Baker: retry once opening node's context on failed checkouts

src/proto_alpha/lib_delegate/client_baking_forge.ml

@@ -21,10 +21,10 @@ let votes_index = 1
 let anonymous_index = 2
 let managers_index = 3
 
-
 type state = {
   genesis: Block_hash.t ;
-  index : Context.index ;
+  context_path: string ;
+  mutable index : Context.index ;
 
   (* see [get_delegates] below to find delegates when the list is empty *)
   delegates: public_key_hash list ;
@@ -38,8 +38,9 @@ type state = {
     (Time.t * (Client_baking_blocks.block_info * int * public_key_hash)) list ;
 }
 
-let create_state genesis index delegates constants best =
+let create_state genesis context_path index delegates constants best =
   { genesis ;
+    context_path ;
     index ;
     delegates ;
     constants ;
@@ -312,7 +313,6 @@ let forge_block cctxt ?(chain = `Main) block
   decode_priority cctxt chain block priority >>=? fun (priority, minimal_timestamp) ->
   unopt_timestamp timestamp minimal_timestamp >>=? fun timestamp ->
 
-
   (* get basic building blocks *)
   let protocol_data = forge_faked_protocol_data ~priority ~seed_nonce_hash in
   classify_operations ?threshold operations_arg >>=? fun operations ->
@@ -449,8 +449,6 @@ let safe_get_unrevealed_nonces cctxt block =
       lwt_warn "Cannot read nonces: %a@." pp_print_error err >>= fun () ->
       Lwt.return []
 
-
-
 let insert_block
     ?max_priority
     ()
@@ -497,12 +495,21 @@ let pop_baking_slots state =
   state.future_slots <- future_slots ;
   slots
 
-
 let filter_invalid_operations (cctxt : #full) state block_info (operations : packed_operation list list) =
   let open Client_baking_simulator in
   lwt_debug "Starting client-side validation %a"
     Block_hash.pp block_info.Client_baking_blocks.hash >>= fun () ->
-  begin_construction cctxt state.index block_info >>=? fun initial_inc ->
+  begin begin_construction cctxt state.index block_info >>= function
+    | Ok inc -> return inc
+    | Error errs ->
+        lwt_log_error "Error while fetching current context : %a"
+          pp_print_error errs >>= fun () ->
+        lwt_log_notice "Retrying to open the context" >>= fun () ->
+        Client_baking_simulator.load_context ~context_path:state.context_path >>= fun index ->
+        begin_construction cctxt index block_info >>=? fun inc ->
+        state.index <- index;
+        return inc
+  end  >>=? fun initial_inc ->
   let endorsements = List.nth operations endorsements_index in
   let votes = List.nth operations votes_index in
   let anonymous = List.nth operations anonymous_index in
@@ -742,7 +749,7 @@ let create
     let constants =
       tzlazy (fun () -> Alpha_services.Constants.all cctxt (`Main, `Head 0)) in
     Client_baking_simulator.load_context ~context_path >>= fun index ->
-    let state = create_state genesis_hash index delegates constants bi in
+    let state = create_state genesis_hash context_path index delegates constants bi in
     return state
   in
 
@@ -755,4 +762,3 @@ let create
     ~compute_timeout
     ~timeout_k:(bake ?threshold ())
     ~event_k:(insert_block ?max_priority ())
-

src/proto_alpha/lib_delegate/client_baking_simulator.ml

@@ -12,6 +12,22 @@ open Alpha_context
 
 module Main = Alpha_environment.Lift(Main)
 
+type error +=
+  | Failed_to_checkout_context
+
+let () =
+  register_error_kind
+    `Permanent
+    ~id:"Client_baking_simulator.failed_to_checkout_context"
+    ~title: "Fail during checkout context"
+    ~description: ""
+    ~pp:(fun ppf () -> Format.fprintf ppf "@[Failed to checkout the context@]")
+    Data_encoding.unit
+    (function
+      | Failed_to_checkout_context -> Some ()
+      | _ -> None)
+    (fun () -> Failed_to_checkout_context)
+
 type incremental = {
   predecessor: Client_baking_blocks.block_info ;
   context : Context.t ;
@@ -25,34 +41,36 @@ let load_context ~context_path =
 
 let begin_construction (_cctxt : #Proto_alpha.full) index predecessor =
   let { Client_baking_blocks.context } = predecessor in
-  Context.checkout_exn index context >>= fun context ->
+  Context.checkout index context >>= function
-  let timestamp = Time.now () in
+  | None -> fail Failed_to_checkout_context
-  let predecessor_hash = predecessor.hash in
+  | Some context ->
-  let header : Tezos_base.Block_header.shell_header = Tezos_base.Block_header.{
+      let timestamp = Time.now () in
-      predecessor = predecessor_hash ;
+      let predecessor_hash = predecessor.hash in
-      proto_level = 0 ;
+      let header : Tezos_base.Block_header.shell_header = Tezos_base.Block_header.{
-      validation_passes = 0 ;
+          predecessor = predecessor_hash ;
-      fitness = predecessor.fitness ;
+          proto_level = 0 ;
-      timestamp ;
+          validation_passes = 0 ;
-      level = Raw_level.to_int32 predecessor.level ;
+          fitness = predecessor.fitness ;
-      context = Context_hash.zero ;
+          timestamp ;
-      operations_hash = Operation_list_list_hash.zero ;
+          level = Raw_level.to_int32 predecessor.level ;
-    } in
+          context = Context_hash.zero ;
-  Main.begin_construction
+          operations_hash = Operation_list_list_hash.zero ;
-    ~predecessor_context: context
+        } in
-    ~predecessor_timestamp: header.timestamp
+      Main.begin_construction
-    ~predecessor_fitness: header.fitness
+        ~predecessor_context: context
-    ~predecessor_level: header.level
+        ~predecessor_timestamp: header.timestamp
-    ~predecessor:predecessor_hash
+        ~predecessor_fitness: header.fitness
-    ~timestamp
+        ~predecessor_level: header.level
-    () >>=? fun state ->
+        ~predecessor:predecessor_hash
-  return {
+        ~timestamp
-    predecessor ;
+        () >>=? fun state ->
-    context ;
+      return {
-    state ;
+        predecessor ;
-    rev_operations = [] ;
+        context ;
-    header ;
+        state ;
-  }
+        rev_operations = [] ;
+        header ;
+      }
 
 let add_operation st ( op : Operation.packed ) =
   Main.apply_operation st.state op >>=? fun (state, _) ->