test(broker): wait for the in-flight projection before asserting it

The reorder test read the snapshot as soon as it received both probe calls and
asserted the hosts were Checking. Dispatch starts each probe while it holds the
registry lock and publishes the in-flight projection at the end of that same
critical section, so a received call is not yet a visible Checking state and
the test asserted on the pre-dispatch publication. Under load it failed about
two runs in three; the failure printed LastAttempt zero and availability
unknown, which is exactly that earlier publication.

Wait for the published Checking state per endpoint, and keep the diagnostic
message so a future failure shows the projection it saw. Verified with 150
loaded repetitions (six concurrent benches of 25 runs) green, against four of
six failing before.

commit: 2dfda5e0f4686cfa576078b2647581986b67928d
author: brice · 33d ago · 2026-09-19 10:59 UTC
parent: 7655e2e
signature: GPG VerifiedCD88F9E564E0102646FB6077A1A07C7C7C3E349D · arch-gpg Verified for the committer email

1 file · +8 -1

M internal/usecase/broker/registry_test.go +8 -1


@@ -864,9 +864,16 @@ func TestRegistryUnchangedSetHostsIsNoOpAndReorderRepublishes(t *testing.T) {
864	864		calls[call.registration.Endpoint] = call
865	865		}
866	866		require.Len(t, calls, 2)
	867	+	// Dispatch starts each probe while it holds the registry lock and publishes
	868	+	// the in-flight projection at the end of the same critical section, so a
	869	+	// received call is not yet a visible Checking state. Wait for the
	870	+	// publication instead of assuming it has landed.
	871	+	for endpoint := range calls {
	872	+	waitHost(t, registry, endpoint, func(host ports.BrokerDaemonObservation) bool { return host.Checking })
	873	+	}
867	874		before := registry.Snapshot()
868	875		for _, host := range before.Daemons {
869		-	require.True(t, host.Checking)
	876	+	require.Truef(t, host.Checking, "host %q is not checking: %+v", host.Endpoint, before.Daemons)
870	877		}
871	878
872	879		// Identical membership in the same order never republishes and never

 		calls[call.registration.Endpoint] = call
 	}
 	require.Len(t, calls, 2)
+	// Dispatch starts each probe while it holds the registry lock and publishes
+	// the in-flight projection at the end of the same critical section, so a
+	// received call is not yet a visible Checking state. Wait for the
+	// publication instead of assuming it has landed.
+	for endpoint := range calls {
+		waitHost(t, registry, endpoint, func(host ports.BrokerDaemonObservation) bool { return host.Checking })
+	}
 	before := registry.Snapshot()
 	for _, host := range before.Daemons {
-		require.True(t, host.Checking)
+		require.Truef(t, host.Checking, "host %q is not checking: %+v", host.Endpoint, before.Daemons)
 	}
 	// Identical membership in the same order never republishes and never