MINOR: mux-h1: pass CO_RFL_READ_ONCE to the lower layers when relevant
When we're in H1_MSG_RQBEFORE or H1_MSG_RPBEFORE, we know that the
first message is highly likely the only one and that it's pointless
to try to perform a second recvfrom() to complete a first partial
read. This is similar to what used to be done in the older I/O methods
with the CF_READ_DONTWAIT flag on the channel. So let's pass
CO_RFL_READ_ONCE to the transport layer during rcv_buf() in this case.
By doing so, in a test involving keep-alive connections with a non-null
client think time, we remove 20% of the recvfrom() calls, all of which
used to systematically fail. More precisely, we observe a drop from 5.0
recvfrom() per request with 60% failure to 4.0 per request with 50%
failure.
diff --git a/src/mux_h1.c b/src/mux_h1.c
index f175ac9..3128501 100644
--- a/src/mux_h1.c
+++ b/src/mux_h1.c
@@ -1986,6 +1986,7 @@
struct h1s *h1s = h1c->h1s;
size_t ret = 0, max;
int rcvd = 0;
+ int flags = 0;
TRACE_ENTER(H1_EV_H1C_RECV, h1c->conn);
@@ -2021,6 +2022,10 @@
if (b_data(&h1c->ibuf) > 0 && b_data(&h1c->ibuf) < 128)
b_slow_realign(&h1c->ibuf, trash.area, 0);
+ /* avoid useless reads after first responses */
+ if (h1s && (h1s->req.state == H1_MSG_RQBEFORE || h1s->res.state == H1_MSG_RPBEFORE))
+ flags |= CO_RFL_READ_ONCE;
+
max = buf_room_for_htx_data(&h1c->ibuf);
if (max) {
if (h1c->flags & H1C_F_IN_FULL) {
@@ -2035,7 +2040,7 @@
*/
h1c->ibuf.head = sizeof(struct htx);
}
- ret = conn->xprt->rcv_buf(conn, conn->xprt_ctx, &h1c->ibuf, max, 0);
+ ret = conn->xprt->rcv_buf(conn, conn->xprt_ctx, &h1c->ibuf, max, flags);
}
if (ret > 0) {
TRACE_DATA("data received", H1_EV_H1C_RECV, h1c->conn,,, (size_t[]){ret});