From b7c2546e5b220e016d1f99791397f56033682cd9 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Gr=C3=A9goire=20Henry?= <greg@hnr.gr>
Date: Tue, 5 Mar 2019 17:29:09 +0100
Subject: [PATCH] Daemon: `await_bootstrapped_node` now retries on
 `Connection_failed`

This allows the baker/endorser/accuser to wait a litlle bit for the
node to initialize itself and to open its RPC port.

By default, the deamon retries 6 times with a total waiting time
around 20 seconds, before to fail.
---
 src/proto_alpha/lib_delegate/client_daemon.ml | 24 ++++++++++++++++++-
 src/proto_alpha/lib_delegate/dune             |  4 +++-
 2 files changed, 26 insertions(+), 2 deletions(-)

diff --git a/src/proto_alpha/lib_delegate/client_daemon.ml b/src/proto_alpha/lib_delegate/client_daemon.ml
index 54ed9c3ee..cf17838c6 100644
--- a/src/proto_alpha/lib_delegate/client_daemon.ml
+++ b/src/proto_alpha/lib_delegate/client_daemon.ml
@@ -23,11 +23,33 @@
 (*                                                                           *)
 (*****************************************************************************)
 
+let rec retry (cctxt: #Proto_alpha.full) ~delay ~tries f x =
+  f x >>= function
+  | Ok _ as r -> Lwt.return r
+  | Error (RPC_client.Request_failed
+             { error = Connection_failed _ ; _ } :: _) as err
+    when tries > 0 -> begin
+      cctxt#message
+        "Connection refused, retrying in %.2f seconds..."
+        delay >>= fun () ->
+      Lwt.pick
+        [ (Lwt_unix.sleep delay >|= fun () -> `Continue) ;
+          (Lwt_exit.termination_thread >|= fun _ -> `Killed) ;
+        ] >>= function
+      | `Killed ->
+          Lwt.return err
+      | `Continue ->
+          retry cctxt ~delay:(delay *. 1.5) ~tries:(tries - 1) f x
+    end
+  | Error _ as err ->
+      Lwt.return err
+
 let await_bootstrapped_node (cctxt: #Proto_alpha.full) =
   (* Waiting for the node to be synchronized *)
   cctxt#message "Waiting for the node to be synchronized with its \
                  peers..." >>= fun () ->
-  Shell_services.Monitor.bootstrapped cctxt >>=? fun _ ->
+  retry cctxt ~tries:5 ~delay:1.
+    Shell_services.Monitor.bootstrapped cctxt >>=? fun _ ->
   cctxt#message "Node synchronized." >>= fun () ->
   return_unit
 
diff --git a/src/proto_alpha/lib_delegate/dune b/src/proto_alpha/lib_delegate/dune
index 6de77c06d..bb3ed553b 100644
--- a/src/proto_alpha/lib_delegate/dune
+++ b/src/proto_alpha/lib_delegate/dune
@@ -9,6 +9,7 @@
             tezos-client-alpha
             tezos-client-commands
             tezos-storage
+            tezos-rpc-http
             tezos-rpc)
  (library_flags (:standard -linkall))
  (modules (:standard \
@@ -22,7 +23,8 @@
                    -open Tezos_client_alpha
                    -open Tezos_client_commands
                    -open Tezos_storage
-                   -open Tezos_rpc)))
+                   -open Tezos_rpc
+                   -open Tezos_rpc_http)))
 
 (library
  (name tezos_baking_alpha_commands)