Hi,
This is a bug. If I remember correctly, this caused keepalive_timeout.cc to
fail sometimes and that is why it was rolled back.
With that said, you could also configure ping timeout which would apply to
keep-alive as well.
On Tuesday, July 21, 2026 at 3:35:59 AM UTC+5:30 Jonathan Patchell wrote:
> I have been testing failover scenarios on AIX where I bring down the
> network on the member in a cluster the client is connected to. It looks
> like there is a bug with the keep-alive timeouts where the timer isn’t
> being set. This causes the failover to take fairly long (~75 seconds) on
> AIX as there aren’t any OS-level socket errors until then.
>
> I have found that the following change causes the keep-alive timeouts to
> work properly causing quicker failovers respecting the
> GRPC_ARG_KEEPALIVE_TIMEOUT_MS value on AIX.
>
> I have been using grpc v1.80.0.
>
> diff --git a/src/core/ext/transport/chttp2/transport/writing.cc
> b/src/core/ext/transport/chttp2/transport/writing.cc
>
> index 30f6caa033..7a288c7066 100644
>
> --- a/src/core/ext/transport/chttp2/transport/writing.cc
>
> +++ b/src/core/ext/transport/chttp2/transport/writing.cc
>
> @@ -799,7 +799,7 @@ void grpc_chttp2_end_write(grpc_chttp2_transport* t,
> grpc_error_handle error) {
>
>
>
> if (t->keepalive_incoming_data_wanted &&
>
> t->keepalive_timeout < t->ping_timeout &&
>
> - t->keepalive_ping_timeout_handle !=
>
> + t->keepalive_ping_timeout_handle ==
>
> grpc_event_engine::experimental::EventEngine::TaskHandle::
>
> kInvalid) {
>
> if (GRPC_TRACE_FLAG_ENABLED(http2_ping) ||
>
> `keepalive_ping_timeout_handle != kInvalid` can never be true when the
> handle is first being armed (it starts as `kInvalid`), so the watchdog
> timer is never scheduled.
>
> It looks like this was a bug that was identified and fixed in
> https://github.com/grpc/grpc/commit/b086f689effef7276bef6456f599aee1c12489d4.
> However, there was an automated rollback to the original version in
> https://github.com/grpc/grpc/commit/4f9b6f15f391433ef12952fe9578b9b50e159990
> .
>
> The bug is platform-agnostic dead code, but the bug seems to be only
> visible on AIX (I don’t see such issues on Linux).
>
> Could someone confirm whether this is indeed a bug worth fixing upstream?
> Happy to submit a PR with this fix.
>
--
You received this message because you are subscribed to the Google Groups
"grpc.io" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To view this discussion visit
https://groups.google.com/d/msgid/grpc-io/ae8c3f56-e8e6-41ac-8cec-b5152880afb0n%40googlegroups.com.
