cluster/distribute: bring in directory-spread-count option

global spread count option is given through volume file

Signed-off-by: Amar Tumballi <amar@gluster.com>
Signed-off-by: Anand Avati <avati@gluster.com>

BUG: 2257 (enhance distribute to control the spread count (ie, control the hashing range))
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=2257

5 files changed, 80 insertions, 4 deletions

diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 237a0789426..d5ee7ad1b90 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -2062,8 +2062,8 @@ err:
 
 
 static int
-dht_fix_layout_cbk (call_frame_t *frame, void *cookie,
-                    xlator_t *this, int32_t op_ret, int32_t op_errno)
+dht_common_setxattr_cbk (call_frame_t *frame, void *cookie,
+                         xlator_t *this, int32_t op_ret, int32_t op_errno)
 {
         DHT_STACK_UNWIND (setxattr, frame, op_ret, op_errno);
 
@@ -2082,6 +2082,9 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
         int           op_errno = EINVAL;
         int           ret      = -1;
         data_t       *tmp      = NULL;
+        uint32_t      dir_spread = 0;
+        char          value[4096] = {0,};
+
 
         VALIDATE_OR_GOTO (frame, err);
         VALIDATE_OR_GOTO (this, err);
@@ -2123,11 +2126,36 @@ dht_setxattr (call_frame_t *frame, xlator_t *this,
                         goto err;
                 }
 
-                dht_fix_directory_layout (frame, dht_fix_layout_cbk,
+                dht_fix_directory_layout (frame, dht_common_setxattr_cbk,
                                           layout);
                 return 0;
         }
 
+        tmp = dict_get (xattr, "distribute.directory-spread-count");
+        if (tmp) {
+                /* Setxattr value is packed as 'binary', not string */
+                memcpy (value, tmp->data, ((tmp->len < 4095)?tmp->len:4095));
+                ret = gf_string2uint32 (value, &dir_spread);
+                if (!ret && ((dir_spread <= conf->subvolume_cnt) &&
+                             (dir_spread > 0))) {
+                        layout->spread_cnt = dir_spread;
+
+                        ret = loc_dup (loc, &local->loc);
+                        if (ret == -1) {
+                                op_errno = ENOMEM;
+                                goto err;
+                        }
+                        dht_fix_directory_layout (frame,
+                                                  dht_common_setxattr_cbk,
+                                                  layout);
+                        return 0;
+                }
+                gf_log (this->name, GF_LOG_ERROR,
+                        "wrong 'directory-spread-count' value (%s)", value);
+                op_errno = ENOTSUP;
+                goto err;
+        }
+
         local->call_cnt = layout->cnt;
 
         for (i = 0; i < layout->cnt; i++) {
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index c85ba9cfcc2..43056c78c9f 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -41,6 +41,9 @@ typedef int (*dht_selfheal_dir_cbk_t) (call_frame_t *frame, void *cookie,
 
 
 struct dht_layout {
+        int               spread_cnt;  /* layout spread count per directory,
+                                          is controlled by 'setxattr()' with
+                                          special key */
         int               cnt;
         int               preset;
         int               gen;
@@ -171,6 +174,10 @@ struct dht_conf {
         char           vol_uuid[UUID_SIZE + 1];
         gf_boolean_t   assert_no_child_down;
         time_t        *subvol_up_time;
+
+        /* This is the count used as the distribute layout for a directory */
+        /* Will be a global flag to control the layout spread count */
+        uint32_t       dir_spread_cnt;
 };
 typedef struct dht_conf dht_conf_t;
 
diff --git a/xlators/cluster/dht/src/dht-layout.c b/xlators/cluster/dht/src/dht-layout.c
index 16767adb9fa..e1ab6c7f170 100644
--- a/xlators/cluster/dht/src/dht-layout.c
+++ b/xlators/cluster/dht/src/dht-layout.c
@@ -52,8 +52,11 @@ dht_layout_new (xlator_t *this, int cnt)
 
         layout->type = DHT_HASH_TYPE_DM;
         layout->cnt = cnt;
-        if (conf)
+
+        if (conf) {
+                layout->spread_cnt = conf->dir_spread_cnt;
                 layout->gen = conf->gen;
+        }
 
         layout->ref = 1;
 out:
diff --git a/xlators/cluster/dht/src/dht-selfheal.c b/xlators/cluster/dht/src/dht-selfheal.c
index a20526067b4..3d4f13c6f3f 100644
--- a/xlators/cluster/dht/src/dht-selfheal.c
+++ b/xlators/cluster/dht/src/dht-selfheal.c
@@ -482,6 +482,9 @@ dht_get_layout_count (xlator_t *this, dht_layout_t *layout)
                 }
         }
 
+        count = ((layout->spread_cnt) ? layout->spread_cnt :
+                 ((count) ? count : 1));
+
         return count;
 }
 
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index 3abba0dfe40..42f0c0d75e8 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -290,6 +290,7 @@ reconfigure (xlator_t *this, dict_t *options)
         gf_boolean_t     search_unhashed;
         uint32_t         temp_free_disk = 0;
         int              ret = -1;
+        uint32_t         dir_spread = 0;
 
         GF_VALIDATE_OR_GOTO ("dht", this, out);
         GF_VALIDATE_OR_GOTO ("dht", options, out);
@@ -340,6 +341,22 @@ reconfigure (xlator_t *this, dict_t *options)
                        " min-free-disk reconfigured to %s",
                        temp_str);
         }
+
+        if (dict_get_str (options, "directory-layout-spread", &temp_str) == 0) {
+                ret = gf_string2uint32 (temp_str, &dir_spread);
+                if (ret ||
+                    (dir_spread > conf->subvolume_cnt) ||
+                    (dir_spread < 1)) {
+                        gf_log (this->name, GF_LOG_ERROR,
+                                "wrong 'directory-layout-spread' option given "
+                                "(%s). setting to earlier value (%d)",
+                                temp_str, conf->dir_spread_cnt);
+                        ret = -1;
+                        goto out;
+                }
+                conf->dir_spread_cnt = dir_spread;
+        }
+
         ret = 0;
 out:
         return ret;
@@ -433,6 +450,21 @@ init (xlator_t *this)
                 }
         }
 
+        conf->dir_spread_cnt = conf->subvolume_cnt;
+        if (dict_get_str (this->options, "directory-layout-spread",
+                          &temp_str) == 0) {
+                ret = gf_string2uint32 (temp_str, &conf->dir_spread_cnt);
+                if (ret ||
+                    (conf->dir_spread_cnt > conf->subvolume_cnt) ||
+                    (conf->dir_spread_cnt < 1)) {
+                        gf_log (this->name, GF_LOG_WARNING,
+                                "wrong 'directory-layout-spread' option given "
+                                "(%s). setting it to subvolume count",
+                                temp_str);
+                        conf->dir_spread_cnt = conf->subvolume_cnt;
+                }
+        }
+
         conf->assert_no_child_down = 0;
 
         ret = dict_get_str_boolean (this->options, "assert-no-child-down", 0);
@@ -574,5 +606,8 @@ struct volume_options options[] = {
         { .key = {"assert-no-child-down"},
           .type = GF_OPTION_TYPE_BOOL
         },
+        { .key  = {"directory-layout-spread"},
+          .type = GF_OPTION_TYPE_INT,
+        },
         { .key  = {NULL} },
 };