glusterd: Support distributed replicated volumes on hot tier

We did not set up the graph properly for hot tiers with replicated
subvolumes. Also add check that the file has not already been moved
by another replicated brick on the same node.

Change-Id: I9adef565ab60f6774810962d912168b77a6032fa
BUG: 1206517
Signed-off-by: Dan Lambright <dlambrig@redhat.com>
Reviewed-on: http://review.gluster.org/10054
Reviewed-by: Joseph Fernandes <josferna@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Kaleb KEITHLEY <kkeithle@redhat.com>

8 files changed, 97 insertions, 35 deletions

diff --git a/cli/src/cli-cmd-parser.c b/cli/src/cli-cmd-parser.c
index 7c03ae228f..a334fd931b 100644
--- a/cli/src/cli-cmd-parser.c
+++ b/cli/src/cli-cmd-parser.c
@@ -1441,7 +1441,7 @@ out:
 
 int32_t
 cli_cmd_volume_add_brick_parse (const char **words, int wordcount,
-                                dict_t **options)
+                                dict_t **options, int *ret_type)
 {
         dict_t  *dict = NULL;
         char    *volname = NULL;
@@ -1559,6 +1559,9 @@ parse_bricks:
         *options = dict;
 
 out:
+        if (ret_type)
+                *ret_type = type;
+
         if (ret) {
                 gf_log ("cli", GF_LOG_ERROR, "Unable to parse add-brick CLI");
                 if (dict)
diff --git a/cli/src/cli-cmd-volume.c b/cli/src/cli-cmd-volume.c
index 5436c76bcc..9c61c3f541 100644
--- a/cli/src/cli-cmd-volume.c
+++ b/cli/src/cli-cmd-volume.c
@@ -787,7 +787,7 @@ cli_cmd_volume_add_brick_cbk (struct cli_state *state,
         if (!frame)
                 goto out;
 
-        ret = cli_cmd_volume_add_brick_parse (words, wordcount, &options);
+        ret = cli_cmd_volume_add_brick_parse (words, wordcount, &options, 0);
         if (ret) {
                 cli_usage_out (word->pattern);
                 parse_error = 1;
@@ -847,12 +847,13 @@ cli_cmd_volume_attach_tier_cbk (struct cli_state *state,
         int                     parse_error = 0;
         gf_answer_t             answer = GF_ANSWER_NO;
         cli_local_t             *local = NULL;
+        int                     type = 0;
 
         frame = create_frame (THIS, THIS->ctx->pool);
         if (!frame)
                 goto out;
 
-        ret = cli_cmd_volume_add_brick_parse (words, wordcount, &options);
+        ret = cli_cmd_volume_add_brick_parse (words, wordcount, &options, &type);
         if (ret) {
                 cli_usage_out (word->pattern);
                 parse_error = 1;
@@ -872,7 +873,7 @@ cli_cmd_volume_attach_tier_cbk (struct cli_state *state,
         if (ret)
                 goto out;
 
-        ret = dict_set_int32 (options, "type", GF_CLUSTER_TYPE_TIER);
+        ret = dict_set_int32 (options, "type", type);
         if (ret)
                 goto out;
 
diff --git a/cli/src/cli.h b/cli/src/cli.h
index ed2bc4aba8..60bbcb9f62 100644
--- a/cli/src/cli.h
+++ b/cli/src/cli.h
@@ -249,7 +249,7 @@ cli_cmd_ganesha_parse (struct cli_state *state, const char **words,
 
 int32_t
 cli_cmd_volume_add_brick_parse (const char **words, int wordcount,
-                                dict_t **options);
+                                dict_t **options, int *type);
 
 int32_t
 cli_cmd_volume_remove_brick_parse (const char **words, int wordcount,
diff --git a/tests/basic/tier/tier.t b/tests/basic/tier/tier.t
index 5884641235..d1e1041f87 100755
--- a/tests/basic/tier/tier.t
+++ b/tests/basic/tier/tier.t
@@ -3,6 +3,12 @@
 . $(dirname $0)/../../include.rc
 . $(dirname $0)/../../volume.rc
 
+LAST_BRICK=3
+CACHE_BRICK_FIRST=4
+CACHE_BRICK_LAST=5
+DEMOTE_TIMEOUT=12
+PROMOTE_TIMEOUT=5
+
 function file_on_slow_tier {
     s=$(md5sum $1)
     for i in `seq 0 $LAST_BRICK`; do
@@ -19,8 +25,9 @@ function file_on_fast_tier {
     local ret="1"
 
     s1=$(md5sum $1)
-    s2=$(md5sum $B0/${V0}${CACHE_BRICK}/$1)
-    if [ -e $B0/${V0}${CACHE_BRICK}/$1 ] && ! [ "$s1" == "$s2" ]; then
+    s2=$(md5sum $B0/${V0}${CACHE_BRICK_FIRST}/$1)
+
+    if [ -e $B0/${V0}${CACHE_BRICK_FIRST}/$1 ] && ! [ "$s1" == "$s2" ]; then
         echo "0"
     else
         echo "1"
@@ -45,17 +52,14 @@ function confirm_vol_stopped {
     fi
 }
 
-LAST_BRICK=1
-CACHE_BRICK=2
-DEMOTE_TIMEOUT=12
-PROMOTE_TIMEOUT=5
 cleanup
 
 
 TEST glusterd
 TEST pidof glusterd
-TEST $CLI volume create $V0 $H0:$B0/${V0}{0..$LAST_BRICK}
-TEST $CLI volume attach-tier $V0 $H0:$B0/${V0}${CACHE_BRICK}
+
+TEST $CLI volume create $V0 replica 2 $H0:$B0/${V0}{0..$LAST_BRICK}
+TEST $CLI volume attach-tier $V0 replica 2 $H0:$B0/${V0}$CACHE_BRICK_FIRST $H0:$B0/${V0}$CACHE_BRICK_LAST
 TEST $CLI volume start $V0
 TEST $CLI volume set $V0 features.ctr-enabled on
 TEST $GFS --volfile-id=/$V0 --volfile-server=$H0 $M0;
@@ -120,7 +124,7 @@ TEST $CLI volume detach-tier $V0
 # temporarily comment out
 #TEST ! [ -e $M0/d1/data.txt ]
 
-EXPECT "0" confirm_tier_removed ${V0}${CACHE_BRICK}
+EXPECT "0" confirm_tier_removed ${V0}${CACHE_BRICK_FIRST}
 
 EXPECT_WITHIN $REBALANCE_TIMEOUT "0" confirm_vol_stopped $V0
 
diff --git a/xlators/cluster/dht/src/tier.c b/xlators/cluster/dht/src/tier.c
index f8b32f1619..fe18d7b91a 100644
--- a/xlators/cluster/dht/src/tier.c
+++ b/xlators/cluster/dht/src/tier.c
@@ -142,12 +142,17 @@ tier_migrate_using_query_file (void *_args)
         int total_status                        = 0;
         FILE *queryFILE                         = NULL;
         char *link_str                          = NULL;
+        xlator_t *src_subvol                    = NULL;
+        dht_conf_t   *conf                      = NULL;
 
         GF_VALIDATE_OR_GOTO ("tier", query_cbk_args, out);
         GF_VALIDATE_OR_GOTO ("tier", query_cbk_args->this, out);
         this = query_cbk_args->this;
         GF_VALIDATE_OR_GOTO (this->name, query_cbk_args->defrag, out);
         GF_VALIDATE_OR_GOTO (this->name, query_cbk_args->queryFILE, out);
+        GF_VALIDATE_OR_GOTO (this->name, this->private, out);
+
+        conf = this->private;
 
         defrag = query_cbk_args->defrag;
 
@@ -300,8 +305,29 @@ tier_migrate_using_query_file (void *_args)
                         inode_unref (loc.inode);
                         loc.inode = linked_inode;
 
+                        /*
+                         * Do not promote/demote if file already is where it
+                         * should be. This shall become a skipped count.
+                         */
+                        src_subvol = dht_subvol_get_cached(this, loc.inode);
+
+                        if (query_cbk_args->is_promotion &&
+                             src_subvol == conf->subvolumes[1]) {
+                                per_link_status = -1;
+                                goto error;
+                        }
+
+                        if (!query_cbk_args->is_promotion &&
+                            src_subvol == conf->subvolumes[0]) {
+                                per_link_status = -1;
+                                goto error;
+                        }
+
                         gf_msg (this->name, GF_LOG_INFO, 0,
-                                DHT_MSG_LOG_TIER_STATUS, "Tier migrate file %s",
+                                DHT_MSG_LOG_TIER_STATUS, "Tier %d"
+                                " src_subvol %s file %s",
+                                query_cbk_args->is_promotion,
+                                src_subvol->name,
                                 loc.name);
 
                         if (tier_check_same_node (this, &loc, defrag)) {
@@ -587,6 +613,7 @@ tier_demote (void *args)
 
         query_cbk_args.this = demotion_args->this;
         query_cbk_args.defrag = demotion_args->defrag;
+        query_cbk_args.is_promotion = 0;
 
         /*Build the query file using bricklist*/
         ret = tier_build_migration_qfile(demotion_args, &query_cbk_args,
@@ -622,6 +649,7 @@ static void
 
         query_cbk_args.this = promotion_args->this;
         query_cbk_args.defrag = promotion_args->defrag;
+        query_cbk_args.is_promotion = 1;
 
         /*Build the query file using bricklist*/
         ret = tier_build_migration_qfile(promotion_args, &query_cbk_args,
@@ -641,7 +669,7 @@ out:
         return NULL;
 }
 
-static void
+static int
 tier_get_bricklist (xlator_t *xl, dict_t *bricklist)
 {
         xlator_list_t  *child = NULL;
@@ -690,14 +718,17 @@ tier_get_bricklist (xlator_t *xl, dict_t *bricklist)
                         if (dict_add_dynstr_with_alloc(bricklist, "brick",
                                                        db_path))
                                 goto out;
+
+                        ret = 0;
+                        goto out;
                 }
         }
 
         for (child = xl->children; child; child = child->next) {
-                tier_get_bricklist(child->xlator, bricklist);
+                ret = tier_get_bricklist(child->xlator, bricklist);
         }
 out:
-        return;
+        return ret;
 }
 
 int
diff --git a/xlators/cluster/dht/src/tier.h b/xlators/cluster/dht/src/tier.h
index 6dc830dabf..7d4ea6b7eb 100644
--- a/xlators/cluster/dht/src/tier.h
+++ b/xlators/cluster/dht/src/tier.h
@@ -50,6 +50,7 @@ typedef struct _query_cbk_args {
         xlator_t *this;
         gf_defrag_info_t *defrag;
         FILE *queryFILE;
+        int is_promotion;
 } query_cbk_args_t;
 
 int
diff --git a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
index b32dbb8913..165622f87e 100644
--- a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
+++ b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
@@ -465,6 +465,9 @@ __glusterd_handle_add_brick (rpcsvc_request_t *req)
                         ret = -1;
                         goto out;
                 }
+
+                ret = dict_get_int32 (dict, "type", &type);
+
                 goto brick_val;
         }
 
@@ -1835,6 +1838,7 @@ glusterd_op_perform_attach_tier (dict_t *dict,
 {
         int                                     ret = 0;
         int                                     replica_count = 0;
+        int                                     type = 0;
 
         /*
          * Store the new (cold) tier's structure until the graph is generated.
@@ -1853,6 +1857,9 @@ glusterd_op_perform_attach_tier (dict_t *dict,
         else
                 volinfo->tier_info.hot_replica_count  = 1;
         volinfo->tier_info.hot_brick_count     = count;
+        ret = dict_get_int32 (dict, "type", &type);
+        volinfo->tier_info.hot_type      = type;
+        ret = dict_set_int32 (dict, "type", GF_CLUSTER_TYPE_TIER);
 
         return ret;
 }
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index 8a61da2e8e..bbac2a2283 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -3107,11 +3107,13 @@ volume_volgen_graph_build_clusters (volgen_graph_t *graph,
         /* All other cases, it will have one or the other cluster type */
         switch (volinfo->type) {
         case GF_CLUSTER_TYPE_REPLICATE:
-                clusters = volgen_link_bricks_from_list_tail (graph, volinfo,
-                                                        replicate_args[0],
-                                                        replicate_args[1],
-                                                        volinfo->brick_count,
-                                                        volinfo->replica_count);
+                clusters = volgen_link_bricks_from_list_tail
+                        (graph, volinfo,
+                         replicate_args[0],
+                         replicate_args[1],
+                         volinfo->brick_count,
+                         volinfo->replica_count);
+
                 if (clusters < 0)
                         goto out;
                 break;
@@ -3284,12 +3286,12 @@ volume_volgen_graph_build_clusters_tier (volgen_graph_t *graph,
         volinfo->type           = volinfo->tier_info.cold_type;
         sprintf (volinfo->volname, "%s-cold", st_volname);
 
-        ret = volume_volgen_graph_build_clusters (graph, volinfo, _gf_false);
+        ret = volume_volgen_graph_build_clusters (graph, volinfo, is_quotad);
         if (ret)
                 goto out;
         cxl = first_of(graph);
 
-        volinfo->type           = GF_CLUSTER_TYPE_TIER;
+        volinfo->type           = volinfo->tier_info.hot_type;
         volinfo->brick_count    = volinfo->tier_info.hot_brick_count;
         volinfo->replica_count  = volinfo->tier_info.hot_replica_count;
         volinfo->dist_leaf_count = glusterd_get_dist_leaf_count(volinfo);
@@ -3297,21 +3299,34 @@ volume_volgen_graph_build_clusters_tier (volgen_graph_t *graph,
 
         sprintf (volinfo->volname, "%s-hot", st_volname);
 
-        if (volinfo->dist_leaf_count == 1) {
-                dist_count = volinfo->brick_count / volinfo->dist_leaf_count;
-                ret = volgen_link_bricks_from_list_head (graph,  volinfo,
-                                                         "cluster/distribute",
-                                                         "%s-dht",
-                                                         dist_count,
-                                                         dist_count);
+        dist_count = volinfo->brick_count / volinfo->dist_leaf_count;
+
+        if (volinfo->dist_leaf_count != 1) {
+                ret = volgen_link_bricks_from_list_head
+                        (graph, volinfo,
+                         "cluster/replicate",
+                         "%s-replicate-%d",
+                         volinfo->brick_count,
+                         volinfo->replica_count);
+                if (ret != -1)
+                        volgen_link_bricks_from_list_tail (graph,  volinfo,
+                                                           "cluster/distribute",
+                                                           "%s-dht",
+                                                           dist_count,
+                                                           dist_count);
         } else {
-                ret = volume_volgen_graph_build_clusters (graph,
-                                                          volinfo,
-                                                          _gf_false);
+                ret = volgen_link_bricks_from_list_head (graph,  volinfo,
+                                                 "cluster/distribute",
+                                                 "%s-dht",
+                                                 dist_count,
+                                                 dist_count);
         }
+        if (ret == -1)
+                goto out;
 
         hxl = first_of(graph);
 
+        volinfo->type           = GF_CLUSTER_TYPE_TIER;
         xl = volgen_graph_add_nolink (graph, "cluster/tier", "%s",
                                       "tier-dht", 0);
         gf_asprintf(&rule, "%s-hot-dht", st_volname);