replication: fix insane lag after reconnect

Applier must not update the replication lag upon receiving a packet
containing an error, because those packets don't have timestamps.

Closes #2965

src/box/applier.cc

@@ -421,8 +421,6 @@ applier_subscribe(struct applier *applier)
 		}
 
 		coio_read_xrow(coio, &iobuf->in, &row);
-		applier->lag = ev_now(loop()) - row.tm;
-		applier->last_row_time = ev_monotonic_now(loop());
 
 		if (iproto_type_is_error(row.type))
 			xrow_decode_error_xc(&row);  /* error */
@@ -437,6 +435,10 @@ applier_subscribe(struct applier *applier)
 				  int2str(row.replica_id),
 				  tt_uuid_str(&REPLICASET_UUID));
 		}
+
+		applier->lag = ev_now(loop()) - row.tm;
+		applier->last_row_time = ev_monotonic_now(loop());
+
 		if (vclock_get(&replicaset_vclock, row.replica_id) < row.lsn) {
 			/**
 			 * Promote the replica set vclock before

test/replication/errinj.result

@@ -235,6 +235,10 @@ box.info.replication[1].upstream.status
 ---
 - follow
 ...
+box.info.replication[1].upstream.lag
+---
+- 0
+...
 -- wait for ack timeout
 while box.info.replication[1].upstream.message ~= 'timed out' do fiber.sleep(0.0001) end
 ---

test/replication/errinj.test.lua

@@ -102,6 +102,7 @@ test_run:cmd("switch replica")
 -- wait for reconnect
 while box.info.replication[1].upstream.status ~= 'follow' do fiber.sleep(0.0001) end
 box.info.replication[1].upstream.status
+box.info.replication[1].upstream.lag
 -- wait for ack timeout
 while box.info.replication[1].upstream.message ~= 'timed out' do fiber.sleep(0.0001) end