wip changelog

.gitignore

@@ -7,3 +7,6 @@ _opam
 *.install
 *.exe
 *.tmp
+src/lib/version.ml
+src/lib/.git_index_path
+src/lib/.git_index.lnk

CHANGES.md

@@ -1,4 +1,36 @@
 
+## 0.20
+
+- major refactor: split library into `opentelemetry.core`, `opentelemetry`,
+  `opentelemetry.util`, `opentelemetry.emitter`, `opentelemetry.atomic`, revamp internals
+- per-signal providers: separate trace, meter, and logger providers replace
+  the single monolithic exporter
+- `opentelemetry.ambient-context` is now a standalone library, once again
+- new `opentelemetry-client-ocurl-lwt` package
+- client: split `opentelemetry-client-sync` off of the main client library
+- client: add support for `http/json` protocol alongside `http/protobuf`
+- client: add HTTP retry with exponential backoff
+- client: overhaul bounded queue; introduce generic consumer framework
+- client: add `Exporter_add_batching`, `Emitter_add_batching`, `Emitter_sample`,
+  `Emitter_limit_interval` combinators. Batching is factored out of individual
+  client libraries.
+- client: add sampler as an emitter transformer
+- client: add `exporter_stdout` and `debug_exporter`
+- client: add `self_metrics` and `self_debug` to exporters
+- client: add `after_shutdown` callback in ocurl/ocurl-lwt clients
+- `Span.dummy`: inert span that is never modified
+- `Span.record_exception` now also sets the span status to error
+- `Span.set_span_status` added in `opentelemetry.trace`
+- `Span`: carry flags to `span_link`
+- `Span`: now mutable thanks to ocaml-protoc 4.0, replaces old `Scope.t` entirely
+- `Meter.emit` and `Meter_provider.emit_l` added
+- emitter: add `flat_map`, `tap`, `to_list`, `enabled` combinators
+- clock abstraction added; `ptime` used by default in logger and metrics
+- interval limiter used for `metrics_callbacks`
+- update to OTEL spec 1.8.0
+- update semantic conventions
+- various bug fixes and performance improvements
+
 ## 0.12
 
 - breaking: change `Collector.cleanup` so it takes a callback

src/client-cohttp-eio/opentelemetry_client_cohttp_eio.ml

@@ -186,7 +186,8 @@ let setup_ ~sw ~config env : unit =
 
   Opentelemetry.Self_debug.log Opentelemetry.Self_debug.Info (fun () ->
       "opentelemetry: cohttp-eio exporter installed");
-  Opentelemetry_client.Self_trace.set_enabled config.self_trace
+  Opentelemetry_client.Self_trace.set_enabled config.self_trace;
+  if config.self_metrics then Opentelemetry.Sdk.setup_self_metrics ()
 
 let setup ?(config = Config.make ()) ?(enable = true) ~sw env =
   if enable && not config.sdk_disabled then setup_ ~sw ~config env

src/client-cohttp-lwt/opentelemetry_client_cohttp_lwt.ml

@@ -120,6 +120,7 @@ let setup_ ~config () : unit =
   Opentelemetry.Self_debug.log Opentelemetry.Self_debug.Info (fun () ->
       "opentelemetry: cohttp-lwt exporter installed");
   Opentelemetry_client.Self_trace.set_enabled config.self_trace;
+  if config.self_metrics then Opentelemetry.Sdk.setup_self_metrics ();
   ()
 
 let setup ?(config = Config.make ()) ?(enable = true) () =

src/client-ocurl-lwt/opentelemetry_client_ocurl_lwt.ml

@@ -95,6 +95,7 @@ let setup_ ~config () : Exporter.t =
   Opentelemetry.Self_debug.log Opentelemetry.Self_debug.Info (fun () ->
       "opentelemetry: ocurl-lwt exporter installed");
   Opentelemetry_client.Self_trace.set_enabled config.self_trace;
+  if config.self_metrics then Opentelemetry.Sdk.setup_self_metrics ();
 
   exp
 

src/client-ocurl/opentelemetry_client_ocurl.ml

@@ -100,6 +100,7 @@ let setup_ ~config () : OTEL.Exporter.t =
       "opentelemetry: ocurl exporter installed");
 
   OTELC.Self_trace.set_enabled config.common.self_trace;
+  if config.common.self_metrics then Opentelemetry.Sdk.setup_self_metrics ();
   exporter
 
 let remove_exporter () : unit =

src/client/exporter_config.ml

@@ -26,6 +26,7 @@ type t = {
   metrics: Opentelemetry.Provider_config.t;
   logs: Opentelemetry.Provider_config.t;
   self_trace: bool;
+  self_metrics: bool;
   http_concurrency_level: int option;
   retry_max_attempts: int;
   retry_initial_delay_ms: float;
@@ -65,6 +66,7 @@ let pp out (self : t) : unit =
     log_level;
     sdk_disabled;
     self_trace;
+    self_metrics;
     url_traces;
     url_metrics;
     url_logs;
@@ -91,7 +93,8 @@ let pp out (self : t) : unit =
   in
   Format.fprintf out
     "{@[ debug=%B;@ log_level=%a;@ sdk_disabled=%B;@ self_trace=%B;@ \
-     url_traces=%S;@ url_metrics=%S;@ url_logs=%S;@ @[<2>headers=@,\
+     self_metrics=%B;@ url_traces=%S;@ url_metrics=%S;@ url_logs=%S;@ \
+     @[<2>headers=@,\
      %a@];@ @[<2>headers_traces=@,\
      %a@];@ @[<2>headers_metrics=@,\
      %a@];@ @[<2>headers_logs=@,\
@@ -100,8 +103,8 @@ let pp out (self : t) : unit =
      logs=%a;@ http_concurrency_level=%a;@ retry_max_attempts=%d;@ \
      retry_initial_delay_ms=%.0f;@ retry_max_delay_ms=%.0f;@ \
      retry_backoff_multiplier=%.1f @]}"
-    debug pp_log_level log_level sdk_disabled self_trace url_traces url_metrics
-    url_logs ppheaders headers ppheaders headers_traces ppheaders
+    debug pp_log_level log_level sdk_disabled self_trace self_metrics url_traces
+    url_metrics url_logs ppheaders headers ppheaders headers_traces ppheaders
     headers_metrics ppheaders headers_logs pp_protocol protocol timeout_ms
     timeout_traces_ms timeout_metrics_ms timeout_logs_ms pp_provider_config
     traces pp_provider_config metrics pp_provider_config logs ppiopt
@@ -135,6 +138,7 @@ type 'k make =
   ?timeout_metrics_ms:int ->
   ?timeout_logs_ms:int ->
   ?self_trace:bool ->
+  ?self_metrics:bool ->
   ?http_concurrency_level:int ->
   ?retry_max_attempts:int ->
   ?retry_initial_delay_ms:float ->
@@ -245,9 +249,9 @@ module Env () : ENV = struct
       ?(protocol = get_protocol_from_env "OTEL_EXPORTER_OTLP_PROTOCOL")
       ?(timeout_ms = get_timeout_from_env "OTEL_EXPORTER_OTLP_TIMEOUT" 10_000)
       ?timeout_traces_ms ?timeout_metrics_ms ?timeout_logs_ms
-      ?(self_trace = false) ?http_concurrency_level ?(retry_max_attempts = 3)
-      ?(retry_initial_delay_ms = 100.) ?(retry_max_delay_ms = 5000.)
-      ?(retry_backoff_multiplier = 2.0) =
+      ?(self_trace = false) ?(self_metrics = false) ?http_concurrency_level
+      ?(retry_max_attempts = 3) ?(retry_initial_delay_ms = 100.)
+      ?(retry_max_delay_ms = 5000.) ?(retry_backoff_multiplier = 2.0) =
     let batch_timeout_ = Mtime.Span.(batch_timeout_ms * ms) in
     let traces =
       match traces with
@@ -375,6 +379,7 @@ module Env () : ENV = struct
         metrics;
         logs;
         self_trace;
+        self_metrics;
         http_concurrency_level;
         retry_max_attempts;
         retry_initial_delay_ms;

src/client/exporter_config.mli

@@ -72,6 +72,10 @@ type t = {
       (** If true, the OTEL library will perform some self-instrumentation.
           Default [false].
           @since 0.7 *)
+  self_metrics: bool;
+      (** If true, the OTEL library will regularly emit metrics about itself.
+          Default [false].
+          @since NEXT_RELEASE *)
   http_concurrency_level: int option;
       (** How many HTTP requests can be done simultaneously (at most)? This can
           be used to represent the size of a pool of workers where each worker
@@ -123,6 +127,7 @@ type 'k make =
   ?timeout_metrics_ms:int ->
   ?timeout_logs_ms:int ->
   ?self_trace:bool ->
+  ?self_metrics:bool ->
   ?http_concurrency_level:int ->
   ?retry_max_attempts:int ->
   ?retry_initial_delay_ms:float ->

src/client/util_resources.ml

@@ -2,14 +2,19 @@
 
 open Common_
 
+open struct
+  let mk_resources ?service_name ?attrs () =
+    let attributes = OTEL.Globals.mk_attributes ?service_name ?attrs () in
+    Proto.Resource.make_resource ~attributes ()
+end
+
 let make_resource_logs ?service_name ?attrs (logs : Proto.Logs.log_record list)
     : Proto.Logs.resource_logs =
-  let attributes = OTEL.Globals.mk_attributes ?service_name ?attrs () in
-  let resource = Proto.Resource.make_resource ~attributes () in
   let ll =
     Proto.Logs.make_scope_logs ~scope:OTEL.Globals.instrumentation_library
       ~log_records:logs ()
   in
+  let resource = mk_resources ?service_name ?attrs () in
   Proto.Logs.make_resource_logs ~resource ~scope_logs:[ ll ] ()
 
 let make_resource_spans ?service_name ?attrs spans : Proto.Trace.resource_spans
@@ -18,8 +23,7 @@ let make_resource_spans ?service_name ?attrs spans : Proto.Trace.resource_spans
     Proto.Trace.make_scope_spans ~scope:OTEL.Globals.instrumentation_library
       ~spans ()
   in
-  let attributes = OTEL.Globals.mk_attributes ?service_name ?attrs () in
-  let resource = Proto.Resource.make_resource ~attributes () in
+  let resource = mk_resources ?service_name ?attrs () in
   Proto.Trace.make_resource_spans ~resource ~scope_spans:[ ils ] ()
 
 (** Aggregate metrics into a {!Proto.Metrics.resource_metrics} *)
@@ -29,6 +33,5 @@ let make_resource_metrics ?service_name ?attrs (l : OTEL.Metrics.t list) :
   let lm =
     make_scope_metrics ~scope:OTEL.Globals.instrumentation_library ~metrics:l ()
   in
-  let attributes = OTEL.Globals.mk_attributes ?service_name ?attrs () in
-  let resource = Proto.Resource.make_resource ~attributes () in
+  let resource = mk_resources ?service_name ?attrs () in
   Proto.Metrics.make_resource_metrics ~scope_metrics:[ lm ] ~resource ()

src/lib/dune

@@ -1,3 +1,26 @@
+(rule
+ (target .git_index_path)
+ (deps (universe) gen_git_index_path.sh)
+ (action
+  (with-stdout-to
+   .git_index_path
+   (run sh gen_git_index_path.sh))))
+
+(rule
+ (target .git_index.lnk)
+ (action
+  (run ln -sf %{read-lines:.git_index_path} %{target})))
+
+(rule
+ (target version.ml)
+ (deps
+  (file .git_index.lnk)
+  gen_version.sh)
+ (action
+  (with-stdout-to
+   version.ml
+   (run sh gen_version.sh))))
+
 (library
  (name opentelemetry)
  (public_name opentelemetry)
@@ -24,4 +47,5 @@
   mtime
   mtime.clock.os
   pbrt
+  unix
   threads))

src/lib/gen_git_index_path.sh

@@ -0,0 +1,2 @@
+#!/bin/sh
+git rev-parse --git-path index 2>/dev/null || echo /dev/null

src/lib/gen_version.sh

@@ -0,0 +1,6 @@
+#!/bin/sh
+v=$(git describe --exact-match HEAD 2>/dev/null | sed 's/^v//')
+[ -z "$v" ] && v=dev
+h=$(git rev-parse HEAD 2>/dev/null)
+[ -z "$h" ] && h=unknown
+printf 'let version = "%s"\nlet git_hash = "%s"\n' "$v" "$h"

src/lib/globals.ml

@@ -20,8 +20,11 @@ let service_instance_id = ref None
     @since 0.12 *)
 let service_version = ref None
 
+(** @since NEXT_RELEASE *)
+let sdk_version : string = Version.(spf "%s at %s" version git_hash)
+
 let instrumentation_library =
-  make_instrumentation_scope ~version:"%%VERSION_NUM%%" ~name:"ocaml-otel" ()
+  make_instrumentation_scope ~version:sdk_version ~name:"opentelemetry" ()
 
 (** Global attributes, initially set via OTEL_RESOURCE_ATTRIBUTES and modifiable
     by the user code. They will be attached to each outgoing metrics/traces. *)
@@ -58,6 +61,15 @@ open struct
       ]
 
   let runtime_attributes_converted = List.map Key_value.conv runtime_attributes
+
+  let sdk_attributes =
+    [
+      "telemetry.sdk.language", `String "ocaml";
+      "telemetry.sdk.name", `String "opentelemetry";
+      "telemetry.sdk.version", `String sdk_version;
+    ]
+
+  let sdk_attributes_converted = List.map Key_value.conv sdk_attributes
 end
 
 (** Attributes about the OCaml runtime. See
@@ -65,9 +77,12 @@ end
 *)
 let[@inline] get_runtime_attributes () = runtime_attributes
 
+(** Build main global attributes, to be used in resource attributes (shared by
+    all spans/metrics/logs in each outgoing batch) *)
 let mk_attributes ?(service_name = !service_name) ?(attrs = []) () : _ list =
   let l = List.rev_map Key_value.conv attrs in
   let l = List.rev_append runtime_attributes_converted l in
+  let l = List.rev_append sdk_attributes_converted l in
   let l =
     make_key_value ~key:Conventions.Attributes.Service.name
       ~value:(String_value service_name) ()

src/lib/opentelemetry.ml

@@ -100,6 +100,7 @@ type key_value = Key_value.t
 (** {2 Global settings} *)
 
 module Globals = Globals
+module Version = Version
 
 (** {2 Traces and Spans} *)
 

src/lib/sdk.ml

@@ -90,9 +90,35 @@ let set ?(traces = Provider_config.default) ?(metrics = Provider_config.default)
   Log_provider.set logger
 
 let self_metrics () : Metrics.t list =
+  let now = Clock.now_main () in
+  let emitter_metrics =
+    Emitter.self_metrics (Trace_provider.get ()).emit ~now
+    @ Emitter.self_metrics (Meter_provider.get ()).emit ~now
+    @ Emitter.self_metrics (Log_provider.get ()).emit ~now
+  in
   match get () with
-  | None -> []
-  | Some exp -> exp.Exporter.self_metrics ()
+  | None -> emitter_metrics
+  | Some exp -> exp.Exporter.self_metrics () @ emitter_metrics
+
+open struct
+  let self_metrics_enabled = Atomic.make false
+end
+
+(** Regularly emit metrics about the OTEL SDK. Idempotent. *)
+let setup_self_metrics () =
+  if not (Atomic.exchange self_metrics_enabled true) then (
+    Self_debug.log Info (fun () -> "enabling self metrics");
+    let interval_limiter =
+      Interval_limiter.create ~min_interval:Mtime.Span.(10 * s) ()
+    in
+    let on_tick () =
+      if Interval_limiter.make_attempt interval_limiter then (
+        let ms = self_metrics () in
+        Meter_provider.emit_l ms
+      )
+    in
+    Globals.add_on_tick_callback on_tick
+  )
 
 (* Permanent tick callback to drive batch timeouts on provider emitters *)
 let () =

src/lib/version.mli

@@ -0,0 +1,7 @@
+val version : string
+(** Version of the library, e.g. ["0.12"]. ["dev"] if not built from a release
+    tag. *)
+
+val git_hash : string
+(** Full git commit hash at build time, e.g. ["b92159c1..."]. ["unknown"] if git
+    was unavailable. *)

tests/client/test_client_lib.ml

@@ -8,7 +8,7 @@ let test_config_printing () =
   in
   let expected =
     "{ debug=false; log_level=info; sdk_disabled=false; self_trace=false;\n\
-    \ url_traces=\"http://localhost:4318/v1/traces\";\n\
+    \ self_metrics=false; url_traces=\"http://localhost:4318/v1/traces\";\n\
     \ url_metrics=\"http://localhost:4318/v1/metrics\";\n\
     \ url_logs=\"http://localhost:4318/v1/logs\"; headers=[]; headers_traces=[];\n\
     \ headers_metrics=[]; headers_logs=[]; protocol=http/protobuf;\n\

tests/logs/test_logs_e2e.expected

@@ -7,8 +7,8 @@
   scope_logs =
     [{ scope =
          Some(
-           { name = "ocaml-otel";
-             version = "%%VERSION_NUM%%";
+           { name = "opentelemetry";
+             version = "";
              attributes = [];
              dropped_attributes_count = 0 (* absent *);
            });

tests/logs/test_logs_e2e.ml

@@ -42,6 +42,11 @@ let tests (signal_batches : Client.Resource_signal.t list) =
                            lr)
                          sl.log_records
                      in
+                     Option.iter
+                       (fun sc ->
+                         Opentelemetry_proto.Common
+                         .instrumentation_scope_set_version sc "")
+                       sl.scope;
                      let sl = L.copy_scope_logs sl in
                      L.scope_logs_set_log_records sl masked_log_records;
                      sl)