Re: [PATCH 1/2] mm: add per-zone lru list stat

[Minchan Kim]

On Tue, Jul 19, 2016 at 05:48:57PM +0100, Mel Gorman wrote:
> On Wed, Jul 20, 2016 at 12:50:32AM +0900, Minchan Kim wrote:
> > While I did stress test with hackbench, I got OOM message frequently
> > which didn't ever happen in zone-lru.
> > 
> 
> This one also showed pgdat going unreclaimable early. Have you tried any
> of the three oom-related patches I sent to Joonsoo to see what impact,
> if any, it had?

Before the result, I want to say goal of this patch, again.
Without per-zone lru stat, it's really hard to debug OOM problem in
multiple zones system so regardless of solving the problem, we should add
per-zone lru stat for debuggability of OOM which has been never perfect
solution, ever.

You sent 3 patches in that thread and first one was same I had applied
when I found this problem firstly. It didn't solve the problem.

So I tested last one

diff --git a/mm/vmscan.c b/mm/vmscan.c
index a6f31617a08c..0dc443b52228 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -1415,7 +1415,7 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
 	LIST_HEAD(pages_skipped);
 
 	for (scan = 0; scan < nr_to_scan && nr_taken < nr_to_scan &&
-					!list_empty(src); scan++) {
+					!list_empty(src);) {
 		struct page *page;
 
 		page = lru_to_page(src);
@@ -1428,6 +1428,9 @@ static unsigned long isolate_lru_pages(unsigned long nr_to_scan,
 			nr_skipped[page_zonenum(page)]++;
 			continue;
 		}
+`
+		/* Pages skipped do not contribute to scan */
+		scan++;
 
 		switch (__isolate_lru_page(page, mode)) {
 		case 0:

The result is not OOM but hackbench stalls forever.
When I parse vmstat for every 2sec, I found pgskip_high velocity is too
high(i.e., 100000000 pages per 2 sec) while pgscan_direct and pgdeactiation is
really low(i.e., 30 pages per 2 sec).
The reason why it doesn't trigger OOM is a small amout of pages(i.e. 20 pages
per sec) are freed so NR_PAGES_SCANNED is always reset to zero.

> 
> -- 
> Mel Gorman
> SUSE Labs