[Devel] [PATCH RHEL7 COMMIT] mm/vmscan: Don't shrink active lrus on the first try.
Konstantin Khorenko
khorenko at virtuozzo.com
Wed Feb 20 16:32:24 MSK 2019
The commit is pushed to "branch-rh7-3.10.0-957.1.3.vz7.83.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh7-3.10.0-957.1.3.vz7.83.13
------>
commit ef120a1ee4af49d4db22d8f8cf671275670a6f8a
Author: Andrey Ryabinin <aryabinin at virtuozzo.com>
Date: Wed Feb 20 16:32:22 2019 +0300
mm/vmscan: Don't shrink active lrus on the first try.
Currently we scan all cgroups and shrink active list when
inactive list is too low. It might be better to reclaim
only inactive pages from cgroups that have big enough
inactive lru and fallback to active only if all inactive
lrus are too small.
https://pmc.acronis.com/browse/VSTOR-19037
Signed-off-by: Andrey Ryabinin <aryabinin at virtuozzo.com>
khorenko@: the patch is intended to handle following situation:
- cgroup A: unlimited, some fat applications are run inside =>
exec filecache + anon (both - mostly active), but almost no
non-exec pagecache
- neighbour cgroup B: unlimited, generates a lot of pagecache
(for example runs dd)
- cgroup C, parent for A and B: limited either by memsw or by pagecache
(vz specific)
On unpatched kernel the configuration above leads to the situation when
pagecache generated by cgroup B pressures and trashes active exec filecache
from cgroup A.
---
mm/vmscan.c | 9 ++++++---
1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/mm/vmscan.c b/mm/vmscan.c
index f64c8775ac9e..2ba4a11a67f7 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -2043,12 +2043,15 @@ static unsigned long shrink_list(enum lru_list lru, unsigned long nr_to_scan,
struct scan_control *sc)
{
if (is_active_lru(lru)) {
- if (inactive_list_is_low(lruvec, is_file_lru(lru), memcg, true))
+ if (sc->may_thrash && inactive_list_is_low(lruvec,
+ is_file_lru(lru), memcg, true))
shrink_active_list(nr_to_scan, lruvec, sc, lru);
return 0;
}
-
- return shrink_inactive_list(nr_to_scan, lruvec, sc, lru);
+ if (sc->may_thrash || !inactive_list_is_low(lruvec, is_file_lru(lru),
+ memcg, false))
+ return shrink_inactive_list(nr_to_scan, lruvec, sc, lru);
+ return 0;
}
#ifdef CONFIG_MEMCG
More information about the Devel
mailing list