From 77e33509c8603b67340f9d754b16b1fbaec7ac15 Mon Sep 17 00:00:00 2001 From: Willy Tarreau Date: Wed, 19 Apr 2023 17:19:28 +0200 Subject: [PATCH] MINOR: listener: resync with the thread index before heavy calculations During heavy accept competition, the CAS will occasionally fail and we'll have to go through all the calculation again. While the first two loops look heavy, they're almost never taken so they're quite cheap. However the rest of the operation is heavy because we have to consult connection counts and queue indexes for other threads, so better double-check if the index is still valid before continuing. Tests show that it's more efficient do retry half-way like this. --- src/listener.c | 10 ++++++++++ 1 file changed, 10 insertions(+) diff --git a/src/listener.c b/src/listener.c index 8e8a1f132..c3f1914fd 100644 --- a/src/listener.c +++ b/src/listener.c @@ -1324,6 +1324,16 @@ void listener_accept(struct listener *l) break; } + /* tests show that it's worth checking that other threads have not + * already changed the index to save the rest of the calculation, + * or we'd have to redo it anyway. + */ + n1 = _HA_ATOMIC_LOAD(thr_idx_ptr); + if (n0 != n1) { + n0 = n1; + continue; + } + /* here we have (r1,g1,t1) that designate the first receiver, its * thread group and local thread, and (r2,g2,t2) that designate * the second receiver, its thread group and local thread.