Reset worker errors, retry failed workers

ebin/gmhive_client.app

@@ -1,6 +1,6 @@
 {application,gmhive_client,
              [{description,"Gajumaru Hive Client"},
-              {vsn,"0.4.7"},
+              {vsn,"0.4.8"},
               {registered,[]},
               {applications,[kernel,stdlib,sasl,gproc,inets,ssl,enoise,
                              gmconfig,gmhive_protocol,gmhive_worker]},

src/gmhc_app.erl

@@ -1,5 +1,6 @@
 %% -*- mode: erlang; erlang-indent-level: 4; indent-tabs-mode: nil -*-
 -module(gmhc_app).
+-vsn("0.4.8").
 
 -behaviour(application).
 

src/gmhc_config.erl

@@ -1,4 +1,5 @@
 -module(gmhc_config).
+-vsn("0.4.8").
 
 -export([ load_config/0
         , get_config/1

src/gmhc_config_schema.erl

@@ -1,4 +1,5 @@
 -module(gmhc_config_schema).
+-vsn("0.4.8").
 
 -export([ schema/0
         , to_json/0 ]).

src/gmhc_connector.erl

@@ -1,4 +1,5 @@
 -module(gmhc_connector).
+-vsn("0.4.8").
 
 -behaviour(gen_server).
 

src/gmhc_connectors_sup.erl

@@ -1,4 +1,5 @@
 -module(gmhc_connectors_sup).
+-vsn("0.4.8").
 -behavior(supervisor).
 
 -export([ start_link/0

src/gmhc_counters.erl

@@ -1,4 +1,5 @@
 -module(gmhc_counters).
+-vsn("0.4.8").
 
 -export([ initialize/0 ]).
 

src/gmhc_eureka.erl

@@ -1,4 +1,5 @@
 -module(gmhc_eureka).
+-vsn("0.4.8").
 
 -export([get_pool_address/0]).
 

src/gmhc_events.erl

@@ -1,4 +1,5 @@
 -module(gmhc_events).
+-vsn("0.4.8").
 
 -export([subscribe/1,
          ensure_subscribed/1,

src/gmhc_handler.erl

@@ -1,4 +1,5 @@
 -module(gmhc_handler).
+-vsn("0.4.8").
 -behavior(gen_server).
 
 -export([ start_link/0

src/gmhc_server.erl

@@ -1,4 +1,5 @@
 -module(gmhc_server).
+-vsn("0.4.8").
 
 -behaviour(gen_server).
 
@@ -44,7 +45,7 @@
 
 -define(CONNECTED(S), map_size(S#st.connected) > 0).
 
--define(MAX_ERRORS, 5).
+-define(MAX_ERRORS, 50).
 
 connected(Id, Type) ->
     gen_server:call(?MODULE, {connected, Id, Type}).
@@ -120,7 +121,7 @@ handle_cast({from_pool, #{via := Connector,
     catch
        Cat:Err:St ->
            ?LOG_ERROR("CAUGHT ~p:~p / ~p", [Cat, Err, St]),
-	   {noreply, S}
+           {noreply, S}
     end;
 handle_cast({disconnected, Id}, #st{connected = Conn} = S) ->
     ?LOG_DEBUG("disconnected: ~p", [Id]),
@@ -154,11 +155,23 @@ handle_info({'EXIT', Pid, Reason}, #st{ workers = Workers
             gmhc_events:publish(error, ?ERR_EVT(#{error => worker_error,
                                                    data  => Reason})),
             Ws1 = incr_worker_error(W, Workers),
+            erlang:start_timer(100, self(), check_workers),
             {noreply, S#st{workers = Ws1}};
         false ->
             %% ?LOG_DEBUG("EXIT apparently not from worker?? (~p)", [Pid]),
             {noreply, S}
     end;
+handle_info({timeout, _, check_workers}, #st{workers = Workers} = S) ->
+    case [W || #worker{cand = undefined} = W <- Workers] of
+        [] ->
+            {noreply, S};
+        Idle ->
+            S1 = maybe_request_nonces(S),
+            S2 = lists:foldl(fun(W, Sx) ->
+                                 maybe_restart_worker(W, Sx)
+                             end, S1, Idle),
+            {noreply, S2}
+    end;
 handle_info(Msg, St) ->
     ?LOG_DEBUG("Unknown msg: ~p", [Msg]),
     {noreply, St}.
@@ -226,14 +239,14 @@ handle_worker_result({worker_result, Result}, W, S) ->
     case Result of
         {solutions, Solutions} ->
             {Cont, S1} = report_solutions_(Solutions, W, S),
-            maybe_continue(Cont, W, S1);
+            maybe_continue(Cont, reset_errors(W), S1);
         {solution, Nonce, Solution} ->
             %% report_solution(Nonce, Solution, W, S),
             {Cont, S1} = report_solutions_([{Nonce, Solution}], W, S),
-            maybe_continue(Cont, W, S1);
+            maybe_continue(Cont, reset_errors(W), S1);
         {no_solution, Nonce} ->
             report_no_solution(Nonce, W, S),
-            maybe_restart_worker(W, S);
+            maybe_restart_worker(reset_errors(W), S);
         {error, S} ->
             ?LOG_DEBUG("Worker ~p reported error as normal", [W#worker.index]),
             gmhc_events:publish(error, ?ERR_EVT(#{error => worker_error,
@@ -261,6 +274,9 @@ report_solutions_(Solutions, W, S) ->
             {error, S}
     end.
 
+reset_errors(#worker{} = W) ->
+    W#worker{errors = 0}.
+
 reset_worker(#worker{index = I} = W, Ws) ->
     W1 = reset_worker_(W),
     lists:keyreplace(I, #worker.index, Ws, W1).

src/gmhc_sup.erl

@@ -1,5 +1,6 @@
 %% -*- mode: erlang; erlang-indent-level: 4; indent-tabs-mode: nil -*-
 -module(gmhc_sup).
+-vsn("0.4.8").
 
 -behaviour(supervisor).
 

src/gmhc_workers.erl

@@ -8,6 +8,7 @@
 %%%-------------------------------------------------------------------
 
 -module(gmhc_workers).
+-vsn("0.4.8").
 
 -export([
           get_worker_configs/0

src/gmhive_client.erl

@@ -1,4 +1,5 @@
 -module(gmhive_client).
+-vsn("0.4.8").
 
 -export([ connect/1
         , disconnect/1

zomp.meta

@@ -4,7 +4,7 @@
 {prefix,"gmhc"}.
 {author,"Ulf Wiger, QPQ AG"}.
 {desc,"Gajumaru Hive Client"}.
-{package_id,{"uwiger","gmhive_client",{0,4,7}}}.
+{package_id,{"uwiger","gmhive_client",{0,4,8}}}.
 {deps,[{"uwiger","gmhive_worker",{0,4,0}},
        {"uwiger","gmcuckoo",{1,2,3}},
        {"otpr","eblake2",{1,0,1}},