LU-18504 llite: fix whole-file readahead RPC limit

We try to increase whole-file readahead to match RPC size,
by using the connect data from sbi->ll_dt_exp.
But that's not the right location, the lov layer
could not touch/change the ocd_brw_size in connect data,
it stays 64M as the largest possible RPC size,
which could be too large for whole-file readahead.

Introduce a new key KEY_MAX_PAGES_PER_RPC in lov_get_info,
to iterate every osc obd to get max pages per rpc replied
from OST.

After obd_connect in client_common_fill_super, there
are chances that the connect rpc reply hasn't been
processed so the optimal ocd_brw_size replied from server
hasn't been updated yet, delay checking max pages per
rpc to give us a better chance to get the replies from OST.

Bump up SBI_DEFAULT_READAHEAD_WHOLE_MAX to 4M, which is
the same to DT_DEF_BRW_SIZE from osd for now.

Fixes: 627d0133d9 ("LU-7990 llite: increase whole-file readahead to RPC size")
Signed-off-by: Li Dongyang <dongyangli@ddn.com>
Change-Id: I172bf81ac803971d5f0d50fe00ae4fdfb0d26485
Reviewed-on: https://review.whamcloud.com/c/fs/lustre-release/+/57249
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Qian Yingjin <qian@ddn.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>

diff --git a/lustre/include/obd.h b/lustre/include/obd.h
index 3ea36fa..558e8df 100644 (file)
--- a/lustre/include/obd.h
+++ b/lustre/include/obd.h
@@ -856,6 +856,7 @@ static inline bool obd_mdt_recovery_abort(struct obd_device *obd)
 #define KEY_CACHE_LRU_SHRINK   "cache_lru_shrink"
 #define KEY_OSP_CONNECTED      "osp_connected"
 #define KEY_FID2IDX            "fid2idx"
+#define KEY_MAX_PAGES_PER_RPC  "max_pages_per_rpc"
 
 #define KEY_UNEVICT_CACHE_SHRINK       "unevict_cache_shrink"
 

diff --git a/lustre/llite/llite_internal.h b/lustre/llite/llite_internal.h
index eae98ef..32bae6d 100644 (file)
--- a/lustre/llite/llite_internal.h
+++ b/lustre/llite/llite_internal.h
@@ -736,7 +736,7 @@ static inline void ll_inode_unlock(struct inode *inode)
 #define SBI_DEFAULT_READ_AHEAD_PER_FILE_MAX    MiB_TO_PAGES(256UL)
 
 /* default read-ahead full files smaller than limit on the second read */
-#define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX       MiB_TO_PAGES(2UL)
+#define SBI_DEFAULT_READ_AHEAD_WHOLE_MAX       MiB_TO_PAGES(4UL)
 
 /* default range pages */
 #define SBI_DEFAULT_RA_RANGE_PAGES             MiB_TO_PAGES(1ULL)

diff --git a/lustre/llite/llite_lib.c b/lustre/llite/llite_lib.c
index 76820db..969700c 100644 (file)
--- a/lustre/llite/llite_lib.c
+++ b/lustre/llite/llite_lib.c
@@ -678,17 +678,6 @@ retry_connect:
 
        sbi->ll_dt_exp->exp_connect_data = *data;
 
-       /* Don't change value if it was specified in the config log */
-       if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
-               sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
-                       max_t(unsigned long, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
-                             (data->ocd_brw_size >> PAGE_SHIFT));
-               if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
-                   sbi->ll_ra_info.ra_max_pages_per_file)
-                       sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
-                               sbi->ll_ra_info.ra_max_pages_per_file;
-       }
-
        err = client_fid_init(sbi->ll_dt_exp->exp_obd, sbi->ll_dt_exp,
                              LUSTRE_SEQ_METADATA);
        if (err) {
@@ -833,6 +822,26 @@ retry_connect:
        OBD_FREE_PTR(data);
        OBD_FREE_PTR(osfs);
 
+       /* Don't change value if it was specified in the config log */
+       if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages == -1) {
+               u32 max_pages_per_rpc;
+
+               size = sizeof(max_pages_per_rpc);
+               err = obd_get_info(NULL, sbi->ll_dt_exp,
+                                  sizeof(KEY_MAX_PAGES_PER_RPC),
+                                  KEY_MAX_PAGES_PER_RPC, &size,
+                                  &max_pages_per_rpc);
+               if (err)
+                       max_pages_per_rpc = 0;
+               sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
+                       max_t(u32, SBI_DEFAULT_READ_AHEAD_WHOLE_MAX,
+                             max_pages_per_rpc);
+               if (sbi->ll_ra_info.ra_max_read_ahead_whole_pages >
+                   sbi->ll_ra_info.ra_max_pages_per_file)
+                       sbi->ll_ra_info.ra_max_read_ahead_whole_pages =
+                               sbi->ll_ra_info.ra_max_pages_per_file;
+       }
+
        if (sbi->ll_dt_obd) {
                err = sysfs_create_link(&sbi->ll_kset.kobj,
                                        &sbi->ll_dt_obd->obd_kset.kobj,

diff --git a/lustre/lov/lov_obd.c b/lustre/lov/lov_obd.c
index 5842c49..f4c8584 100644 (file)
--- a/lustre/lov/lov_obd.c
+++ b/lustre/lov/lov_obd.c
@@ -1134,6 +1134,34 @@ static int lov_get_info(const struct lu_env *env, struct obd_export *exp,
                *((u32 *)val) = lov_mds_md_size(def_stripe_count, LOV_MAGIC_V3);
        } else if (KEY_IS(KEY_TGT_COUNT)) {
                *((int *)val) = lov->desc.ld_tgt_count;
+       } else if (KEY_IS(KEY_MAX_PAGES_PER_RPC)) {
+               struct lov_tgt_desc *tgt;
+               struct client_obd *cli;
+               struct obd_import *imp;
+               u32 result = 0;
+               u32 i;
+
+               for (i = 0; i < lov->desc.ld_tgt_count; i++) {
+                       tgt = lov->lov_tgts[i];
+
+                       /* OST was disconnected */
+                       if (tgt == NULL || tgt->ltd_exp == NULL)
+                               continue;
+
+                       cli = &tgt->ltd_obd->u.cli;
+                       imp = cli->cl_import;
+
+                       if (imp == NULL || imp->imp_state != LUSTRE_IMP_FULL)
+                               continue;
+
+                       if (result == 0)
+                               result = cli->cl_max_pages_per_rpc;
+                       else
+                               result = min_t(u32, cli->cl_max_pages_per_rpc,
+                                              result);
+               }
+
+               *((u32 *)val) = result;
        } else {
                rc = -EINVAL;
        }