diff --git a/connectd/connectd.c b/connectd/connectd.c
index f66abd13f7bb..5232b6fe74ff 100644
--- a/connectd/connectd.c
+++ b/connectd/connectd.c
@@ -208,7 +208,7 @@ void destroy_peer(struct peer *peer)
 {
 	assert(!peer->draining);
 
-	if (!peer_htable_del(&peer->daemon->peers, peer))
+	if (!peer_htable_del(peer->daemon->peers, peer))
 		abort();
 
 	/* Tell gossipd to stop asking this peer gossip queries */
@@ -257,7 +257,7 @@ static struct peer *new_peer(struct daemon *daemon,
 
 	/* Now we own it */
 	tal_steal(peer, peer->to_peer);
-	peer_htable_add(&daemon->peers, peer);
+	peer_htable_add(daemon->peers, peer);
 	tal_add_destructor(peer, destroy_peer);
 
 	return peer;
@@ -282,7 +282,7 @@ struct io_plan *peer_connected(struct io_conn *conn,
 	bool option_gossip_queries;
 
 	/* We remove any previous connection immediately, on the assumption it's dead */
-	peer = peer_htable_get(&daemon->peers, id);
+	peer = peer_htable_get(daemon->peers, id);
 	if (peer)
 		tal_free(peer);
 
@@ -1724,7 +1724,7 @@ static void try_connect_peer(struct daemon *daemon,
 	struct connecting *connect;
 
 	/* Already existing?  Must have crossed over, it'll know soon. */
-	if (peer_htable_get(&daemon->peers, id))
+	if (peer_htable_get(daemon->peers, id))
 		return;
 
 	/* If we're trying to connect it right now, that's OK. */
@@ -1827,7 +1827,7 @@ static void peer_discard(struct daemon *daemon, const u8 *msg)
 
 	/* We should stay in sync with lightningd, but this can happen
 	 * under stress. */
-	peer = peer_htable_get(&daemon->peers, &id);
+	peer = peer_htable_get(daemon->peers, &id);
 	if (!peer)
 		return;
 	/* If it's reconnected already, it will learn soon. */
@@ -1852,7 +1852,7 @@ static void peer_final_msg(struct io_conn *conn,
 
 	/* This can happen if peer hung up on us (or wrong counter
 	 * if it reconnected). */
-	peer = peer_htable_get(&daemon->peers, &id);
+	peer = peer_htable_get(daemon->peers, &id);
 	if (peer && peer->counter == counter)
 		multiplex_final_msg(peer, take(finalmsg));
 }
@@ -1868,7 +1868,7 @@ static void dev_connect_memleak(struct daemon *daemon, const u8 *msg)
 
 	/* Now delete daemon and those which it has pointers to. */
 	memleak_scan_obj(memtable, daemon);
-	memleak_scan_htable(memtable, &daemon->peers.raw);
+	memleak_scan_htable(memtable, &daemon->peers->raw);
 
 	found_leak = dump_memleak(memtable, memleak_status_broken);
 	daemon_conn_send(daemon->master,
@@ -1995,7 +1995,7 @@ static struct io_plan *recv_gossip(struct io_conn *conn,
 		status_failed(STATUS_FAIL_GOSSIP_IO, "Unknown msg %i",
 			      fromwire_peektype(msg));
 
-	peer = peer_htable_get(&daemon->peers, &dst);
+	peer = peer_htable_get(daemon->peers, &dst);
 	if (peer)
 		inject_peer_msg(peer, take(gossip_msg));
 
@@ -2007,7 +2007,7 @@ static struct io_plan *recv_gossip(struct io_conn *conn,
 #if DEVELOPER
 static void memleak_daemon_cb(struct htable *memtable, struct daemon *daemon)
 {
-	memleak_scan_htable(memtable, &daemon->peers.raw);
+	memleak_scan_htable(memtable, &daemon->peers->raw);
 }
 #endif /* DEVELOPER */
 
@@ -2023,7 +2023,8 @@ int main(int argc, char *argv[])
 	/* Allocate and set up our simple top-level structure. */
 	daemon = tal(NULL, struct daemon);
 	daemon->connection_counter = 1;
-	peer_htable_init(&daemon->peers);
+	daemon->peers = tal(daemon, struct peer_htable);
+	peer_htable_init(daemon->peers);
 	memleak_add_helper(daemon, memleak_daemon_cb);
 	list_head_init(&daemon->connecting);
 	timers_init(&daemon->timers, time_mono());
diff --git a/connectd/connectd.h b/connectd/connectd.h
index 61555e3dc506..c940887451d4 100644
--- a/connectd/connectd.h
+++ b/connectd/connectd.h
@@ -142,7 +142,7 @@ struct daemon {
 
 	/* Peers that we've handed to `lightningd`, which it hasn't told us
 	 * have disconnected. */
-	struct peer_htable peers;
+	struct peer_htable *peers;
 
 	/* Peers we are trying to reach */
 	struct list_head connecting;
diff --git a/connectd/multiplex.c b/connectd/multiplex.c
index 0758513c5e78..ecfcffe538f5 100644
--- a/connectd/multiplex.c
+++ b/connectd/multiplex.c
@@ -592,7 +592,7 @@ void send_custommsg(struct daemon *daemon, const u8 *msg)
 		master_badmsg(WIRE_CONNECTD_CUSTOMMSG_OUT, msg);
 
 	/* Races can happen: this might be gone by now. */
-	peer = peer_htable_get(&daemon->peers, &id);
+	peer = peer_htable_get(daemon->peers, &id);
 	if (peer)
 		inject_peer_msg(peer, take(custommsg));
 }
@@ -1242,7 +1242,7 @@ void peer_connect_subd(struct daemon *daemon, const u8 *msg, int fd)
 		master_badmsg(WIRE_CONNECTD_PEER_CONNECT_SUBD, msg);
 
 	/* Races can happen: this might be gone by now (or reconnected!). */
-	peer = peer_htable_get(&daemon->peers, &id);
+	peer = peer_htable_get(daemon->peers, &id);
 	if (!peer || peer->counter != counter) {
 		close(fd);
 		return;
@@ -1276,7 +1276,7 @@ void send_manual_ping(struct daemon *daemon, const u8 *msg)
 	if (!fromwire_connectd_ping(msg, &id, &num_pong_bytes, &len))
 		master_badmsg(WIRE_CONNECTD_PING, msg);
 
-	peer = peer_htable_get(&daemon->peers, &id);
+	peer = peer_htable_get(daemon->peers, &id);
 	if (!peer) {
 		daemon_conn_send(daemon->master,
 				 take(towire_connectd_ping_reply(NULL,
diff --git a/connectd/onion_message.c b/connectd/onion_message.c
index a20119deaafb..85c7c44b959a 100644
--- a/connectd/onion_message.c
+++ b/connectd/onion_message.c
@@ -29,7 +29,7 @@ void onionmsg_req(struct daemon *daemon, const u8 *msg)
 
 	/* Even though lightningd checks for valid ids, there's a race
 	 * where it might vanish before we read this command. */
-	peer = peer_htable_get(&daemon->peers, &id);
+	peer = peer_htable_get(daemon->peers, &id);
 	if (peer) {
 		u8 *omsg = towire_onion_message(NULL, &blinding, onionmsg);
 		inject_peer_msg(peer, take(omsg));
@@ -86,7 +86,7 @@ void handle_onion_message(struct daemon *daemon,
 
 		/* FIXME: Handle short_channel_id! */
 		node_id_from_pubkey(&next_node_id, &next_node);
-		next_peer = peer_htable_get(&daemon->peers, &next_node_id);
+		next_peer = peer_htable_get(daemon->peers, &next_node_id);
 		if (!next_peer) {
 			status_peer_debug(&peer->id,
 					  "onion msg: unknown next peer %s",