Merge pull request #25 from Cjen1/improve_restart_perf

Improve the failure recovery properties

.github/workflows/workflow.yml

@@ -1,33 +1,24 @@
-name: Build and runtest
+name: "CI"
 
 on:
-  push:
-    branches:
-      - main
   pull_request:
+  push:
+    branches: [main]
 
 jobs:
-   build:
-     name: Check build
-     runs-on: ubuntu-22.04
-     steps:
-      - uses: actions/checkout@v3
-      - uses: DeterminateSystems/nix-installer-action@main
-      - uses: DeterminateSystems/magic-nix-cache-action@main
-      - run: nix develop -c dune build
-   format:
-     name: Check formatting
-     runs-on: ubuntu-22.04
-     steps:
-      - uses: actions/checkout@v3
-      - uses: DeterminateSystems/nix-installer-action@main
-      - uses: DeterminateSystems/magic-nix-cache-action@main
-      - run: nix develop -c dune build @fmt
-   runtest:
-     name: Run tests
-     runs-on: ubuntu-22.04
-     steps:
-      - uses: actions/checkout@v3
-      - uses: DeterminateSystems/nix-installer-action@main
-      - uses: DeterminateSystems/magic-nix-cache-action@main
-      - run: nix develop -c dune runtest
+  tests:
+    runs-on: ubuntu-latest
+    steps:
+    - uses: actions/checkout@v3
+
+    - name: Install Nix
+      uses: DeterminateSystems/nix-installer-action@v4
+    - name: Run the Magic Nix Cache
+      uses: DeterminateSystems/magic-nix-cache-action@v1
+    - name: Run tests
+      run: |
+        nix develop -c dune runtest
+    - name: Format
+      run: |
+        nix develop -c dune build @fmt
+

bin/cli.ml

@@ -13,7 +13,9 @@ let run op sockaddrs id retry_timeout =
     sockaddrs
     |> List.mapi (fun idx addr ->
            ( idx
-           , fun sw -> (Eio.Net.connect ~sw env#net addr :> Eio.Flow.two_way_ty Eio.Flow.two_way) ) )
+           , fun sw ->
+               ( Eio.Net.connect ~sw env#net addr
+                 :> Eio.Flow.two_way_ty Eio.Flow.two_way ) ) )
   in
   Eio.traceln "Creating conns to: %a"
     Fmt.(braces @@ list ~sep:comma Eio.Net.Sockaddr.pp)

bin/dune

@@ -1,16 +1,19 @@
 (executable
-  (name cli)
-  (modules cli)
-  (libraries ocons.core cmdliner))
+ (name cli)
+ (modules cli)
+ (libraries ocons.core cmdliner))
+
 (executable
-  (name bench)
-  (modules bench)
-  (libraries ocons.core cmdliner))
+ (name bench)
+ (modules bench)
+ (libraries ocons.core cmdliner))
+
 (executable
-  (name echo)
-  (modules echo)
-  (libraries ocons.core cmdliner))
+ (name echo)
+ (modules echo)
+ (libraries ocons.core cmdliner))
+
 (executable
-  (name test_alloc)
-  (modules test_alloc)
-  (libraries core core_bench core_unix.command_unix))
+ (name test_alloc)
+ (modules test_alloc)
+ (libraries core core_bench core_unix.command_unix))

bin/test_alloc.ml

@@ -9,8 +9,7 @@ let alloc_test =
   let test `init =
     let large = make_data 1000 in
     let extern = ref None in
-    fun () -> 
-      extern := Some large
+    fun () -> extern := Some large
   in
   Bench.Test.create_with_initialization ~name:"alloc_with_reference" test
 

impl/bench/dune

@@ -1,6 +1,6 @@
 (executable
-  (name ctree_bench)
-  (modules ctree_bench)
-  (libraries impl_core core_bench.inline_benchmarks)
-  (preprocess (pps ppx_jane ppx_deriving.show))
-  )
+ (name ctree_bench)
+ (modules ctree_bench)
+ (libraries impl_core core_bench.inline_benchmarks)
+ (preprocess
+  (pps ppx_jane ppx_deriving.show)))

impl/lib/conspire_command_tree.ml

@@ -7,19 +7,20 @@ module type Value = sig
 end
 
 module CommandTree (Value : Value) = struct
-  module Key = struct 
+  module Key = struct
     include Md5
-    let pp ppf v =
-      Fmt.pf ppf "%s" (Md5.to_hex v)
+
+    let pp ppf v = Fmt.pf ppf "%s" (Md5.to_hex v)
   end
+
   type key = Key.t [@@deriving show, bin_io, equal, compare]
 
   let make_key =
-    let open struct 
+    let open struct
       type relevant_key_data = key * Value.t [@@deriving bin_io]
     end in
     fun parent_key value ->
-    Md5.digest_bin_prot bin_writer_relevant_key_data (parent_key, value)
+      Md5.digest_bin_prot bin_writer_relevant_key_data (parent_key, value)
 
   (* Map of vector clocks to values
      Aim to replicate this to other nodes
@@ -44,7 +45,8 @@ module CommandTree (Value : Value) = struct
 
   let root_key = Md5.digest_string ""
 
-  let get_key_of_node node = match node with None -> root_key | Some {key; _} -> key
+  let get_key_of_node node =
+    match node with None -> root_key | Some {key; _} -> key
 
   let get_idx_of_node node =
     match node with None -> 0 | Some {node= idx, _, _; _} -> idx
@@ -224,7 +226,7 @@ module CommandTree (Value : Value) = struct
         Some GT
     | Some ({node= ia, _, _; _} as na), Some ({node= ib, _, _; _} as nb) -> (
         let rec on_path t curr ({key= kt; node= it, _, _; _} as target) =
-          assert (get_idx_of_node curr >= it);
+          assert (get_idx_of_node curr >= it) ;
           match curr with
           | None ->
               false

impl/lib/conspire_dc.ml

@@ -2,10 +2,13 @@ open! Core
 open Types
 module Time = Time_float_unix
 
+let pp_time_float_unix : Time.t Fmt.t =
+ fun ppf v -> Fmt.pf ppf "%0.5f" (Utils.time_to_float v)
+
 module Value = struct
   let pp_command = Command.pp
 
-  type t = command list * Time.t
+  type t = command list * (Time.t[@printer pp_time_float_unix])
   [@@deriving compare, equal, hash, bin_io, sexp, show]
 
   let empty = ([], Time.epoch)
@@ -26,13 +29,10 @@ end
 *)
 
 module DelayReorderBuffer = struct
-  let pp_time_float_unix : Time.t Fmt.t =
-   fun ppf v -> Fmt.pf ppf "%0.5f" (Utils.time_to_float v)
-
   type 'a t =
     { mutable store: 'a list Map.M(Time).t
           [@polyprinter fun pa -> Utils.pp_map pp_time_float_unix pa]
-    ; mutable hwm: Time.t
+    ; mutable hwm: Time.t [@printer pp_time_float_unix]
     ; interval: Time.Span.t
     ; compare: 'a -> 'a -> int }
   [@@deriving show {with_path= false}]
@@ -121,7 +121,7 @@ module Types = struct
     ; clock= (clock :> float Eio.Time.clock_ty Eio.Time.clock) }
 
   type message =
-    | Commands of (Command.t list * Time.t)
+    | Commands of (Command.t list * (Time.t[@printer pp_time_float_unix]))
     | Conspire of Conspire.message
   [@@deriving show, bin_io]
 

impl/lib/conspire_f.ml

@@ -128,11 +128,15 @@ module Make (Value : Value) = struct
 
   type t =
     { rep: Rep.rep
-    ; other_nodes_state: state Map.M(Int).t [@printer Utils.pp_map Fmt.int pp_state]
+    ; other_nodes_state: state Map.M(Int).t
+          [@printer Utils.pp_map Fmt.int pp_state]
     ; config: config [@opaque]
     ; commit_log: Value.t Log.t }
   [@@deriving show {with_path= false}]
 
+  let reporter_conflict, run_c =
+    Ocons_core.Utils.InternalReporter.rate_reporter "conflict"
+
   let acceptor_reply t src =
     let local = t.rep.state in
     let remote = Map.find_exn t.other_nodes_state src in
@@ -148,6 +152,7 @@ module Make (Value : Value) = struct
         let res = CTree.compare_keys t.rep.store local.vval remote.vval in
         match res with
         | None ->
+            reporter_conflict () ;
             (* conflict *)
             Utils.dtraceln "CONFLICT from %d" src ;
             Utils.dtraceln "local %a does not prefix of remote %a"
@@ -240,8 +245,7 @@ module Make (Value : Value) = struct
     check_commit t
 
   let acceptor_term_tick t term' =
-    if t.rep.state.term < term' then
-      t.rep.state.term <- term'
+    if t.rep.state.term < term' then t.rep.state.term <- term'
 
   let handle_steady_state t src (msg : Rep.success) =
     let option_bind o ~f = Option.value_map o ~default:(Ok ()) ~f in
@@ -261,9 +265,8 @@ module Make (Value : Value) = struct
         set_state remote new_state ;
         acceptor_reply t src ;
         check_commit t ;
-        check_conflict_recovery t;
-        acceptor_term_tick t new_state.term;
-      ) ;
+        check_conflict_recovery t ;
+        acceptor_term_tick t new_state.term ) ;
     Result.return ()
 
   let handle_message t src (msg : Rep.message) :
@@ -280,6 +283,7 @@ module Make (Value : Value) = struct
         Error `MustAck
 
   let create (config : config) =
+    run_c := true ;
     let rep = Rep.create config.other_replica_ids in
     let other_nodes_state =
       List.map config.other_replica_ids ~f:(fun i -> (i, init_state rep.store))

impl/lib/conspire_mp.ml

@@ -91,11 +91,7 @@ module Types = struct
       Conspire_f.
         {node_id; replica_ids; other_replica_ids; replica_count; quorum_size}
     in
-    { conspire
-    ; other_replica_ids
-    ; lower_replica_ids
-    ; fd_timeout
-    ; max_outstanding }
+    {conspire; other_replica_ids; lower_replica_ids; fd_timeout; max_outstanding}
 
   type message = Conspire.message [@@deriving show, bin_io]