gazagnaire.org
Logging, runtime tracing, and the obs performance profiler for OCaml
obs: unhang the producer test suite
Forwarder's stop joins the session domain, so it blocks. test_server_end_to_end called it from inside Eio_main.run, parking that scheduler: the collector's handler fiber could then never drain the socket the session domain was blocked writing to, and neither side ever observed the stop flag. dune test hung there indefinitely -- the suite was killed, never passed.
The tests stop through Eio_unix.run_in_systhread instead, and the blocking contract is stated on the stop type. Observe.setup is unaffected: it defers stop to at_exit, which runs after Eio_main.run has returned.
author
Thomas Gazagnaire
date
(Jul 9, 2026, 7:00 PM UTC)
commit
8557cb05
8557cb05d752295404266b8a44c4a87c92ac2ff3
parent
ecd79810
ecd798109bbbc5aec309f8c0ba436afa308eb146
+13
-3
+7
-1
lib/producer/forwarder.mli
+7
-1
lib/producer/forwarder.mli
···
12
12
type stop = unit -> unit
13
13
(** Stops forwarding: pauses the ring, flushes the backlog, streams the memtrace
14
14
CTF if one was captured, sends {!Observe_protocol.Frame.Bye}, and closes the
15
-
connection. *)
15
+
connection.
16
+
17
+
This blocks until the background domain has joined. An Eio program must
18
+
therefore not call it from a fiber, or it stalls that domain's scheduler
19
+
while the session domain may still be waiting to write; run it with
20
+
[Eio_unix.run_in_systhread] instead. {!Observe.setup} defers it to
21
+
[at_exit], which runs after [Eio_main.run] has returned. *)
16
22
17
23
(** Where to reach the collector, named by [OBS_COLLECTOR]. The first three are
18
24
dialled; the [*-listen] forms make the producer the passive side -- it binds
+6
-2
test/producer/test_forwarder.ml
+6
-2
test/producer/test_forwarder.ml
···
269
269
Gc.full_major ()
270
270
done;
271
271
Eio.Time.sleep clock 0.05;
272
-
stop ();
272
+
(* [stop] joins the session domain. Calling it from a fiber would block
273
+
this scheduler, so the handler fiber could never drain the socket that
274
+
the session domain is blocked writing to, and neither side would ever
275
+
observe the stop flag. *)
276
+
Eio_unix.run_in_systhread stop;
273
277
(* Let the handler fiber drain the Bye and see EOF before the switch ends. *)
274
278
Eio.Time.sleep clock 0.05);
275
279
Alcotest.(check int)
···
300
304
Gc.full_major ()
301
305
done;
302
306
Eio.Time.sleep clock 0.05;
303
-
stop ();
307
+
Eio_unix.run_in_systhread stop;
304
308
Eio.Time.sleep clock 0.05);
305
309
Alcotest.(check int)
306
310
"one producer registered over tcp" 1