Reset worker errors, retry failed workers

Reviewed-on: #8
Reviewed-by: Jarvis Carroll <jarviscarrol@qpq.swiss>

README.md

@@ -1,11 +1,213 @@
 # Gajumaru Hive Client
 
-When running tests locally:
+This application is primarily meant to be used by the [GajuMine](https://git.qpq.swiss/QPQ-AG/gajumine)
+application, but can also be run as a headless worker. It can be built either using `rebar3` or
+using [`zx`](https://zxq9.com/projects/zomp/) (recomended).
 
-`rebar3 shell`
+## Operation
+The `gmhive_client` needs at least a worker public key to operate.
+It first queries the [GajuMining website](https://gajumining.com) to find the address of
+the Gajumaru Hive server, which it then connects to. GajuMining won't provide the address
+unless the public key is a registered worker.
 
-If connecting to the `testnet` hive:
-`GMMP_CLIENT_CONFIG=gmhc_config-testnet.eterm rebar3 shell`
+## Configuration
 
-If connecting to a `testnet` hive running locally:
-`GMHC_CONFIG=gmhc_config-testnet-local.eterm rebar3 shell`
+The Hive Client uses a configuration schema ([see below](#json-schema)).
+At a minimum, a value for `"pubkey"` must be provided. All other configuration
+items have usable defaults.
+
+Examples:
+
+**.json**:
+```json
+{ "pubkey" : "ak_2NQoybA6uyvhu3cdzWFtZnxCZpZvKE5TxjYB3hB7kMWEquwLVY",
+  "extra_pubkeys" : [ "ak_2Dfwyb7ZFhcAELoxgWyMCkVnpZUhwUm6aftKcv6dakagePWyx9" ]
+}
+```
+
+**.eterm**:
+```erlang
+#{ <<"pubkey">> => <<"ak_2NQoybA6uyvhu3cdzWFtZnxCZpZvKE5TxjYB3hB7kMWEquwLVY">>,
+   <<"extra_pubkeys">> : [ <<"ak_2Dfwyb7ZFhcAELoxgWyMCkVnpZUhwUm6aftKcv6dakagePWyx9">> ]
+ }.
+```
+
+Configuration can be provided in a file, where the filename extension defines
+the format: `".json"` for JSON format, `".eterm"` for Erlang term format.
+In the case of Erlang term format, the data should be on "internal JSON form",
+i.e. nested maps, where all strings are of type `binary()`.
+
+By default, the Hive client looks for a configuration file named
+`"gmhive_client_config.[json|eterm]"` in the current working directory.
+A specific configuration file can be identified using the OS environment
+variable `GMHIVE_CLIENT_CONFIG`.
+
+On invocation, individual configuration items can be set in one of two ways:
+
+### OS environment variable overrides
+
+By creating an OS environment variable with the name `GMHC__<variable_name>`,
+an individual variable can be redefined. The name is composed from the 
+configuration key name, converted to uppercase, and with `__` as a delimiter
+for a nested key. The value is coerced into the expected type using the schema.
+Complex values are given in JSON format - be sure to quote it.
+
+Example:
+```
+GMHC__PUBKEY="ak_2Dfwyb7ZFhcAELoxgWyMCkVnpZUhwUm6aftKcv6dakagePWyx9" zx run uwiger-gmhive_client
+```
+
+```
+GMHC__WORKERS='[{"executable": "mean29-avx2"}]' zx run ...
+```
+
+### Command-line arguments
+
+Arguments on the form `-gmhc Key Value` can be added to the command line.
+The key name is derived from the schema, all lowercase and with `__` as delimiter.
+
+Example:
+```
+zx run uwiger-gmhive_client -gmhc pubkey ak_2Dfwyb...
+```
+
+### Considerations
+
+The values for `pool_admin` and `pool` should be left alone. The `"network"`
+configuration item is special. Its default value is `"mainnet"`, which
+indicates that the client should connect to
+[gajumining.com](https://gajumining.com) to find the hive server, but by
+changing its value to `"testnet"`, (or to other networks in the future,) not
+only does this inform the client of the network for which the client should
+mine, but it also changes what URL the client will use to find the hive
+server. (For `"testnet"` this will be `"test.gajumining.com"`.)
+
+## JSON Schema
+
+```json
+{
+    "$schema": "http://json-schema.org/draft-04/schema#",
+    "additionalProperties": false,
+    "properties": {
+        "extra_pubkeys": {
+            "default": [],
+            "description": "Additional worker pubkeys, sharing rewards",
+            "items": {
+                "pattern": "^ak_[1-9A-HJ-NP-Za-km-z]*$",
+                "type": "string"
+            },
+            "type": "array"
+        },
+        "network": {
+            "default": "mainnet",
+            "type": "string"
+        },
+        "pool": {
+            "additionalProperties": false,
+            "properties": {
+                "host": {
+                    "default": "127.0.0.1",
+                    "description": "Hostname of hive server",
+                    "example": "0.0.0.0",
+                    "type": "string"
+                },
+                "id": {
+                    "description": "Pool contract id",
+                    "pattern": "^ct_[1-9A-HJ-NP-Za-km-z]*$",
+                    "type": "string"
+                },
+                "port": {
+                    "default": 17888,
+                    "description": "Hive server listen port",
+                    "minimum": 1,
+                    "type": "integer"
+                }
+            },
+            "type": "object"
+        },
+        "pool_admin": {
+            "additionalProperties": false,
+            "properties": {
+                "default_per_network": {
+                    "additionalProperties": false,
+                    "properties": {
+                        "mainnet": {
+                            "default": "https://gajumining.com/api/workers/{CLIENT_ID}",
+                            "type": "string"
+                        },
+                        "testnet": {
+                            "default": "https://test.gajumining.com/api/workers/{CLIENT_ID}",
+                            "type": "string"
+                        }
+                    },
+                    "type": "object"
+                },
+                "url": {
+                    "default": "https://test.gajumining.com/api/workers/{CLIENT_ID}",
+                    "description": "URL of Eureka worker api",
+                    "type": "string"
+                }
+            },
+            "type": "object"
+        },
+        "pubkey": {
+            "description": "Primary client pubkey",
+            "pattern": "^ak_[1-9A-HJ-NP-Za-km-z]*$",
+            "type": "string"
+        },
+        "type": {
+            "default": "worker",
+            "description": "monitor mode can be used to see if a pool is alive",
+            "enum": [
+                "worker",
+                "monitor"
+            ],
+            "type": "string"
+        },
+        "workers": {
+            "default": [
+                { "executable": "mean29-generic" }
+            ],
+            "description": "Definitions of workers' configurations. If no worker are configured one worker is used as default, i.e. 'mean29-generic' executable without any extra args.",
+            "items": {
+                "additionalProperties": false,
+                "properties": {
+                    "executable": {
+                        "default": "mean29-generic",
+                        "description": "Executable binary of the worker. Can be a fully qualified path, but the software may apply default logic to locate a plain basename.",
+                        "type": "string"
+                    },
+                    "extra_args": {
+                        "default": "",
+                        "description": "Extra arguments to pass to the worker executable binary. The safest choice is specifying no arguments i.e. empty string.",
+                        "type": "string"
+                    },
+                    "hex_encoded_header": {
+                        "default": false,
+                        "description": "Hexadecimal encode the header argument that is send to the worker executable. CUDA executables expect hex encoded header.",
+                        "type": "boolean"
+                    },
+                    "instances": {
+                        "description": "Instances used by the worker in case of Multi-GPU mining. Numbers on the configuration list represent GPU devices that are to be addressed by the worker.",
+                        "example": [0,1,2,3],
+                        "items": { "type": "integer" },
+                        "minItems": 1,
+                        "type": "array"
+                    },
+                    "repeats": {
+                        "default": 1,
+                        "description": "Number of tries to do in each worker context - WARNING: it should be set so the worker process runs for 3-5s or else the node risk missing out on new micro blocks.",
+                        "type": "integer"
+                    }
+                },
+                "required": [
+                    "executable"
+                ],
+                "type": "object"
+            },
+            "type": "array"
+        }
+    },
+    "type": "object"
+}
+```

ebin/gmhive_client.app

@@ -1,6 +1,6 @@
 {application,gmhive_client,
              [{description,"Gajumaru Hive Client"},
-              {vsn,"0.4.3"},
+              {vsn,"0.4.8"},
               {registered,[]},
               {applications,[kernel,stdlib,sasl,gproc,inets,ssl,enoise,
                              gmconfig,gmhive_protocol,gmhive_worker]},

rebar.config

@@ -10,7 +10,7 @@
         {gmhive_protocol,
          {git, "https://git.qpq.swiss/QPQ-AG/gmhive_protocol.git",
           {ref, "818ce33"}}},
-        {gmhive_worker, {git, "https://git.qpq.swiss/QPQ-AG/gmhive_worker", {ref, "255ef59"}}},
+        {gmhive_worker, {git, "https://git.qpq.swiss/QPQ-AG/gmhive_worker", {ref, "fac460714f"}}},
         {gmconfig, {git, "https://git.qpq.swiss/QPQ-AG/gmconfig.git",
                     {ref, "38620ff9e2"}}},
         {gproc, "1.0.0"},

rebar.lock

@@ -29,7 +29,7 @@
   0},
  {<<"gmhive_worker">>,
   {git,"https://git.qpq.swiss/QPQ-AG/gmhive_worker",
-       {ref,"255ef59ccd7f795d2d25f2d0ebcf24e3251b6f36"}},
+       {ref,"fac460714fc228eb0b723a3f292a44aec77f094a"}},
   0},
  {<<"gmserialization">>,
   {git,"https://git.qpq.swiss/QPQ-AG/gmserialization.git",

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
@@ -129,12 +130,15 @@ call_connector(Req0) ->
             gmhc_connector:send(ViaId, #{call => Req#{ id => Id }}),
             receive
                 {from_pool, #{reply := #{ id := Id, result := Result }}} ->
+                    erlang:demonitor(MRef),
                     Result;
                 {from_pool, #{error := #{ id := Id } = Error}} ->
+                    erlang:demonitor(MRef),
                     {error, maps:remove(id, Error)};
                 {'DOWN', MRef, _, _, _} ->
                     {error, no_connection}
             after 5000 ->
+                    erlang:demonitor(MRef),
                     {error, {timeout, process_info(self(), messages)}}
             end
     end.

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}).
@@ -107,17 +108,20 @@ handle_cast({from_pool, #{via := Connector,
     %% We could check whether we have already received the candidate ...
     %% For now, stop all workers, restart with new candidate
     try
-        % Most of the time we don't want to stop the worker. If we do, though, then
-        % we need to do it more carefully than this, or memory usage will triple.
-        % Workers1 = stop_workers(Workers),
-        {Workers2, Cand1} = assign_nonces(Workers, Cand),
-        #st{candidate = Cand2} = S1 = maybe_request_nonces(S#st{candidate = Cand1}),
-        NewWorkers = [spawn_worker(W, Cand2) || W <- Workers2],
-        {noreply, S1#st{workers = NewWorkers}}
+        %% Most of the time we don't want to stop the worker. If we do, though, then
+        %% we need to do it more carefully than this, or memory usage will triple.
+        %% Workers1 = stop_workers(Workers),
+        %%
+        %% Nonces may be [], in which case we need to request new nonces first.
+        #st{candidate = Cand1} = S1 = maybe_request_nonces(S#st{candidate = Cand}),
+        {Workers2, Cand2} = assign_nonces(Workers, Cand1),
+        #st{candidate = Cand3} = S2 = maybe_request_nonces(S1#st{candidate = Cand2}),
+        NewWorkers = [spawn_worker(W, Cand3) || W <- Workers2],
+        {noreply, S2#st{workers = NewWorkers}}
     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]),
@@ -151,11 +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}.
@@ -223,14 +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,
@@ -258,6 +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).
@@ -272,13 +291,14 @@ incr_worker_error(#worker{errors = Es, index = I} = W, Ws) ->
     W1 = reset_worker_(W#worker{errors = Es+1}),
     lists:keyreplace(I, #worker.index, Ws, W1).
 
-maybe_continue(stopped, _, S) ->
-    S;
+%% maybe_continue(stopped, _, S) ->
+%%     S;
 maybe_continue(continue, W, S) ->
     maybe_restart_worker(W, S);
 maybe_continue(error, W, S) ->
     ?LOG_INFO("Won't restart worker ~p due to error", [W#worker.index]),
-    S.
+    Ws = reset_worker(W, S#st.workers),
+    S#st{workers = Ws}.
 
 maybe_restart_worker(#worker{index = I} = W, #st{candidate = C} = S) ->
     case maps:get(nonces, C) of
@@ -314,7 +334,7 @@ stop_workers_for_seq(Seq, Workers) ->
 stop_worker(#worker{pid = Pid} = W) when is_pid(Pid) ->
     MRef = erlang:monitor(process, Pid),
     ?LOG_DEBUG("Will stop worker ~p (MRef = ~p)", [Pid, MRef]),
-    exit(Pid, kill),
+    exit(Pid, shutdown),
     receive
         {'EXIT', Pid, _} -> ok;
         {'DOWN', MRef, process, Pid, _} -> ok

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

@@ -4,9 +4,9 @@
 {prefix,"gmhc"}.
 {author,"Ulf Wiger, QPQ AG"}.
 {desc,"Gajumaru Hive Client"}.
-{package_id,{"uwiger","gmhive_client",{0,4,3}}}.
-{deps,[{"uwiger","gmcuckoo",{1,2,3}},
-       {"uwiger","gmhive_worker",{0,3,0}},
+{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}},
        {"otpr","base58",{0,1,1}},
        {"otpr","gmserialization",{0,1,3}},