From: Max Kellermann <max.kellermann@ionos.com>
Date: Thu, 10 Oct 2024 07:31:53 +0000 (+0200)
Subject: MDS/CDir: return as early as possible from CDir::should_split_fast()
X-Git-Tag: testing/wip-pdonnell-testing-20241106.142118-debug~24^2
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=27feea13ff0fe9690bc099e3692ffbd5db5e51e4;p=ceph-ci.git

MDS/CDir: return as early as possible from CDir::should_split_fast()

All we want to know is whether we're above the `fast_limit`; if we're
above that, we don't need to know how many exactly.

By returning early instead of iterating over all entries, a lot of CPU
time can be saved.  In a microbenchmark where `fio` was used to create
thousands of files, the CPU usage of `CDir::should_split_fast()` went
from 6% to less than 1% in the `perf report`.

Signed-off-by: Max Kellermann <max.kellermann@ionos.com>
---

diff --git a/src/mds/CDir.cc b/src/mds/CDir.cc
index f000da7928a..0095b5276e9 100644
--- a/src/mds/CDir.cc
+++ b/src/mds/CDir.cc
@@ -3811,10 +3811,13 @@ bool CDir::should_split_fast() const
     const CDentry *dn = p.second;
     if (!dn->get_projected_linkage()->is_null()) {
       effective_size++;
+
+      if (effective_size > fast_limit) [[unlikely]]
+	return true;
     }
   }
 
-  return effective_size > fast_limit;
+  return false;
 }
 
 bool CDir::should_merge() const