'lookup-unhashed' option of distribute should be 'auto' by default.

* Added 'auto' option, older boolean options works as they used to.

* This option should make 'create' rate faster, also handles
  self-healing of linkfile properly in case of scaling to
  more servers or filesystem is getting full.

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

BUG: 584 (automatically configure 'lookup-unhashed' option)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=584

5 files changed, 76 insertions, 22 deletions

diff --git a/xlators/cluster/dht/src/dht-common.c b/xlators/cluster/dht/src/dht-common.c
index 9c5b9006b04..3dc3609ac01 100644
--- a/xlators/cluster/dht/src/dht-common.c
+++ b/xlators/cluster/dht/src/dht-common.c
@@ -667,6 +667,8 @@ dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         loc_t        *loc           = NULL;
         call_frame_t *prev          = NULL;
         int           ret           = 0;
+        uint64_t      tmp_layout    = 0;
+        dht_layout_t *parent_layout = NULL;
 
         conf  = this->private;
 
@@ -675,11 +677,21 @@ dht_lookup_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
         loc   = &local->loc;
 
 	if (ENTRY_MISSING (op_ret, op_errno)) {
-		if (conf->search_unhashed) {
+                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_ON) {
 			local->op_errno = ENOENT;
 			dht_lookup_everywhere (frame, this, loc);
 			return 0;
 		}
+                if ((conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) &&
+                    (loc->parent)) {
+                        ret = inode_ctx_get (loc->parent, this, &tmp_layout);
+                        parent_layout = (dht_layout_t *)(long)tmp_layout;
+                        if (parent_layout->search_unhashed) {
+                                local->op_errno = ENOENT;
+                                dht_lookup_everywhere (frame, this, loc);
+                                return 0;
+                        }
+                }
 	}
 
  	if (op_ret == 0) {
@@ -2046,15 +2058,19 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
 	xlator_t     *next_subvol = NULL;
         off_t         next_offset = 0;
 	int           count = 0;
-
+        dht_layout_t *layout = 0;
+        dht_conf_t   *conf   = NULL;
+        xlator_t     *subvol = 0;
 
 	INIT_LIST_HEAD (&entries.list);
 	prev = cookie;
 	local = frame->local;
+	conf  = this->private;
 
 	if (op_ret < 0)
 		goto done;
 
+        layout = dht_layout_get (this, local->fd->inode);
 	list_for_each_entry (orig_entry, (&orig_entries->list), list) {
                 next_offset = orig_entry->d_off;
 
@@ -2071,6 +2087,17 @@ dht_readdirp_cbk (call_frame_t *frame, void *cookie, xlator_t *this, int op_ret,
                         goto unwind;
                 }
 
+                /* Do this if conf->search_unhashed is set to "auto" */
+                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
+                        subvol = dht_layout_search (this, layout,
+                                                    orig_entry->d_name);
+                        if (!subvol || (subvol != prev->this)) {
+                                /* TODO: Count the number of entries which need
+                                   linkfile to prove its existance in fs */
+                                layout->search_unhashed++;
+                        }
+                }
+
                 entry->d_stat = orig_entry->d_stat;
 
                 dht_itransform (this, prev->this, orig_entry->d_ino,
@@ -2133,15 +2160,20 @@ dht_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
 	xlator_t     *next_subvol = NULL;
         off_t         next_offset = 0;
 	int           count = 0;
-
+        dht_layout_t *layout = 0;
+        dht_conf_t   *conf   = NULL;
+        xlator_t     *subvol = 0;
 
 	INIT_LIST_HEAD (&entries.list);
 	prev = cookie;
 	local = frame->local;
+	conf  = this->private;
 
 	if (op_ret < 0)
 		goto done;
 
+        layout = dht_layout_get (this, local->fd->inode);
+
 	list_for_each_entry (orig_entry, (&orig_entries->list), list) {
                 next_offset = orig_entry->d_off;
 
@@ -2158,6 +2190,17 @@ dht_readdir_cbk (call_frame_t *frame, void *cookie, xlator_t *this,
                         goto unwind;
                 }
 
+                /* Do this if conf->search_unhashed is set to "auto" */
+                if (conf->search_unhashed == GF_DHT_LOOKUP_UNHASHED_AUTO) {
+                        subvol = dht_layout_search (this, layout,
+                                                    orig_entry->d_name);
+                        if (!subvol || (subvol != prev->this)) {
+                                /* TODO: Count the number of entries which need
+                                   linkfile to prove its existance in fs */
+                                layout->search_unhashed++;
+                        }
+                }
+
                 dht_itransform (this, prev->this, orig_entry->d_ino,
                                 &entry->d_ino);
                 dht_itransform (this, prev->this, orig_entry->d_off,
diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index f8a60b52038..ca4a1d33e10 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -25,6 +25,8 @@
 #ifndef _DHT_H
 #define _DHT_H
 
+#define GF_DHT_LOOKUP_UNHASHED_ON   1
+#define GF_DHT_LOOKUP_UNHASHED_AUTO 2
 
 typedef int (*dht_selfheal_dir_cbk_t) (call_frame_t *frame, void *cookie,
 				       xlator_t *this,
@@ -37,6 +39,7 @@ struct dht_layout {
         int               gen;
 	int               type;
         int               ref;   /* use with dht_conf_t->layout_lock */
+        int               search_unhashed;
         struct {
 		int       err;   /* 0 = normal
 				   -1 = dir exists and no xattr
diff --git a/xlators/cluster/dht/src/dht.c b/xlators/cluster/dht/src/dht.c
index d947ae8145b..d7f187457fc 100644
--- a/xlators/cluster/dht/src/dht.c
+++ b/xlators/cluster/dht/src/dht.c
@@ -259,10 +259,11 @@ init (xlator_t *this)
                 goto err;
         }
 
-	conf->search_unhashed = 0;
-
+	conf->search_unhashed =  GF_DHT_LOOKUP_UNHASHED_AUTO;
 	if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-		gf_string2boolean (temp_str, &conf->search_unhashed);
+                /* If option is not "auto", other options _should_ be boolean */
+                if (strcasecmp (temp_str, "auto"))
+                        gf_string2boolean (temp_str, &conf->search_unhashed);
 	}
 
 	conf->unhashed_sticky_bit = 0;
@@ -409,8 +410,10 @@ struct xlator_cbks cbks = {
 
 
 struct volume_options options[] = {
-        { .key  = {"lookup-unhashed"}, 
-	  .type = GF_OPTION_TYPE_BOOL 
+        { .key  = {"lookup-unhashed"},
+          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
+                    "on", "off"},
+	  .type = GF_OPTION_TYPE_STR
 	},
         { .key  = {"min-free-disk"},
           .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
diff --git a/xlators/cluster/dht/src/nufa.c b/xlators/cluster/dht/src/nufa.c
index 5fe233dfe6e..96cc5fe3e2e 100644
--- a/xlators/cluster/dht/src/nufa.c
+++ b/xlators/cluster/dht/src/nufa.c
@@ -560,12 +560,11 @@ init (xlator_t *this)
                 goto err;
         }
 
-	conf->search_unhashed = 0;
-
-	if (dict_get_str (this->options, "lookup-unhashed",
-			  &temp_str) == 0) {
-		gf_string2boolean (temp_str,
-				   &conf->search_unhashed);
+	conf->search_unhashed =  GF_DHT_LOOKUP_UNHASHED_AUTO;
+	if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
+                /* If option is not "auto", other options _should_ be boolean */
+                if (strcasecmp (temp_str, "auto"))
+                        gf_string2boolean (temp_str, &conf->search_unhashed);
 	}
 
         ret = dht_init_subvolumes (this, conf);
@@ -734,12 +733,14 @@ struct xlator_cbks cbks = {
 
 
 struct volume_options options[] = {
+        { .key  = {"lookup-unhashed"},
+          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
+                    "on", "off"},
+	  .type = GF_OPTION_TYPE_STR
+	},
 	{ .key  = {"local-volume-name"},
 	  .type = GF_OPTION_TYPE_XLATOR
 	},
-        { .key  = {"lookup-unhashed"},
-	  .type = GF_OPTION_TYPE_BOOL
-	},
         { .key  = {"min-free-disk"},
           .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
         },
diff --git a/xlators/cluster/dht/src/switch.c b/xlators/cluster/dht/src/switch.c
index 6670f0e3b3b..5a36f8dab04 100644
--- a/xlators/cluster/dht/src/switch.c
+++ b/xlators/cluster/dht/src/switch.c
@@ -887,9 +887,11 @@ init (xlator_t *this)
                 goto err;
         }
 
-	conf->search_unhashed = 1;
+	conf->search_unhashed =  GF_DHT_LOOKUP_UNHASHED_AUTO;
 	if (dict_get_str (this->options, "lookup-unhashed", &temp_str) == 0) {
-		gf_string2boolean (temp_str, &conf->search_unhashed);
+                /* If option is not "auto", other options _should_ be boolean */
+                if (strcasecmp (temp_str, "auto"))
+                        gf_string2boolean (temp_str, &conf->search_unhashed);
 	}
 
 	conf->unhashed_sticky_bit = 0;
@@ -1039,12 +1041,14 @@ struct xlator_cbks cbks = {
 
 
 struct volume_options options[] = {
+        { .key  = {"lookup-unhashed"},
+          .value = {"auto", "yes", "no", "enable", "disable", "1", "0",
+                    "on", "off"},
+	  .type = GF_OPTION_TYPE_STR
+	},
 	{ .key  = {"pattern.switch.case"},
 	  .type = GF_OPTION_TYPE_ANY
 	},
-        { .key  = {"lookup-unhashed"},
-	  .type = GF_OPTION_TYPE_BOOL
-	},
         { .key  = {"min-free-disk"},
           .type = GF_OPTION_TYPE_PERCENT_OR_SIZET,
         },