openingd to wait there. */
 	peer = peer_by_id(ld, &id);
+	if (peer) {
+		/* When a peer disconnects, we give subds time to clean themselves up
+		 * (this lets connectd ensure they've seen the final messages).  But
+		 * now it's reconnected, we've gotta force them out.  This might free
+		 * the peer! */
+		peer_channels_cleanup(peer);
+		peer = peer_by_id(ld, &id);
+	}
+
 	if (!peer) {
 		/* If we connected to them, we know this is a good address. */
 		peer = new_peer(ld, 0, &id, &hook_payload->addr,
diff --git a/lightningd/peer_control.h b/lightningd/peer_control.h
index 88a5f8866..84d11f2ca 100644
--- a/lightningd/peer_control.h
+++ b/lightningd/peer_control.h
@@ -165,6 +165,9 @@ command_find_channel(struct command *cmd,
 		     const char *buffer, const jsmntok_t *tok,
 		     struct channel **channel);
 
+/* We do this lazily, when reconnecting */
+void peer_channels_cleanup(struct peer *peer);
+
 /* Ancient (0.7.0 and before) releases could create invalid commitment txs! */
 bool invalid_last_tx(const struct bitcoin_tx *tx);