gluster volume create: option for stripe + replicate volume

So RAID01 like option is a possibility

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

BUG: 3040 (need a way to create volumes with 'stripe+replicate' setup..)
URL: http://bugs.gluster.com/cgi-bin/bugzilla3/show_bug.cgi?id=3040

5 files changed, 149 insertions, 24 deletions

diff --git a/cli/src/cli-cmd-parser.c b/cli/src/cli-cmd-parser.c
index 3e331848d..2c1897550 100644
--- a/cli/src/cli-cmd-parser.c
+++ b/cli/src/cli-cmd-parser.c
@@ -158,6 +158,7 @@ cli_cmd_volume_create_parse (const char **words, int wordcount, dict_t **options
         int     ret = -1;
         gf1_cluster_type type = GF_CLUSTER_TYPE_NONE;
         int     count = 1;
+        int     sub_count = 1;
         int     brick_index = 0;
         int     i = 0;
         char    *trans_type = NULL;
@@ -204,19 +205,23 @@ cli_cmd_volume_create_parse (const char **words, int wordcount, dict_t **options
                                 goto out;
         }
 
-
-        ret = dict_set_str (dict, "volname", volname);
-        if (ret)
-                goto out;
-
         if (wordcount < 4) {
                 ret = -1;
                 goto out;
         }
+
+        if (wordcount < 6) {
+                /* seems no options are given, go directly to the parse_brick */
+                brick_index = 3;
+                type = GF_CLUSTER_TYPE_NONE;
+                trans_type = gf_strdup ("tcp");
+                goto parse_bricks;
+        }
+
         w = str_getunamb (words[3], opwords_cl);
         if (!w) {
                 type = GF_CLUSTER_TYPE_NONE;
-                brick_index = 3;
+                index = 3;
         } else if ((strcmp (w, "replica")) == 0) {
                 type = GF_CLUSTER_TYPE_REPLICATE;
                 if (wordcount < 5) {
@@ -232,7 +237,7 @@ cli_cmd_volume_create_parse (const char **words, int wordcount, dict_t **options
                 ret = dict_set_int32 (dict, "replica-count", count);
                 if (ret)
                         goto out;
-                brick_index = 5;
+                index = 5;
         } else if ((strcmp (w, "stripe")) == 0) {
                 type = GF_CLUSTER_TYPE_STRIPE;
                 if (wordcount < 5) {
@@ -248,24 +253,78 @@ cli_cmd_volume_create_parse (const char **words, int wordcount, dict_t **options
                 ret = dict_set_int32 (dict, "stripe-count", count);
                 if (ret)
                         goto out;
-                brick_index = 5;
-        } else
+                index = 5;
+        } else {
                 GF_ASSERT (!"opword mismatch");
-
-        ret = dict_set_int32 (dict, "type", type);
-        if (ret)
+                ret = -1;
                 goto out;
+        }
 
-        if (type)
-                index = 5;
-        else
-                index = 3;
+        /* Ie, how many bricks together forms a subvolume of distribute */
+        sub_count = count;
+        /* reset the count value now */
+        count = 1;
+
+        /* It can be a 'stripe x replicate' (ie, RAID01) type of volume */
+        w = str_getunamb (words[5], opwords_cl);
+        if (w && ((strcmp (w, "replica")) == 0)) {
+                if (type == GF_CLUSTER_TYPE_REPLICATE) {
+                        cli_out ("'replica' option is already given");
+                        ret = -1;
+                        goto out;
+                }
+                /* The previous one should have been 'stripe' option */
+                type = GF_CLUSTER_TYPE_STRIPE_REPLICATE;
 
-        if (wordcount < (index + 1)) {
+                if (wordcount < 7) {
+                        ret = -1;
+                        goto out;
+                }
+                count = strtol (words[6], NULL, 0);
+                if (!count || (count < 2)) {
+                        cli_out ("replica count should be greater than 1");
+                        ret = -1;
+                        goto out;
+                }
+
+                ret = dict_set_int32 (dict, "replica-count", count);
+                if (ret)
+                        goto out;
+                index = 7;
+        } else if (w && ((strcmp (w, "stripe")) == 0)) {
+                if (type == GF_CLUSTER_TYPE_STRIPE) {
+                        cli_out ("'stripe' option is already given");
+                        ret = -1;
+                        goto out;
+                }
+
+                /* The previous one should have been 'replica' option */
+                type = GF_CLUSTER_TYPE_STRIPE_REPLICATE;
+
+                if (wordcount < 7) {
+                        ret = -1;
+                        goto out;
+                }
+                count = strtol (words[6], NULL, 0);
+                if (!count || (count < 2)) {
+                        cli_out ("stripe count should be greater than 1");
+                        ret = -1;
+                        goto out;
+                }
+                ret = dict_set_int32 (dict, "stripe-count", count);
+                if (ret)
+                        goto out;
+                index = 7;
+        } else if (w) {
+                GF_ASSERT (!"opword mismatch");
                 ret = -1;
                 goto out;
         }
 
+        sub_count *= count;
+
+        brick_index = index;
+
         if (str_getunamb (words[index], opwords_tr)) {
                 brick_index = index+2;
                 if (wordcount < (index + 2)) {
@@ -290,10 +349,7 @@ cli_cmd_volume_create_parse (const char **words, int wordcount, dict_t **options
                 trans_type = gf_strdup ("tcp");
         }
 
-        ret = dict_set_dynstr (dict, "transport", trans_type);
-        if (ret)
-                goto out;
-
+parse_bricks:
         ret = cli_cmd_bricks_parse (words, wordcount, brick_index, &bricks,
                                     &brick_count);
         if (ret)
@@ -307,17 +363,35 @@ cli_cmd_volume_create_parse (const char **words, int wordcount, dict_t **options
                 goto out;
         }
 
-        if (brick_count % count) {
+        if (brick_count % sub_count) {
                 if (type == GF_CLUSTER_TYPE_STRIPE)
                         cli_out ("number of bricks is not a multiple of "
                                  "stripe count");
                 else if (type == GF_CLUSTER_TYPE_REPLICATE)
                         cli_out ("number of bricks is not a multiple of "
                                  "replica count");
+                else
+                        cli_out ("number of bricks given doesn't match "
+                                 "required count");
+
                 ret = -1;
                 goto out;
         }
 
+        /* Everything if parsed fine. start setting info in dict */
+        ret = dict_set_str (dict, "volname", volname);
+        if (ret)
+                goto out;
+
+        ret = dict_set_int32 (dict, "type", type);
+        if (ret)
+                goto out;
+
+        ret = dict_set_dynstr (dict, "transport", trans_type);
+        if (ret)
+                goto out;
+
+
         ret = dict_set_dynstr (dict, "bricks", bricks);
         if (ret)
                 goto out;
diff --git a/rpc/xdr/src/cli1-xdr.h b/rpc/xdr/src/cli1-xdr.h
index 80a2b4633..01b4e461c 100644
--- a/rpc/xdr/src/cli1-xdr.h
+++ b/rpc/xdr/src/cli1-xdr.h
@@ -37,6 +37,7 @@ enum gf1_cluster_type {
 	GF_CLUSTER_TYPE_NONE = 0,
 	GF_CLUSTER_TYPE_STRIPE = 0 + 1,
 	GF_CLUSTER_TYPE_REPLICATE = 0 + 2,
+	GF_CLUSTER_TYPE_STRIPE_REPLICATE = 0 + 3,
 };
 typedef enum gf1_cluster_type gf1_cluster_type;
 
diff --git a/xlators/mgmt/glusterd/src/glusterd-op-sm.c b/xlators/mgmt/glusterd/src/glusterd-op-sm.c
index 0cf94171c..47f2c73c9 100644
--- a/xlators/mgmt/glusterd/src/glusterd-op-sm.c
+++ b/xlators/mgmt/glusterd/src/glusterd-op-sm.c
@@ -2707,6 +2707,17 @@ glusterd_op_create_volume (dict_t *dict, char **op_errstr)
                                       &sub_count);
                 if (ret)
                         goto out;
+        } else if (GF_CLUSTER_TYPE_STRIPE_REPLICATE == volinfo->type) {
+                ret = dict_get_int32 (dict, "stripe-count",
+                                      &volinfo->stripe_count);
+                if (ret)
+                        goto out;
+                ret = dict_get_int32 (dict, "replica-count",
+                                      &volinfo->replica_count);
+                if (ret)
+                        goto out;
+
+                sub_count = volinfo->stripe_count * volinfo->replica_count;
         }
 
         ret = dict_get_str (dict, "transport", &trans_type);
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index a090f46de..607fe1daf 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -1506,6 +1506,7 @@ static int
 client_graph_builder (volgen_graph_t *graph, glusterd_volinfo_t *volinfo,
                       dict_t *set_dict, void *param)
 {
+        int                      sub_count          = 0;
         int                      dist_count         = 0;
         char                     transt[16]         = {0,};
         char                    *volname            = NULL;
@@ -1576,7 +1577,8 @@ client_graph_builder (volgen_graph_t *graph, glusterd_volinfo_t *volinfo,
                 return -1;
         }
 
-        if (volinfo->sub_count > 1) {
+        sub_count = volinfo->sub_count;
+        if (sub_count > 1) {
                 switch (volinfo->type) {
                 case GF_CLUSTER_TYPE_REPLICATE:
                         cluster_args = replicate_args;
@@ -1584,6 +1586,11 @@ client_graph_builder (volgen_graph_t *graph, glusterd_volinfo_t *volinfo,
                 case GF_CLUSTER_TYPE_STRIPE:
                         cluster_args = stripe_args;
                         break;
+                case GF_CLUSTER_TYPE_STRIPE_REPLICATE:
+                        /* Replicate after the clients, then stripe */
+                        sub_count = volinfo->replica_count;
+                        cluster_args = replicate_args;
+                        break;
                 default:
                         gf_log ("", GF_LOG_ERROR, "volume inconsistency: "
                                 "unrecognized clustering type");
@@ -1595,7 +1602,7 @@ client_graph_builder (volgen_graph_t *graph, glusterd_volinfo_t *volinfo,
                 txl = first_of (graph);
                 for (trav = txl; trav->next; trav = trav->next);
                 for (;; trav = trav->prev) {
-                        if (i % volinfo->sub_count == 0) {
+                        if (i % sub_count == 0) {
                                 xl = volgen_graph_add_nolink (graph,
                                                               cluster_args[0],
                                                               cluster_args[1],
@@ -1613,8 +1620,38 @@ client_graph_builder (volgen_graph_t *graph, glusterd_volinfo_t *volinfo,
                                 break;
                         i++;
                 }
+
+                if (GF_CLUSTER_TYPE_STRIPE_REPLICATE == volinfo->type) {
+                        sub_count = volinfo->stripe_count;
+                        cluster_args = stripe_args;
+
+                        i = 0;
+                        txl = first_of (graph);
+                        for (trav = txl; --j; trav = trav->next);
+                        for (;; trav = trav->prev) {
+                                if (i % sub_count == 0) {
+                                        xl = volgen_graph_add_nolink (graph,
+                                                                      cluster_args[0],
+                                                                      cluster_args[1],
+                                                                      volname, j);
+                                        if (!xl)
+                                                return -1;
+                                        j++;
+                                }
+
+                                ret = volgen_xlator_link (xl, trav);
+                                if (ret)
+                                        return -1;
+
+                                if (trav == txl)
+                                        break;
+                                i++;
+                        }
+
+                }
         }
 
+
         if (volinfo->sub_count)
                 dist_count = volinfo->brick_count / volinfo->sub_count;
         else
diff --git a/xlators/mgmt/glusterd/src/glusterd.h b/xlators/mgmt/glusterd/src/glusterd.h
index b1027c171..7cbb4bc56 100644
--- a/xlators/mgmt/glusterd/src/glusterd.h
+++ b/xlators/mgmt/glusterd/src/glusterd.h
@@ -187,6 +187,8 @@ struct glusterd_volinfo_ {
         struct list_head        bricks;
         glusterd_volume_status  status;
         int                     sub_count;
+        int                     stripe_count;
+        int                     replica_count;
         int                     port;
         glusterd_store_handle_t *shandle;