diff --git a/.github/workflows/main.yml b/.github/workflows/main.yml
index 3a5c66ec..aebd8efa 100644
--- a/.github/workflows/main.yml
+++ b/.github/workflows/main.yml
@@ -3,7 +3,7 @@ name: build
 on:
   pull_request:
   push:
-    branch:
+    branches:
       - master
 
 jobs:
@@ -13,7 +13,7 @@ jobs:
       matrix:
         os:
           - ubuntu-latest
-          - windows-latest
+            #- windows-latest
             #- macos-latest
         ocaml-compiler:
           - 4.08.x
diff --git a/dune b/dune
index 1c6b45f6..575b5a09 100644
--- a/dune
+++ b/dune
@@ -1,3 +1,3 @@
 (env
  (_
-  (flags :standard -warn-error -a+8 -w +a-4-30-40-41-42-44-70 -strict-sequence)))
+  (flags :standard -warn-error -a+8 -w +a-4-30-40-41-42-44-48-70 -strict-sequence)))
diff --git a/dune-project b/dune-project
index 6d660768..ec02e11c 100644
--- a/dune-project
+++ b/dune-project
@@ -45,8 +45,9 @@
    (opentelemetry (= :version))
    (pbrt (>= 2.3))
    (odoc :with-doc)
+   (ezcurl (>= 0.2.3))
    ocurl)
-  (synopsis "Collector client for opentelemetry, using http + ocurl"))
+  (synopsis "Collector client for opentelemetry, using http + ezcurl"))
 
 (package
  (name opentelemetry-cohttp-lwt)
diff --git a/opentelemetry-client-ocurl.opam b/opentelemetry-client-ocurl.opam
index 30305dc4..1f1b4c4d 100644
--- a/opentelemetry-client-ocurl.opam
+++ b/opentelemetry-client-ocurl.opam
@@ -1,7 +1,7 @@
 # This file is generated by dune, edit dune-project instead
 opam-version: "2.0"
 version: "0.4"
-synopsis: "Collector client for opentelemetry, using http + ocurl"
+synopsis: "Collector client for opentelemetry, using http + ezcurl"
 maintainer: ["the Imandra team and contributors"]
 authors: ["the Imandra team and contributors"]
 license: "MIT"
@@ -14,6 +14,7 @@ depends: [
   "opentelemetry" {= version}
   "pbrt" {>= "2.3"}
   "odoc" {with-doc}
+  "ezcurl" {>= "0.2.3"}
   "ocurl"
 ]
 build: [
diff --git a/src/client-ocurl/b_queue.ml b/src/client-ocurl/b_queue.ml
new file mode 100644
index 00000000..21d6012f
--- /dev/null
+++ b/src/client-ocurl/b_queue.ml
@@ -0,0 +1,69 @@
+type 'a t = {
+  mutex: Mutex.t;
+  cond: Condition.t;
+  q: 'a Queue.t;
+  mutable closed: bool;
+}
+
+exception Closed
+
+let create () : _ t =
+  {
+    mutex = Mutex.create ();
+    cond = Condition.create ();
+    q = Queue.create ();
+    closed = false;
+  }
+
+let close (self : _ t) =
+  Mutex.lock self.mutex;
+  if not self.closed then (
+    self.closed <- true;
+    Condition.broadcast self.cond (* awake waiters so they fail  *)
+  );
+  Mutex.unlock self.mutex
+
+let push (self : _ t) x : unit =
+  Mutex.lock self.mutex;
+  if self.closed then (
+    Mutex.unlock self.mutex;
+    raise Closed
+  ) else (
+    Queue.push x self.q;
+    Condition.signal self.cond;
+    Mutex.unlock self.mutex
+  )
+
+let pop (self : 'a t) : 'a =
+  Mutex.lock self.mutex;
+  let rec loop () =
+    if self.closed then (
+      Mutex.unlock self.mutex;
+      raise Closed
+    ) else if Queue.is_empty self.q then (
+      Condition.wait self.cond self.mutex;
+      (loop [@tailcall]) ()
+    ) else (
+      let x = Queue.pop self.q in
+      Mutex.unlock self.mutex;
+      x
+    )
+  in
+  loop ()
+
+let pop_all (self : 'a t) into : unit =
+  Mutex.lock self.mutex;
+  let rec loop () =
+    if Queue.is_empty self.q then (
+      if self.closed then (
+        Mutex.unlock self.mutex;
+        raise Closed
+      );
+      Condition.wait self.cond self.mutex;
+      (loop [@tailcall]) ()
+    ) else (
+      Queue.transfer self.q into;
+      Mutex.unlock self.mutex
+    )
+  in
+  loop ()
diff --git a/src/client-ocurl/b_queue.mli b/src/client-ocurl/b_queue.mli
new file mode 100644
index 00000000..73bf3edc
--- /dev/null
+++ b/src/client-ocurl/b_queue.mli
@@ -0,0 +1,23 @@
+(** Basic Blocking Queue *)
+
+type 'a t
+
+val create : unit -> _ t
+
+exception Closed
+
+val push : 'a t -> 'a -> unit
+(** [push q x] pushes [x] into [q], and returns [()].
+    @raise Closed if [close q] was previously called.*)
+
+val pop : 'a t -> 'a
+(** [pop q] pops the next element in [q]. It might block until an element comes.
+   @raise Closed if the queue was closed before a new element was available. *)
+
+val pop_all : 'a t -> 'a Queue.t -> unit
+(** [pop_all q into] pops all the elements of [q]
+  and moves them into [into]. It might block until an element comes.
+   @raise Closed if the queue was closed before a new element was available. *)
+
+val close : _ t -> unit
+(** Close the queue, meaning there won't be any more [push] allowed. *)
diff --git a/src/client-ocurl/batch.ml b/src/client-ocurl/batch.ml
new file mode 100644
index 00000000..0be8b1b0
--- /dev/null
+++ b/src/client-ocurl/batch.ml
@@ -0,0 +1,24 @@
+type 'a t = {
+  mutable len: int;
+  mutable l: 'a list list;
+  mutable started: Mtime.t;
+}
+
+let create () = { len = 0; l = []; started = Mtime_clock.now () }
+
+let push self l =
+  if l != [] then (
+    if self.l == [] then self.started <- Mtime_clock.now ();
+    self.l <- l :: self.l;
+    self.len <- self.len + List.length l
+  )
+
+let[@inline] len self = self.len
+
+let[@inline] time_started self = self.started
+
+let pop_all self =
+  let l = self.l in
+  self.l <- [];
+  self.len <- 0;
+  l
diff --git a/src/client-ocurl/batch.mli b/src/client-ocurl/batch.mli
new file mode 100644
index 00000000..2b867b88
--- /dev/null
+++ b/src/client-ocurl/batch.mli
@@ -0,0 +1,14 @@
+(** List of lists with length *)
+
+type 'a t
+
+val create : unit -> 'a t
+
+val push : 'a t -> 'a list -> unit
+
+val len : _ t -> int
+
+val time_started : _ t -> Mtime.t
+(** Time at which the batch most recently became non-empty *)
+
+val pop_all : 'a t -> 'a list list
diff --git a/src/client-ocurl/common_.ml b/src/client-ocurl/common_.ml
index 592a4462..f688ff2b 100644
--- a/src/client-ocurl/common_.ml
+++ b/src/client-ocurl/common_.ml
@@ -1,10 +1,10 @@
 module Atomic = Opentelemetry_atomic.Atomic
 include Opentelemetry.Lock
 
-let[@inline] ( let@ ) f x = f x
-
 let spf = Printf.sprintf
 
+let ( let@ ) = ( @@ )
+
 let tid () = Thread.id @@ Thread.self ()
 
 let debug_ =
@@ -22,11 +22,6 @@ let get_url () = !url
 
 let set_url s = url := s
 
-(** [with_mutex m f] calls [f()] in a section where [m] is locked. *)
-let[@inline] with_mutex_ m f =
-  Mutex.lock m;
-  Fun.protect ~finally:(fun () -> Mutex.unlock m) f
-
 let parse_headers s =
   let parse_header s =
     match String.split_on_char '=' s with
diff --git a/src/client-ocurl/config.ml b/src/client-ocurl/config.ml
index b86e6305..0abca7e1 100644
--- a/src/client-ocurl/config.ml
+++ b/src/client-ocurl/config.ml
@@ -4,58 +4,23 @@ type t = {
   debug: bool;
   url: string;
   headers: (string * string) list;
-  batch_traces: int option;
-  batch_metrics: int option;
-  batch_logs: int option;
   batch_timeout_ms: int;
   bg_threads: int;
   ticker_thread: bool;
 }
 
 let pp out self =
-  let ppiopt = Format.pp_print_option Format.pp_print_int in
   let pp_header ppf (a, b) = Format.fprintf ppf "@[%s: @,%s@]@." a b in
   let ppheaders = Format.pp_print_list pp_header in
-  let {
-    debug;
-    url;
-    headers;
-    batch_traces;
-    batch_metrics;
-    batch_logs;
-    batch_timeout_ms;
-    bg_threads;
-    ticker_thread;
-  } =
+  let { debug; url; headers; batch_timeout_ms; bg_threads; ticker_thread } =
     self
   in
   Format.fprintf out
-    "{@[ debug=%B;@ url=%S;@ headers=%a;@ batch_traces=%a;@ batch_metrics=%a;@ \
-     batch_logs=%a;@ batch_timeout_ms=%d; bg_threads=%d;@ ticker_thread=%B @]}"
-    debug url ppheaders headers ppiopt batch_traces ppiopt batch_metrics ppiopt
-    batch_logs batch_timeout_ms bg_threads ticker_thread
+    "{@[ debug=%B;@ url=%S;@ headers=%a;@ batch_timeout_ms=%d; bg_threads=%d;@ \
+     ticker_thread=%B @]}"
+    debug url ppheaders headers batch_timeout_ms bg_threads ticker_thread
 
 let make ?(debug = !debug_) ?(url = get_url ()) ?(headers = get_headers ())
-    ?(batch_traces = Some 400) ?(batch_metrics = None) ?(batch_logs = Some 400)
-    ?(batch_timeout_ms = 500) ?(thread = true) ?bg_threads
-    ?(ticker_thread = true) () : t =
-  let bg_threads =
-    match bg_threads with
-    | Some n -> max n 0
-    | None ->
-      if thread then
-        4
-      else
-        0
-  in
-  {
-    debug;
-    url;
-    headers;
-    batch_traces;
-    batch_metrics;
-    batch_timeout_ms;
-    batch_logs;
-    bg_threads;
-    ticker_thread;
-  }
+    ?(batch_timeout_ms = 500) ?(bg_threads = 4) ?(ticker_thread = true) () : t =
+  let bg_threads = max 2 (min bg_threads 32) in
+  { debug; url; headers; batch_timeout_ms; bg_threads; ticker_thread }
diff --git a/src/client-ocurl/config.mli b/src/client-ocurl/config.mli
index 957dba13..17a7711e 100644
--- a/src/client-ocurl/config.mli
+++ b/src/client-ocurl/config.mli
@@ -1,3 +1,5 @@
+(** Configuration for the ocurl backend *)
+
 type t = private {
   debug: bool;
   url: string;
@@ -6,23 +8,6 @@ type t = private {
   headers: (string * string) list;
       (** API headers sent to the endpoint. Default is none or
       "OTEL_EXPORTER_OTLP_HEADERS" if set. *)
-  batch_traces: int option;
-      (** Batch traces? If [Some i], then this produces batches of (at most)
-      [i] items. If [None], there is no batching.
-
-      Note that traces and metrics are batched separately.
-      Default [Some 400].
-  *)
-  batch_metrics: int option;
-      (** Batch metrics? If [Some i], then this produces batches of (at most)
-      [i] items. If [None], there is no batching.
-
-      Note that traces and metrics are batched separately.
-      Default [None].
-  *)
-  batch_logs: int option;
-      (** Batch logs? See {!batch_metrics} for details.
-      Default [Some 400] *)
   batch_timeout_ms: int;
       (** Number of milliseconds after which we will emit a batch, even
       incomplete.
@@ -31,10 +16,8 @@ type t = private {
   bg_threads: int;
       (** Are there background threads, and how many? Default [4] *)
   ticker_thread: bool;
-      (** Is there a ticker thread? Default [true].
-      This thread will regularly call [tick()] on the backend, to make
-      sure it makes progress, and regularly send events to the collector.
-      This option is ignored if [bg_threads=0]. *)
+      (** If true, start a thread that regularly checks if signals should
+          be sent to the collector. Default [true] *)
 }
 (** Configuration.
 
@@ -45,20 +28,12 @@ val make :
   ?debug:bool ->
   ?url:string ->
   ?headers:(string * string) list ->
-  ?batch_traces:int option ->
-  ?batch_metrics:int option ->
-  ?batch_logs:int option ->
   ?batch_timeout_ms:int ->
-  ?thread:bool ->
   ?bg_threads:int ->
   ?ticker_thread:bool ->
   unit ->
   t
 (** Make a configuration.
-
-   @param thread if true and [bg_threads] is not provided, we will pick a number
-   of bg threads. Otherwise the number of [bg_threads] superseeds this option.
-
  *)
 
 val pp : Format.formatter -> t -> unit
diff --git a/src/client-ocurl/dune b/src/client-ocurl/dune
index 8e5c873f..a8162b4a 100644
--- a/src/client-ocurl/dune
+++ b/src/client-ocurl/dune
@@ -2,4 +2,4 @@
  (name opentelemetry_client_ocurl)
  (public_name opentelemetry-client-ocurl)
  (libraries opentelemetry opentelemetry.atomic curl pbrt threads mtime
-   mtime.clock.os))
+   mtime.clock.os ezcurl ezcurl.core))
diff --git a/src/client-ocurl/opentelemetry_client_ocurl.ml b/src/client-ocurl/opentelemetry_client_ocurl.ml
index cb30f1ff..e0d6a596 100644
--- a/src/client-ocurl/opentelemetry_client_ocurl.ml
+++ b/src/client-ocurl/opentelemetry_client_ocurl.ml
@@ -4,6 +4,7 @@
  *)
 
 module OT = Opentelemetry
+module Config = Config
 open Opentelemetry
 include Common_
 
@@ -13,15 +14,15 @@ let last_gc_metrics = Atomic.make (Mtime_clock.now ())
 
 let timeout_gc_metrics = Mtime.Span.(20 * s)
 
+(** side channel for GC, appended to metrics batch data *)
 let gc_metrics = AList.make ()
-(* side channel for GC, appended to {!E_metrics}'s data *)
 
-(* capture current GC metrics if {!needs_gc_metrics} is true
-   or it has been a long time since the last GC metrics collection,
-   and push them into {!gc_metrics} for later collection *)
+(** capture current GC metrics if {!needs_gc_metrics} is true
+    or it has been a long time since the last GC metrics collection,
+    and push them into {!gc_metrics} for later collection *)
 let sample_gc_metrics_if_needed () =
   let now = Mtime_clock.now () in
-  let alarm = Atomic.compare_and_set needs_gc_metrics true false in
+  let alarm = Atomic.exchange needs_gc_metrics false in
   let timeout () =
     let elapsed = Mtime.span now (Atomic.get last_gc_metrics) in
     Mtime.Span.compare elapsed timeout_gc_metrics > 0
@@ -36,663 +37,404 @@ let sample_gc_metrics_if_needed () =
     AList.add gc_metrics l
   )
 
-module Config = Config
-
-let _init_curl =
-  lazy
-    (Curl.global_init Curl.CURLINIT_GLOBALALL;
-     at_exit Curl.global_cleanup)
-
-type error =
-  [ `Status of int * Opentelemetry.Proto.Status.status
-  | `Failure of string
-  | `Sysbreak
-  ]
-
 let n_errors = Atomic.make 0
 
 let n_dropped = Atomic.make 0
 
-let report_err_ = function
-  | `Sysbreak -> Printf.eprintf "opentelemetry: ctrl-c captured, stopping\n%!"
-  | `Failure msg ->
-    Format.eprintf "@[<2>opentelemetry: export failed: %s@]@." msg
-  | `Status (code, { Opentelemetry.Proto.Status.code = scode; message; details })
-    ->
-    let pp_details out l =
-      List.iter
-        (fun s -> Format.fprintf out "%S;@ " (Bytes.unsafe_to_string s))
-        l
-    in
-    Format.eprintf
-      "@[<2>opentelemetry: export failed with@ http code=%d@ status \
-       {@[code=%ld;@ message=%S;@ details=[@[%a@]]@]}@]@."
-      code scode
-      (Bytes.unsafe_to_string message)
-      pp_details details
-
-module Httpc : sig
-  type t
-
-  val create : unit -> t
-
-  val send :
-    t ->
-    path:string ->
-    decode:[ `Dec of Pbrt.Decoder.t -> 'a | `Ret of 'a ] ->
-    string ->
-    ('a, error) result
-
-  val cleanup : t -> unit
-end = struct
+(** Something sent to the collector *)
+module Event = struct
   open Opentelemetry.Proto
 
-  let () = Lazy.force _init_curl
-
-  (* TODO: use Curl.Multi, etc. instead? *)
-  type t = {
-    buf_res: Buffer.t;
-    curl: Curl.t;
-  }
-
-  let create () : t = { buf_res = Buffer.create 256; curl = Curl.init () }
-
-  let cleanup self = Curl.cleanup self.curl
-
-  (* send the content to the remote endpoint/path *)
-  let send (self : t) ~path ~decode (bod : string) : ('a, error) result =
-    let { curl; buf_res } = self in
-    Curl.reset curl;
-    if !debug_ then Curl.set_verbose curl true;
-    let full_url = !url ^ path in
-    Curl.set_url curl full_url;
-    Curl.set_httppost curl [];
-    let to_http_header (k, v) = Printf.sprintf "%s: %s" k v in
-    let http_headers = List.map to_http_header !headers in
-    Curl.set_httpheader curl
-      ("Content-Type: application/x-protobuf" :: http_headers);
-    (* write body *)
-    Curl.set_post curl true;
-    Curl.set_postfieldsize curl (String.length bod);
-    Curl.set_readfunction curl
-      (let i = ref 0 in
-       fun n ->
-         if !debug_ then Printf.eprintf "curl asks for %d bytes\n%!" n;
-         let len = min n (String.length bod - !i) in
-         let s = String.sub bod !i len in
-         if !debug_ then Printf.eprintf "gave curl %d bytes\n%!" len;
-         i := !i + len;
-         s);
-    (* read result's body *)
-    Buffer.clear buf_res;
-    Curl.set_writefunction curl (fun s ->
-        Buffer.add_string buf_res s;
-        String.length s);
-    try
-      match Curl.perform curl with
-      | () ->
-        let code = Curl.get_responsecode curl in
-        if !debug_ then
-          Printf.eprintf "result body: %S\n%!" (Buffer.contents buf_res);
-        if code >= 200 && code < 300 then (
-          match decode with
-          | `Ret x -> Ok x
-          | `Dec f ->
-            let dec = Pbrt.Decoder.of_string (Buffer.contents buf_res) in
-            (try Ok (f dec)
-             with e ->
-               let bt = Printexc.get_backtrace () in
-               Error
-                 (`Failure
-                   (spf "decoding failed with:\n%s\n%s" (Printexc.to_string e)
-                      bt)))
-        ) else (
-          let str = Buffer.contents buf_res in
-          let dec = Pbrt.Decoder.of_string str in
-
-          try
-            let status = Status.decode_status dec in
-            Error (`Status (code, status))
-          with e ->
-            let bt = Printexc.get_backtrace () in
-            Error
-              (`Failure
-                (spf
-                   "httpc: decoding of status (url=%S, code=%d) failed with:\n\
-                    %s\n\
-                    status: %S\n\
-                    %s"
-                   full_url code (Printexc.to_string e) str bt))
-        )
-      | exception Sys.Break -> Error `Sysbreak
-      | exception Curl.CurlException (_, code, msg) ->
-        let status =
-          Status.default_status ~code:(Int32.of_int code)
-            ~message:(Bytes.unsafe_of_string msg)
-            ()
-        in
-        Error (`Status (code, status))
-    with
-    | Sys.Break -> Error `Sysbreak
-    | e ->
-      let bt = Printexc.get_backtrace () in
-      Error
-        (`Failure
-          (spf "httpc: post on url=%S failed with:\n%s\n%s" full_url
-             (Printexc.to_string e) bt))
+  type t =
+    | E_metric of Metrics.resource_metrics list
+    | E_trace of Trace.resource_spans list
+    | E_logs of Logs.resource_logs list
+    | E_tick
+    | E_flush_all  (** Flush all batches *)
 end
 
-(** Batch of resources to be pushed later.
-
-    This type is thread-safe. *)
-module Batch : sig
-  type 'a t
-
-  val push : 'a t -> 'a -> bool
-  (** [push batch x] pushes [x] into the batch, and heuristically
-      returns [true] if the batch is ready to be emitted (to know if we should
-      wake up the sending thread, if any) *)
-
-  val push' : 'a t -> 'a -> unit
-
-  val is_ready : now:Mtime.t -> _ t -> bool
-  (** is the batch ready to be sent? This is heuristic. *)
-
-  val pop_if_ready : ?force:bool -> now:Mtime.t -> 'a t -> 'a list option
-  (** Is the batch ready to be emitted? If batching is disabled,
-      this is true as soon as {!is_empty} is false. If a timeout is provided
-      for this batch, then it will be ready if an element has been in it
-      for at least the timeout.
-      @param now passed to implement timeout *)
-
-  val make : ?batch:int -> ?timeout:Mtime.span -> unit -> 'a t
-  (** Create a new batch *)
-end = struct
-  type 'a t = {
-    lock: Mutex.t;
-    mutable size: int;
-    mutable q: 'a list;
-    batch: int option;
-    high_watermark: int;
-    timeout: Mtime.span option;
-    mutable start: Mtime.t;
-  }
-
-  let make ?batch ?timeout () : _ t =
-    Option.iter (fun b -> assert (b > 0)) batch;
-    let high_watermark = Option.fold ~none:100 ~some:(fun x -> x * 10) batch in
-    {
-      lock = Mutex.create ();
-      size = 0;
-      start = Mtime_clock.now ();
-      q = [];
-      batch;
-      timeout;
-      high_watermark;
-    }
-
-  let timeout_expired_ ~now self : bool =
-    match self.timeout with
-    | Some t ->
-      let elapsed = Mtime.span now self.start in
-      Mtime.Span.compare elapsed t >= 0
-    | None -> false
-
-  let is_full_ self : bool =
-    match self.batch with
-    | None -> self.size > 0
-    | Some b -> self.size >= b
-
-  let is_ready ~now self : bool =
-    let@ () = with_mutex_ self.lock in
-    is_full_ self || timeout_expired_ ~now self
-
-  let pop_if_ready ?(force = false) ~now (self : _ t) : _ list option =
-    let@ () = with_mutex_ self.lock in
-    if self.size > 0 && (force || is_full_ self || timeout_expired_ ~now self)
-    then (
-      let l = self.q in
-      self.q <- [];
-      self.size <- 0;
-      assert (l <> []);
-      Some l
-    ) else
-      None
-
-  let push (self : _ t) x : bool =
-    let@ () = with_mutex_ self.lock in
-    if self.size >= self.high_watermark then (
-      (* drop this to prevent queue from growing too fast *)
-      Atomic.incr n_dropped;
-      true
-    ) else (
-      if self.size = 0 && Option.is_some self.timeout then
-        (* current batch starts now *)
-        self.start <- Mtime_clock.now ();
-
-      (* add to queue *)
-      self.size <- 1 + self.size;
-      self.q <- x :: self.q;
-      let ready = is_full_ self in
-      ready
-    )
-
-  let push' self x = ignore (push self x : bool)
-end
-
-(** An emitter. This is used by {!Backend} below to forward traces/metrics/…
-    from the program to whatever collector client we have. *)
-module type EMITTER = sig
+(** Something to be sent via HTTP *)
+module To_send = struct
   open Opentelemetry.Proto
 
-  val push_trace : Trace.resource_spans list -> unit
-
-  val push_metrics : Metrics.resource_metrics list -> unit
-
-  val push_logs : Logs.resource_logs list -> unit
-
-  val set_on_tick_callbacks : (unit -> unit) list ref -> unit
-
-  val tick : unit -> unit
-
-  val cleanup : unit -> unit
+  type t =
+    | Send_metric of Metrics.resource_metrics list list
+    | Send_trace of Trace.resource_spans list list
+    | Send_logs of Logs.resource_logs list list
 end
 
-(* start a thread in the background, running [f()] *)
-let start_bg_thread (f : unit -> unit) : unit =
+(** start a thread in the background, running [f()] *)
+let start_bg_thread (f : unit -> unit) : Thread.t =
   let run () =
     (* block some signals: USR1 USR2 TERM PIPE ALARM STOP, see [$ kill -L] *)
     ignore (Thread.sigmask Unix.SIG_BLOCK [ 10; 12; 13; 14; 15; 19 ] : _ list);
     f ()
   in
-  ignore (Thread.create run () : Thread.t)
+  Thread.create run ()
 
-(* make an emitter.
+let str_to_hex (s : string) : string =
+  let i_to_hex (i : int) =
+    if i < 10 then
+      Char.chr (i + Char.code '0')
+    else
+      Char.chr (i - 10 + Char.code 'a')
+  in
 
-   exceptions inside should be caught, see
-   https://opentelemetry.io/docs/reference/specification/error-handling/ *)
-let mk_emitter ~stop ~(config : Config.t) () : (module EMITTER) =
-  let open Proto in
-  (* local helpers *)
-  let open struct
-    let timeout =
-      if config.batch_timeout_ms > 0 then
-        Some Mtime.Span.(config.batch_timeout_ms * ms)
-      else
-        None
+  let res = Bytes.create (2 * String.length s) in
+  for i = 0 to String.length s - 1 do
+    let n = Char.code (String.get s i) in
+    Bytes.set res (2 * i) (i_to_hex ((n land 0xf0) lsr 4));
+    Bytes.set res ((2 * i) + 1) (i_to_hex (n land 0x0f))
+  done;
+  Bytes.unsafe_to_string res
 
-    let batch_traces : Trace.resource_spans list Batch.t =
-      Batch.make ?batch:config.batch_traces ?timeout ()
+module Backend_impl : sig
+  type t
 
-    let batch_metrics : Metrics.resource_metrics list Batch.t =
-      Batch.make ?batch:config.batch_metrics ?timeout ()
+  val create : stop:bool Atomic.t -> config:Config.t -> unit -> t
 
-    let batch_logs : Logs.resource_logs list Batch.t =
-      Batch.make ?batch:config.batch_logs ?timeout ()
+  val send_event : t -> Event.t -> unit
+
+  val shutdown : t -> unit
+end = struct
+  open Opentelemetry.Proto
+
+  type t = {
+    stop: bool Atomic.t;
+    cleaned: bool Atomic.t;  (** True when we cleaned up after closing *)
+    config: Config.t;
+    q: Event.t B_queue.t;  (** Queue to receive data from the user's code *)
+    mutable main_th: Thread.t option;  (** Thread that listens on [q] *)
+    send_q: To_send.t B_queue.t;  (** Queue for the send worker threads *)
+    mutable send_threads: Thread.t array;  (** Threads that send data via http *)
+  }
+
+  let send_http_ ~stop ~config (client : Curl.t) encoder ~path ~encode x : unit
+      =
+    Pbrt.Encoder.reset encoder;
+    encode x encoder;
+    let data = Pbrt.Encoder.to_string encoder in
+    let url = config.Config.url ^ path in
+    if !debug_ || config.debug then
+      Printf.eprintf "opentelemetry: send http POST to %s (%dB)\n%!" url
+        (String.length data);
+    let headers =
+      ("Content-Type", "application/x-protobuf") :: config.headers
+    in
+    match
+      Ezcurl.post ~headers ~client ~params:[] ~url ~content:(`String data) ()
+    with
+    | Ok { code; _ } when code >= 200 && code < 300 -> ()
+    | Ok { code; body; headers = _; info = _ } ->
+      Atomic.incr n_errors;
+      if !debug_ || config.debug then (
+        let dec = Pbrt.Decoder.of_string body in
+        let body =
+          try
+            let status = Status.decode_status dec in
+            Format.asprintf "%a" Status.pp_status status
+          with _ ->
+            spf "(could not decode status)\nraw bytes: %s" (str_to_hex body)
+        in
+        Printf.eprintf "error while sending:\n  code=%d\n  %s\n%!" code body
+      );
+      ()
+    | exception Sys.Break ->
+      Printf.eprintf "ctrl-c captured, stopping\n%!";
+      Atomic.set stop true
+    | Error (code, msg) ->
+      (* TODO: log error _via_ otel? *)
+      Atomic.incr n_errors;
+
+      Printf.eprintf "opentelemetry: export failed:\n  %s\n  curl code: %s\n%!"
+        msg (Curl.strerror code);
+
+      (* avoid crazy error loop *)
+      Thread.delay 3.
+
+  let send_logs_http ~stop ~config (client : Curl.t) encoder
+      (l : Logs.resource_logs list list) : unit =
+    let l = List.fold_left (fun acc l -> List.rev_append l acc) [] l in
+    let x =
+      Logs_service.default_export_logs_service_request ~resource_logs:l ()
+    in
+    send_http_ ~stop ~config client encoder ~path:"/v1/logs"
+      ~encode:Logs_service.encode_export_logs_service_request x
+
+  let send_metrics_http ~stop ~config curl encoder
+      (l : Metrics.resource_metrics list list) : unit =
+    let l = List.fold_left (fun acc l -> List.rev_append l acc) [] l in
+    let x =
+      Metrics_service.default_export_metrics_service_request ~resource_metrics:l
+        ()
+    in
+    send_http_ ~stop ~config curl encoder ~path:"/v1/metrics"
+      ~encode:Metrics_service.encode_export_metrics_service_request x
+
+  let send_traces_http ~stop ~config curl encoder
+      (l : Trace.resource_spans list list) : unit =
+    let l = List.fold_left (fun acc l -> List.rev_append l acc) [] l in
+    let x =
+      Trace_service.default_export_trace_service_request ~resource_spans:l ()
+    in
+    send_http_ ~stop ~config curl encoder ~path:"/v1/traces"
+      ~encode:Trace_service.encode_export_trace_service_request x
+
+  let[@inline] send_event (self : t) ev : unit = B_queue.push self.q ev
+
+  (** Thread that, in a loop, reads from [q] to get the
+      next message to send via http *)
+  let bg_thread_loop (self : t) : unit =
+    Ezcurl.with_client ?set_opts:None @@ fun client ->
+    let stop = self.stop in
+    let config = self.config in
+    let encoder = Pbrt.Encoder.create () in
+    try
+      while not (Atomic.get stop) do
+        let msg = B_queue.pop self.send_q in
+        match msg with
+        | To_send.Send_trace tr ->
+          send_traces_http ~stop ~config client encoder tr
+        | To_send.Send_metric ms ->
+          send_metrics_http ~stop ~config client encoder ms
+        | To_send.Send_logs logs ->
+          send_logs_http ~stop ~config client encoder logs
+      done
+    with B_queue.Closed -> ()
+
+  type batches = {
+    traces: Proto.Trace.resource_spans Batch.t;
+    logs: Proto.Logs.resource_logs Batch.t;
+    metrics: Proto.Metrics.resource_metrics Batch.t;
+  }
+
+  let batch_max_size_ = 200
+
+  let should_send_batch_ ~config ~now (b : _ Batch.t) : bool =
+    Batch.len b > 0
+    && (Batch.len b >= batch_max_size_
+       ||
+       let timeout = Mtime.Span.(config.Config.batch_timeout_ms * ms) in
+       let elapsed = Mtime.span now (Batch.time_started b) in
+       Mtime.Span.compare elapsed timeout >= 0)
+
+  let main_thread_loop (self : t) : unit =
+    let local_q = Queue.create () in
+    let config = self.config in
+
+    (* keep track of batches *)
+    let batches =
+      {
+        traces = Batch.create ();
+        logs = Batch.create ();
+        metrics = Batch.create ();
+      }
+    in
+
+    let send_metrics () =
+      B_queue.push self.send_q
+        (To_send.Send_metric (Batch.pop_all batches.metrics))
+    in
+
+    let send_logs () =
+      B_queue.push self.send_q (To_send.Send_logs (Batch.pop_all batches.logs))
+    in
+
+    let send_traces () =
+      B_queue.push self.send_q
+        (To_send.Send_trace (Batch.pop_all batches.traces))
+    in
+
+    try
+      while not (Atomic.get self.stop) do
+        (* read multiple events at once *)
+        B_queue.pop_all self.q local_q;
+
+        let now = Mtime_clock.now () in
+
+        (* how to process a single event *)
+        let process_ev (ev : Event.t) : unit =
+          match ev with
+          | Event.E_metric m ->
+            Batch.push batches.metrics m;
+            if should_send_batch_ ~config ~now batches.metrics then
+              send_metrics ()
+          | Event.E_trace tr ->
+            Batch.push batches.traces tr;
+            if should_send_batch_ ~config ~now batches.traces then
+              send_traces ()
+          | Event.E_logs logs ->
+            Batch.push batches.logs logs;
+            if should_send_batch_ ~config ~now batches.logs then send_logs ()
+          | Event.E_tick ->
+            (* check for batches whose timeout expired *)
+            if should_send_batch_ ~config ~now batches.metrics then
+              send_metrics ();
+            if should_send_batch_ ~config ~now batches.logs then send_logs ();
+            if should_send_batch_ ~config ~now batches.traces then
+              send_traces ()
+          | Event.E_flush_all ->
+            if Batch.len batches.metrics > 0 then send_metrics ();
+            if Batch.len batches.logs > 0 then send_logs ();
+            if Batch.len batches.traces > 0 then send_traces ()
+        in
+
+        while not (Queue.is_empty local_q) do
+          let ev = Queue.pop local_q in
+          process_ev ev
+        done
+      done
+    with B_queue.Closed -> ()
+
+  let create ~stop ~config () : t =
+    let n_send_threads = max 2 config.Config.bg_threads in
+    let self =
+      {
+        stop;
+        config;
+        q = B_queue.create ();
+        send_threads = [||];
+        send_q = B_queue.create ();
+        cleaned = Atomic.make false;
+        main_th = None;
+      }
+    in
+
+    let main_th = start_bg_thread (fun () -> main_thread_loop self) in
+    self.main_th <- Some main_th;
+
+    self.send_threads <-
+      Array.init n_send_threads (fun _i ->
+          start_bg_thread (fun () -> bg_thread_loop self));
+
+    self
+
+  let shutdown self : unit =
+    Atomic.set self.stop true;
+    if not (Atomic.exchange self.cleaned true) then (
+      (* empty batches *)
+      send_event self Event.E_flush_all;
+      (* close the incoming queue, wait for the thread to finish
+         before we start cutting off the background threads, so that they
+         have time to receive the final batches *)
+      B_queue.close self.q;
+      Option.iter Thread.join self.main_th;
+      (* close send queues, then wait for all threads *)
+      B_queue.close self.send_q;
+      Array.iter Thread.join self.send_threads
+    )
+end
+
+let mk_backend ~stop ~config () : (module Collector.BACKEND) =
+  let module M = struct
+    open Opentelemetry.Proto
+    open Opentelemetry.Collector
+
+    let backend = Backend_impl.create ~stop ~config ()
+
+    let send_trace : Trace.resource_spans list sender =
+      {
+        send =
+          (fun l ~ret ->
+            Backend_impl.send_event backend (Event.E_trace l);
+            ret ());
+      }
+
+    let last_sent_metrics = Atomic.make (Mtime_clock.now ())
+
+    (* send metrics from time to time *)
+    let timeout_sent_metrics = Mtime.Span.(5 * s)
+
+    let signal_emit_gc_metrics () =
+      if !debug_ || config.debug then
+        Printf.eprintf "opentelemetry: emit GC metrics requested\n%!";
+      Atomic.set needs_gc_metrics true
+
+    let additional_metrics () : Metrics.resource_metrics list =
+      (* add exporter metrics to the lot? *)
+      let last_emit = Atomic.get last_sent_metrics in
+      let now = Mtime_clock.now () in
+      let add_own_metrics =
+        let elapsed = Mtime.span last_emit now in
+        Mtime.Span.compare elapsed timeout_sent_metrics > 0
+      in
+
+      (* there is a possible race condition here, as several threads might update
+         metrics at the same time. But that's harmless. *)
+      if add_own_metrics then (
+        Atomic.set last_sent_metrics now;
+        let open OT.Metrics in
+        [
+          make_resource_metrics
+            [
+              sum ~name:"otel.export.dropped" ~is_monotonic:true
+                [
+                  int
+                    ~start_time_unix_nano:(Mtime.to_uint64_ns last_emit)
+                    ~now:(Mtime.to_uint64_ns now) (Atomic.get n_dropped);
+                ];
+              sum ~name:"otel.export.errors" ~is_monotonic:true
+                [
+                  int
+                    ~start_time_unix_nano:(Mtime.to_uint64_ns last_emit)
+                    ~now:(Mtime.to_uint64_ns now) (Atomic.get n_errors);
+                ];
+            ];
+        ]
+      ) else
+        []
+
+    let send_metrics : Metrics.resource_metrics list sender =
+      {
+        send =
+          (fun m ~ret ->
+            let m = List.rev_append (additional_metrics ()) m in
+            Backend_impl.send_event backend (Event.E_metric m);
+            ret ());
+      }
+
+    let send_logs : Logs.resource_logs list sender =
+      {
+        send =
+          (fun m ~ret ->
+            Backend_impl.send_event backend (Event.E_logs m);
+            ret ());
+      }
 
     let on_tick_cbs_ = Atomic.make (ref [])
 
     let set_on_tick_callbacks = Atomic.set on_tick_cbs_
 
-    let send_http_ (httpc : Httpc.t) encoder ~path ~encode x : unit =
-      Pbrt.Encoder.reset encoder;
-      encode x encoder;
-      let data = Pbrt.Encoder.to_string encoder in
-      match Httpc.send httpc ~path ~decode:(`Ret ()) data with
-      | Ok () -> ()
-      | Error `Sysbreak ->
-        Printf.eprintf "ctrl-c captured, stopping\n%!";
-        Atomic.set stop true
-      | Error err ->
-        (* TODO: log error _via_ otel? *)
-        Atomic.incr n_errors;
-        report_err_ err;
-        (* avoid crazy error loop *)
-        Thread.delay 3.
-
-    let send_metrics_http curl encoder (l : Metrics.resource_metrics list list)
-        =
-      let l = List.fold_left (fun acc l -> List.rev_append l acc) [] l in
-      let x =
-        Metrics_service.default_export_metrics_service_request
-          ~resource_metrics:l ()
-      in
-      send_http_ curl encoder ~path:"/v1/metrics"
-        ~encode:Metrics_service.encode_export_metrics_service_request x
-
-    let send_traces_http curl encoder (l : Trace.resource_spans list list) =
-      let l = List.fold_left (fun acc l -> List.rev_append l acc) [] l in
-      let x =
-        Trace_service.default_export_trace_service_request ~resource_spans:l ()
-      in
-      send_http_ curl encoder ~path:"/v1/traces"
-        ~encode:Trace_service.encode_export_trace_service_request x
-
-    let send_logs_http curl encoder (l : Logs.resource_logs list list) =
-      let l = List.fold_left (fun acc l -> List.rev_append l acc) [] l in
-      let x =
-        Logs_service.default_export_logs_service_request ~resource_logs:l ()
-      in
-      send_http_ curl encoder ~path:"/v1/logs"
-        ~encode:Logs_service.encode_export_logs_service_request x
-
-    (* emit metrics, if the batch is full or timeout lapsed *)
-    let emit_metrics_maybe ~now ?force httpc encoder : bool =
-      match Batch.pop_if_ready ?force ~now batch_metrics with
-      | None -> false
-      | Some l ->
-        let batch = AList.pop_all gc_metrics :: l in
-        send_metrics_http httpc encoder batch;
-        true
-
-    let emit_traces_maybe ~now ?force httpc encoder : bool =
-      match Batch.pop_if_ready ?force ~now batch_traces with
-      | None -> false
-      | Some l ->
-        send_traces_http httpc encoder l;
-        true
-
-    let emit_logs_maybe ~now ?force httpc encoder : bool =
-      match Batch.pop_if_ready ?force ~now batch_logs with
-      | None -> false
-      | Some l ->
-        send_logs_http httpc encoder l;
-        true
-
-    let[@inline] guard_exn_ where f =
-      try f ()
-      with e ->
-        let bt = Printexc.get_backtrace () in
-        Printf.eprintf
-          "opentelemetry-curl: uncaught exception in %s: %s\n%s\n%!" where
-          (Printexc.to_string e) bt
-
-    let emit_all_force (httpc : Httpc.t) encoder =
-      let now = Mtime_clock.now () in
-      ignore (emit_traces_maybe ~now ~force:true httpc encoder : bool);
-      ignore (emit_logs_maybe ~now ~force:true httpc encoder : bool);
-      ignore (emit_metrics_maybe ~now ~force:true httpc encoder : bool)
-
-    let tick_common_ () =
-      if !debug_ then Printf.eprintf "tick (from %d)\n%!" (tid ());
-      sample_gc_metrics_if_needed ();
-      List.iter
-        (fun f ->
-          try f ()
-          with e ->
-            Printf.eprintf "on tick callback raised: %s\n"
-              (Printexc.to_string e))
-        !(Atomic.get on_tick_cbs_);
-      ()
-
-    let setup_ticker_thread ~tick ~finally () =
-      (* thread that calls [tick()] regularly, to help enforce timeouts *)
-      let tick_thread () =
-        let@ () =
-          Fun.protect ~finally:(fun () ->
-              Atomic.set stop true;
-              finally ())
-        in
-        while not @@ Atomic.get stop do
-          Thread.delay 0.5;
-          tick ()
-        done
-      in
-      start_bg_thread tick_thread
-  end in
-  (* setup a global lock *)
-  (let global_lock_ = Mutex.create () in
-   Lock.set_mutex
-     ~lock:(fun () -> Mutex.lock global_lock_)
-     ~unlock:(fun () -> Mutex.unlock global_lock_));
-
-  if config.bg_threads > 0 then (
-    (* lock+condition used for background threads to wait, and be woken up
-       when a batch is ready *)
-    let m = Mutex.create () in
-    let cond = Condition.create () in
-
-    (* loop for the thread that processes events and sends them to collector *)
-    let bg_thread () =
-      let httpc = Httpc.create () in
-      let encoder = Pbrt.Encoder.create () in
-      while not @@ Atomic.get stop do
-        let@ () = guard_exn_ (spf "bg thread[%d] (main loop)" @@ tid ()) in
-
-        let now = Mtime_clock.now () in
-        let do_metrics = emit_metrics_maybe ~now httpc encoder in
-        let do_traces = emit_traces_maybe ~now httpc encoder in
-        let do_logs = emit_logs_maybe ~now httpc encoder in
-        if (not do_metrics) && (not do_traces) && not do_logs then (
-          let@ () = guard_exn_ (spf "bg thread[%d] (waiting)" @@ tid ()) in
-          (* wait for something to happen *)
-          Mutex.lock m;
-          Condition.wait cond m;
-          Mutex.unlock m
-        )
-      done;
-      (* flush remaining events once we exit *)
-      let@ () = guard_exn_ "bg thread (cleanup)" in
-      emit_all_force httpc encoder;
-      Httpc.cleanup httpc
-    in
-
-    for _i = 1 to config.bg_threads do
-      start_bg_thread bg_thread
-    done;
-
-    (* if the bg thread waits, this will wake it up so it can send batches *)
-    let wakeup ~all () =
-      with_mutex_ m (fun () ->
-          if all then
-            Condition.broadcast cond
-          else
-            Condition.signal cond);
-      Thread.yield ()
-    in
-
     let tick () =
-      tick_common_ ();
+      sample_gc_metrics_if_needed ();
+      Backend_impl.send_event backend Event.E_tick;
+      let l = Atomic.get on_tick_cbs_ in
+      List.iter (fun f -> f ()) !l
 
-      let now = Mtime_clock.now () in
-      if Atomic.get stop then
-        wakeup ~all:true ()
-      else if
-        Batch.is_ready ~now batch_metrics
-        || Batch.is_ready ~now batch_traces
-        || Batch.is_ready ~now batch_logs
-      then
-        wakeup ~all:false ()
-    in
+    let cleanup () = Backend_impl.shutdown backend
+  end in
+  (module M)
 
-    if config.ticker_thread then
-      setup_ticker_thread ~tick ~finally:(fun () -> wakeup ~all:true ()) ();
-
-    let module M = struct
-      let push_trace e = if Batch.push batch_traces e then wakeup ~all:false ()
-
-      let push_metrics e =
-        if Batch.push batch_metrics e then wakeup ~all:false ()
-
-      let push_logs e = if Batch.push batch_logs e then wakeup ~all:false ()
-
-      let set_on_tick_callbacks = set_on_tick_callbacks
-
-      let tick = tick
-
-      let cleanup () =
-        Atomic.set stop true;
-        if !debug_ then Printf.eprintf "opentelemetry: exiting…\n%!";
-        wakeup ~all:true ()
-    end in
-    (module M)
-  ) else (
-    let httpc = Httpc.create () in
-    let encoder = Pbrt.Encoder.create () in
-
-    let module M = struct
-      (* we make sure that this is thread-safe, even though we don't have a
-         background thread. There can still be a ticker thread, and there
-         can also be several user threads that produce spans and call
-         the emit functions. *)
-
-      let push_trace e =
-        let@ () = guard_exn_ "push trace" in
-        Batch.push' batch_traces e;
-        let now = Mtime_clock.now () in
-        let@ () = Lock.with_lock in
-        ignore (emit_traces_maybe ~now httpc encoder : bool)
-
-      let push_metrics e =
-        let@ () = guard_exn_ "push metrics" in
-        sample_gc_metrics_if_needed ();
-        Batch.push' batch_metrics e;
-        let now = Mtime_clock.now () in
-        let@ () = Lock.with_lock in
-        ignore (emit_metrics_maybe ~now httpc encoder : bool)
-
-      let push_logs e =
-        let@ () = guard_exn_ "push logs" in
-        Batch.push' batch_logs e;
-        let now = Mtime_clock.now () in
-        let@ () = Lock.with_lock in
-        ignore (emit_logs_maybe ~now httpc encoder : bool)
-
-      let set_on_tick_callbacks = set_on_tick_callbacks
-
-      let tick () =
-        tick_common_ ();
-        sample_gc_metrics_if_needed ();
-        let@ () = Lock.with_lock in
-        let now = Mtime_clock.now () in
-        ignore (emit_traces_maybe ~now httpc encoder : bool);
-        ignore (emit_metrics_maybe ~now httpc encoder : bool);
-        ignore (emit_logs_maybe ~now httpc encoder : bool);
-        ()
-
-      (* make sure we have a ticker thread, if required  *)
-      let () =
-        if config.ticker_thread then
-          setup_ticker_thread ~tick ~finally:ignore ()
-
-      let cleanup () =
-        if !debug_ then Printf.eprintf "opentelemetry: exiting…\n%!";
-        emit_all_force httpc encoder;
-        Httpc.cleanup httpc
-    end in
-    (module M)
-  )
-
-module Backend (Arg : sig
-  val stop : bool Atomic.t
-
-  val config : Config.t
-end)
-() : Opentelemetry.Collector.BACKEND = struct
-  include (val mk_emitter ~stop:Arg.stop ~config:Arg.config ())
-
-  open Opentelemetry.Proto
-  open Opentelemetry.Collector
-
-  let send_trace : Trace.resource_spans list sender =
-    {
-      send =
-        (fun l ~ret ->
-          (if !debug_ then
-            let@ () = Lock.with_lock in
-            Format.eprintf "send spans %a@."
-              (Format.pp_print_list Trace.pp_resource_spans)
-              l);
-          push_trace l;
-          ret ());
-    }
-
-  let last_sent_metrics = Atomic.make (Mtime_clock.now ())
-
-  let timeout_sent_metrics = Mtime.Span.(5 * s)
-  (* send metrics from time to time *)
-
-  let signal_emit_gc_metrics () =
-    if !debug_ then
-      Printf.eprintf "opentelemetry: emit GC metrics requested\n%!";
-    Atomic.set needs_gc_metrics true
-
-  let additional_metrics () : Metrics.resource_metrics list =
-    (* add exporter metrics to the lot? *)
-    let last_emit = Atomic.get last_sent_metrics in
-    let now = Mtime_clock.now () in
-    let add_own_metrics =
-      let elapsed = Mtime.span last_emit now in
-      Mtime.Span.compare elapsed timeout_sent_metrics > 0
-    in
-
-    (* there is a possible race condition here, as several threads might update
-       metrics at the same time. But that's harmless. *)
-    if add_own_metrics then (
-      Atomic.set last_sent_metrics now;
-      let open OT.Metrics in
-      [
-        make_resource_metrics
-          [
-            sum ~name:"otel.export.dropped" ~is_monotonic:true
-              [
-                int
-                  ~start_time_unix_nano:(Mtime.to_uint64_ns last_emit)
-                  ~now:(Mtime.to_uint64_ns now) (Atomic.get n_dropped);
-              ];
-            sum ~name:"otel.export.errors" ~is_monotonic:true
-              [
-                int
-                  ~start_time_unix_nano:(Mtime.to_uint64_ns last_emit)
-                  ~now:(Mtime.to_uint64_ns now) (Atomic.get n_errors);
-              ];
-          ];
-      ]
-    ) else
-      []
-
-  let send_metrics : Metrics.resource_metrics list sender =
-    {
-      send =
-        (fun m ~ret ->
-          (if !debug_ then
-            let@ () = Lock.with_lock in
-            Format.eprintf "send metrics %a@."
-              (Format.pp_print_list Metrics.pp_resource_metrics)
-              m);
-
-          let m = List.rev_append (additional_metrics ()) m in
-          push_metrics m;
-          ret ());
-    }
-
-  let send_logs : Logs.resource_logs list sender =
-    {
-      send =
-        (fun m ~ret ->
-          (if !debug_ then
-            let@ () = Lock.with_lock in
-            Format.eprintf "send logs %a@."
-              (Format.pp_print_list Logs.pp_resource_logs)
-              m);
-
-          push_logs m;
-          ret ());
-    }
-end
+(** thread that calls [tick()] regularly, to help enforce timeouts *)
+let setup_ticker_thread ~stop ~sleep_ms (module B : Collector.BACKEND) () =
+  let sleep_s = float sleep_ms /. 1000. in
+  let tick_loop () =
+    while not @@ Atomic.get stop do
+      Thread.delay sleep_s;
+      B.tick ()
+    done
+  in
+  start_bg_thread tick_loop
 
 let setup_ ?(stop = Atomic.make false) ~(config : Config.t) () =
-  debug_ := config.debug;
-  let module B =
-    Backend
-      (struct
-        let stop = stop
+  let ((module B) as backend) = mk_backend ~stop ~config () in
+  Opentelemetry.Collector.set_backend backend;
+
+  if config.ticker_thread then (
+    let sleep_ms = min 5_000 (max 2 config.batch_timeout_ms) in
+    ignore (setup_ticker_thread ~stop ~sleep_ms backend () : Thread.t)
+  );
 
-        let config = config
-      end)
-      ()
-  in
-  Opentelemetry.Collector.set_backend (module B);
   B.cleanup
 
 let setup ?stop ?(config = Config.make ()) ?(enable = true) () =
diff --git a/src/client-ocurl/opentelemetry_client_ocurl.mli b/src/client-ocurl/opentelemetry_client_ocurl.mli
index 4292586f..7d3ccc0a 100644
--- a/src/client-ocurl/opentelemetry_client_ocurl.mli
+++ b/src/client-ocurl/opentelemetry_client_ocurl.mli
@@ -14,10 +14,6 @@ val get_headers : unit -> (string * string) list
 val set_headers : (string * string) list -> unit
 (** Set http headers that are sent on every http query to the collector. *)
 
-val set_mutex : lock:(unit -> unit) -> unlock:(unit -> unit) -> unit
-(** Set a lock/unlock pair to protect the critical sections
-    of {!Opentelemetry.Collector.BACKEND} *)
-
 module Atomic = Opentelemetry_atomic.Atomic
 module Config = Config
 
diff --git a/tests/bin/emit1.ml b/tests/bin/emit1.ml
index faaeca12..35f16423 100644
--- a/tests/bin/emit1.ml
+++ b/tests/bin/emit1.ml
@@ -3,7 +3,7 @@ module Atomic = Opentelemetry_atomic.Atomic
 
 let spf = Printf.sprintf
 
-let ( let@ ) f x = f x
+let ( let@ ) = ( @@ )
 
 let sleep_inner = ref 0.1
 
@@ -98,21 +98,13 @@ let () =
   let ts_start = Unix.gettimeofday () in
 
   let debug = ref false in
-  let thread = ref true in
   let n_bg_threads = ref 0 in
-  let batch_traces = ref 400 in
-  let batch_metrics = ref 3 in
   let opts =
     [
       "--debug", Arg.Bool (( := ) debug), " enable debug output";
-      "--thread", Arg.Bool (( := ) thread), " use a background thread";
       ( "--stress-alloc",
         Arg.Bool (( := ) stress_alloc_),
         " perform heavy allocs in inner loop" );
-      "--batch-traces", Arg.Int (( := ) batch_traces), " size of traces batch";
-      ( "--batch-metrics",
-        Arg.Int (( := ) batch_metrics),
-        " size of metrics batch" );
       "--sleep-inner", Arg.Set_float sleep_inner, " sleep (in s) in inner loop";
       "--sleep-outer", Arg.Set_float sleep_outer, " sleep (in s) in outer loop";
       "-j", Arg.Set_int n_jobs, " number of parallel jobs";
@@ -123,17 +115,8 @@ let () =
 
   Arg.parse opts (fun _ -> ()) "emit1 [opt]*";
 
-  let some_if_nzero r =
-    if !r > 0 then
-      Some !r
-    else
-      None
-  in
   let config =
     Opentelemetry_client_ocurl.Config.make ~debug:!debug
-      ~batch_traces:(some_if_nzero batch_traces)
-      ~batch_metrics:(some_if_nzero batch_metrics)
-      ~thread:!thread
       ?bg_threads:
         (let n = !n_bg_threads in
          if n = 0 then