TD-788: Adds prometheus support for hackney per host metrics (#32)

src/woody_hackney_prometheus.erl

@@ -0,0 +1,134 @@
+%% Translates hackney metrics for prometheus.
+%% See
+%% https://github.com/benoitc/hackney/tree/1.18.0?tab=readme-ov-file#metrics
+%%
+%% For pool metrics see dedicated collector module
+%% `woody_hackney_prometheus_collector`.
+-module(woody_hackney_prometheus).
+
+-export([
+    new/2,
+    delete/1,
+    increment_counter/1,
+    increment_counter/2,
+    decrement_counter/1,
+    decrement_counter/2,
+    update_histogram/2,
+    update_gauge/2,
+    update_meter/2
+]).
+
+-type name() :: any().
+-type metric() :: counter | histogram | gauge | meter.
+
+%% API
+
+-define(NB_REQUESTS, hackney_nb_requests).
+-define(TOTAL_REQUESTS, hackney_total_requests).
+-define(HOST_NB_REQUESTS, hackney_host_nb_requests).
+-define(HOST_CONNECT_TIMEOUT, hackney_host_connect_timeout).
+-define(HOST_CONNECT_ERROR, hackney_host_connect_error).
+-define(HOST_NEW_CONNECTION, hackney_host_new_connection).
+-define(HOST_REUSE_CONNECTION, hackney_host_reuse_connection).
+
+%% Ignore unsupported metric
+-spec new(metric(), name()) -> ok.
+%% Total counters
+new(counter, [hackney, nb_requests]) ->
+    true = prometheus_gauge:declare([
+        {name, ?NB_REQUESTS},
+        {registry, registry()},
+        {labels, []},
+        {help, "Number of running requests."}
+    ]),
+    %% Per host counters, see
+    %% https://github.com/benoitc/hackney/tree/1.18.0?tab=readme-ov-file#metrics-per-hosts
+    %% NOTE Hackney won't call `metrics:new/3` for those counters
+    true = prometheus_gauge:declare([
+        {name, ?HOST_NB_REQUESTS},
+        {registry, registry()},
+        {labels, [host]},
+        {help, "Number of running requests."}
+    ]),
+    [
+        true = prometheus_counter:declare([
+            {name, Name},
+            {registry, registry()},
+            {labels, [host]},
+            {help, Help}
+        ])
+     || {Name, Help} <- [
+            {?HOST_CONNECT_TIMEOUT, "Number of connect timeout."},
+            {?HOST_CONNECT_ERROR, "Number of timeout errors."},
+            {?HOST_NEW_CONNECTION, "Number of new pool connections per host."},
+            {?HOST_REUSE_CONNECTION, "Number of reused pool connections per host."}
+        ]
+    ],
+    ok;
+new(counter, [hackney, total_requests]) ->
+    true = prometheus_counter:declare([
+        {name, ?TOTAL_REQUESTS},
+        {registry, registry()},
+        {labels, []},
+        {help, "Total number of requests."}
+    ]),
+    ok;
+new(_Type, _Name) ->
+    ok.
+
+-spec delete(name()) -> ok.
+delete(_Name) ->
+    ok.
+
+-spec increment_counter(name()) -> ok.
+increment_counter(Name) ->
+    increment_counter(Name, 1).
+
+-spec increment_counter(name(), pos_integer()) -> ok.
+increment_counter([hackney, nb_requests], Value) ->
+    prometheus_gauge:inc(registry(), ?NB_REQUESTS, [], Value);
+increment_counter([hackney, total_requests], Value) ->
+    prometheus_counter:inc(registry(), ?TOTAL_REQUESTS, [], Value);
+increment_counter([hackney, Host, nb_requests], Value) ->
+    prometheus_gauge:inc(registry(), ?HOST_NB_REQUESTS, [{host, Host}], Value);
+increment_counter([hackney, Host, connect_timeout], Value) ->
+    prometheus_counter:inc(registry(), ?HOST_CONNECT_TIMEOUT, [{host, Host}], Value);
+increment_counter([hackney, Host, connect_error], Value) ->
+    prometheus_counter:inc(registry(), ?HOST_CONNECT_ERROR, [{host, Host}], Value);
+increment_counter([hackney_pool, Host, new_connection], Value) ->
+    prometheus_counter:inc(registry(), ?HOST_NEW_CONNECTION, [{host, Host}], Value);
+increment_counter([hackney_pool, Host, reuse_connection], Value) ->
+    prometheus_counter:inc(registry(), ?HOST_REUSE_CONNECTION, [{host, Host}], Value);
+increment_counter(_Name, _Value) ->
+    ok.
+
+-spec decrement_counter(name()) -> ok.
+decrement_counter(Name) ->
+    decrement_counter(Name, 1).
+
+-spec decrement_counter(name(), pos_integer()) -> ok.
+decrement_counter([hackney, nb_requests], Value) ->
+    prometheus_gauge:dec(registry(), ?NB_REQUESTS, [], Value);
+decrement_counter([hackney, Host, nb_requests], Value) ->
+    prometheus_gauge:dec(registry(), ?HOST_NB_REQUESTS, [{host, Host}], Value);
+decrement_counter(_Name, _Value) ->
+    ok.
+
+-spec update_histogram(name(), fun(() -> ok) | number()) -> ok.
+update_histogram(_Name, Fun) when is_function(Fun, 0) ->
+    Fun();
+update_histogram(_Name, _Value) ->
+    ok.
+
+-spec update_gauge(name(), number()) -> ok.
+update_gauge(_Name, _Value) ->
+    ok.
+
+-spec update_meter(name(), number()) -> ok.
+update_meter(_Name, _Value) ->
+    ok.
+
+%%
+
+registry() ->
+    default.

test/woody_tests_SUITE.erl

@@ -290,8 +290,9 @@ init_per_suite(C) ->
     % dbg:tracer(), dbg:p(all, c),
     % dbg:tpl({woody_joint_workers, '_', '_'}, x),
     %%Apps = genlib_app:start_application_with(woody, [{trace_http_server, true}]),
-    {ok, Apps} = application:ensure_all_started(woody),
+    ok = application:set_env(hackney, mod_metrics, woody_hackney_prometheus),
     {ok, MetricsApps} = application:ensure_all_started(prometheus),
+    {ok, Apps} = application:ensure_all_started(woody),
     ok = woody_ranch_prometheus_collector:setup(),
     ok = woody_hackney_prometheus_collector:setup(),
     {ok, OtelApps} = setup_opentelemetry(),