[lustre-devel] [PATCH 07/29] lustre: readahead: limit over reservation
James Simmons
jsimmons at infradead.org
Sun Apr 25 13:08:14 PDT 2021
From: Wang Shilong <wshilong at ddn.com>
For performance reason, exceeding @ra_max_pages are allowed to
cover current read window, but this should be limited with RPC
size in case a large block size read issued. Trim to RPC boundary.
Otherwise, too many read ahead pages might be issued and
make client short of LRU pages.
Fixes: 35b7c43c21 ("lustre: llite: allow current readahead to exceed reservation")
WC-bug-id: https://jira.whamcloud.com/browse/LU-12142
Signed-off-by: Wang Shilong <wshilong at ddn.com>
Reviewed-on: https://review.whamcloud.com/42060
Reviewed-by: Andreas Dilger <adilger at whamcloud.com>
Reviewed-by: Bobi Jam <bobijam at hotmail.com>
Reviewed-by: Oleg Drokin <green at whamcloud.com>
Signed-off-by: James Simmons <jsimmons at infradead.org>
---
fs/lustre/llite/lproc_llite.c | 10 ++++++++--
fs/lustre/llite/rw.c | 8 ++++++++
2 files changed, 16 insertions(+), 2 deletions(-)
diff --git a/fs/lustre/llite/lproc_llite.c b/fs/lustre/llite/lproc_llite.c
index ec241a4..4ce6fab 100644
--- a/fs/lustre/llite/lproc_llite.c
+++ b/fs/lustre/llite/lproc_llite.c
@@ -455,6 +455,7 @@ static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
struct super_block *sb = m->private;
struct ll_sb_info *sbi = ll_s2sbi(sb);
struct cl_client_cache *cache = sbi->ll_cache;
+ struct ll_ra_info *ra = &sbi->ll_ra_info;
long max_cached_mb;
long unused_mb;
@@ -462,17 +463,22 @@ static int ll_max_cached_mb_seq_show(struct seq_file *m, void *v)
max_cached_mb = PAGES_TO_MiB(cache->ccc_lru_max);
unused_mb = PAGES_TO_MiB(atomic_long_read(&cache->ccc_lru_left));
mutex_unlock(&cache->ccc_max_cache_mb_lock);
+
seq_printf(m,
"users: %d\n"
"max_cached_mb: %ld\n"
"used_mb: %ld\n"
"unused_mb: %ld\n"
- "reclaim_count: %u\n",
+ "reclaim_count: %u\n"
+ "max_read_ahead_mb: %lu\n"
+ "used_read_ahead_mb: %d\n",
refcount_read(&cache->ccc_users),
max_cached_mb,
max_cached_mb - unused_mb,
unused_mb,
- cache->ccc_lru_shrinkers);
+ cache->ccc_lru_shrinkers,
+ PAGES_TO_MiB(ra->ra_max_pages),
+ PAGES_TO_MiB(atomic_read(&ra->ra_cur_pages)));
return 0;
}
diff --git a/fs/lustre/llite/rw.c b/fs/lustre/llite/rw.c
index 8bba97f..2d08767 100644
--- a/fs/lustre/llite/rw.c
+++ b/fs/lustre/llite/rw.c
@@ -788,6 +788,14 @@ static int ll_readahead(const struct lu_env *env, struct cl_io *io,
vio->vui_ra_start_idx + vio->vui_ra_pages - 1;
pages_min = vio->vui_ra_start_idx + vio->vui_ra_pages -
ria->ria_start_idx;
+ /**
+ * For performance reason, exceeding @ra_max_pages
+ * are allowed, but this should be limited with RPC
+ * size in case a large block size read issued. Trim
+ * to RPC boundary.
+ */
+ pages_min = min(pages_min, ras->ras_rpc_pages -
+ (ria->ria_start_idx % ras->ras_rpc_pages));
}
/* don't over reserved for mmap range read */
--
1.8.3.1
More information about the lustre-devel
mailing list