mgmt/glusterd: Store arbiter-count and restore it

Problem:
1) Glusterd doesn't remember about arbiter information of replica volume in
   store.  When glusterd goes down and comes backup, arbiter volumes will
   become replica volumes.

2) Glusterd doesn't import/export arbiter information to/from the other peers.

3) Volume info doesn't show any arbiter count in the output.

Fix:
1) Persist arbiter information in glusterd-store
2) Import/Export arbiter information of the volume
3) Change volume info output to show arbiter count.

Change-Id: I2db81e73d2694b01f7d07b08a17b41ad5a55c361
BUG: 1276675
Signed-off-by: Pranith Kumar K <pkarampu@redhat.com>
Reviewed-on: http://review.gluster.org/12475
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Ravishankar N <ravishankar@redhat.com>
Reviewed-by: Jeff Darcy <jdarcy@redhat.com>

8 files changed, 144 insertions, 13 deletions

diff --git a/cli/src/cli-rpc-ops.c b/cli/src/cli-rpc-ops.c
index 41efde80312..45fbb93867d 100644
--- a/cli/src/cli-rpc-ops.c
+++ b/cli/src/cli-rpc-ops.c
@@ -579,14 +579,22 @@ out:
 void
 gf_cli_print_number_of_bricks (int type, int brick_count, int dist_count,
                                int stripe_count, int replica_count,
-                               int disperse_count, int redundancy_count)
+                               int disperse_count, int redundancy_count,
+                               int arbiter_count)
 {
        if (type == GF_CLUSTER_TYPE_STRIPE_REPLICATE) {
-               cli_out ("Number of Bricks: %d x %d x %d = %d",
-                        (brick_count / dist_count),
-                        stripe_count,
-                        replica_count,
-                        brick_count);
+               if (arbiter_count == 0) {
+                       cli_out ("Number of Bricks: %d x %d x %d = %d",
+                                (brick_count / dist_count),
+                                stripe_count,
+                                replica_count,
+                                brick_count);
+               } else {
+                       cli_out ("Number of Bricks: %d x %d x (%d + %d) = %d",
+                                (brick_count / dist_count),
+                                stripe_count, replica_count - arbiter_count,
+                                arbiter_count, brick_count);
+               }
        } else if (type == GF_CLUSTER_TYPE_NONE ||
                   type == GF_CLUSTER_TYPE_TIER) {
                cli_out ("Number of Bricks: %d", brick_count);
@@ -598,9 +606,16 @@ gf_cli_print_number_of_bricks (int type, int brick_count, int dist_count,
        } else {
                /* For both replicate and stripe, dist_count is
                   good enough */
-               cli_out ("Number of Bricks: %d x %d = %d",
-                        (brick_count / dist_count),
-                        dist_count, brick_count);
+               if (arbiter_count == 0) {
+                       cli_out ("Number of Bricks: %d x %d = %d",
+                                (brick_count / dist_count),
+                                dist_count, brick_count);
+               } else {
+                       cli_out ("Number of Bricks: %d x (%d + %d) = %d",
+                                (brick_count / dist_count),
+                                dist_count - arbiter_count, arbiter_count,
+                                brick_count);
+               }
        }
 
 }
@@ -693,7 +708,7 @@ gf_cli_print_tier_info (dict_t *dict, int i, int brick_count)
                  cli_vol_type_str[vol_type]);
         gf_cli_print_number_of_bricks (hot_type,
                         hot_brick_count, hot_dist_count, 0,
-                        hot_replica_count, 0, 0);
+                        hot_replica_count, 0, 0, 0);
 
         ret = print_brick_details (dict, i, 1, hot_brick_count);
         if (ret)
@@ -711,7 +726,7 @@ gf_cli_print_tier_info (dict_t *dict, int i, int brick_count)
         gf_cli_print_number_of_bricks (cold_type,
                 cold_brick_count,
                 cold_dist_count, 0, cold_replica_count,
-                cold_disperse_count, cold_redundancy_count);
+                cold_disperse_count, cold_redundancy_count, 0);
 
         ret = print_brick_details (dict, i, hot_brick_count+1,
                                    brick_count);
@@ -737,6 +752,7 @@ gf_cli_get_volume_cbk (struct rpc_req *req, struct iovec *iov,
         int32_t                    replica_count        = 0;
         int32_t                    disperse_count       = 0;
         int32_t                    redundancy_count     = 0;
+        int32_t                    arbiter_count        = 0;
         int32_t                    vol_type             = 0;
         int32_t                    transport            = 0;
         char                      *volume_id_str        = NULL;
@@ -900,6 +916,11 @@ xml_output:
                 if (ret)
                         goto out;
 
+                snprintf (key, sizeof(key), "volume%d.arbiter_count", i);
+                ret = dict_get_int32 (dict, key, &arbiter_count);
+                if (ret)
+                        goto out;
+
                 snprintf (key, 256, "volume%d.transport", i);
                 ret = dict_get_int32 (dict, key, &transport);
                 if (ret)
@@ -957,7 +978,8 @@ next:
 #endif
                 gf_cli_print_number_of_bricks (type, brick_count,
                                 dist_count, stripe_count, replica_count,
-                                disperse_count, redundancy_count);
+                                disperse_count, redundancy_count,
+                                arbiter_count);
 
                 cli_out ("Transport-type: %s",
                          ((transport == 0)?"tcp":
diff --git a/tests/basic/glusterd/arbiter-volume-probe.t b/tests/basic/glusterd/arbiter-volume-probe.t
new file mode 100644
index 00000000000..cb05f4ada42
--- /dev/null
+++ b/tests/basic/glusterd/arbiter-volume-probe.t
@@ -0,0 +1,25 @@
+#!/bin/bash
+
+. $(dirname $0)/../../include.rc
+. $(dirname $0)/../../cluster.rc
+
+#This tests if the arbiter-count is transferred to the other peer.
+function check_peers {
+        $CLI_1 peer status | grep 'Peer in Cluster (Connected)' | wc -l
+}
+
+cleanup;
+
+TEST launch_cluster 2;
+TEST $CLI_1 peer probe $H2;
+
+EXPECT_WITHIN $PROBE_TIMEOUT 1 check_peers
+
+kill_glusterd 2
+$CLI_1 volume create $V0 replica 3 arbiter 1 $H0:$B0/b{1..3}
+TEST $glusterd_2
+EXPECT_WITHIN $PROBE_TIMEOUT 1 check_peers
+EXPECT "1 x \(2 \+ 1\) = 3" volinfo_field_1 $V0 "Number of Bricks"
+EXPECT "1 x \(2 \+ 1\) = 3" volinfo_field_2 $V0 "Number of Bricks"
+
+cleanup;
diff --git a/tests/basic/glusterd/arbiter-volume.t b/tests/basic/glusterd/arbiter-volume.t
new file mode 100644
index 00000000000..03f9aca2daf
--- /dev/null
+++ b/tests/basic/glusterd/arbiter-volume.t
@@ -0,0 +1,32 @@
+#!/bin/bash
+
+. $(dirname $0)/../../include.rc
+. $(dirname $0)/../../volume.rc
+
+# This command tests the volume create command validation for arbiter volumes.
+
+cleanup;
+TEST glusterd
+TEST pidof glusterd
+
+TEST $CLI volume create $V0 replica 3 arbiter 1 $H0:$B0/b1 $H0:$B0/b2 $H0:$B0/b3
+EXPECT "1 x \(2 \+ 1\) = 3" volinfo_field $V0 "Number of Bricks"
+
+TEST $CLI volume delete $V0
+TEST $CLI volume create $V0 replica 3 arbiter 1 $H0:$B0/b{4..9}
+EXPECT "2 x \(2 \+ 1\) = 6" volinfo_field $V0 "Number of Bricks"
+
+TEST $CLI volume delete $V0
+TEST $CLI volume create $V0 stripe 2 replica 3 arbiter 1 $H0:$B0/b{10..15}
+EXPECT "1 x 2 x \(2 \+ 1\) = 6" volinfo_field $V0 "Number of Bricks"
+
+TEST $CLI volume delete $V0
+TEST rm -rf $B0/b{1..3}
+TEST $CLI volume create $V0 replica 3 arbiter 1 $H0:$B0/b1 $H0:$B0/b2 $H0:$B0/b3
+EXPECT "1 x \(2 \+ 1\) = 3" volinfo_field $V0 "Number of Bricks"
+TEST killall -15 glusterd
+TEST glusterd
+TEST pidof glusterd
+EXPECT "1 x \(2 \+ 1\) = 3" volinfo_field $V0 "Number of Bricks"
+
+#cleanup
diff --git a/tests/cluster.rc b/tests/cluster.rc
index 6fd74bc8ba4..d210ca2932e 100644
--- a/tests/cluster.rc
+++ b/tests/cluster.rc
@@ -154,3 +154,19 @@ function cluster_volinfo_field()
     local field=$3;
     eval  \$CLI_$1  volume info $vol | grep "^$field: " | sed 's/.*: //';
 }
+
+function volinfo_field_1()
+{
+    local vol=$1;
+    local field=$2;
+
+    $CLI_1 volume info $vol | grep "^$field: " | sed 's/.*: //';
+}
+
+function volinfo_field_2()
+{
+    local vol=$1;
+    local field=$2;
+
+    $CLI_2 volume info $vol | grep "^$field: " | sed 's/.*: //';
+}
diff --git a/xlators/mgmt/glusterd/src/glusterd-handler.c b/xlators/mgmt/glusterd/src/glusterd-handler.c
index d3f3be20a9b..40704b74512 100644
--- a/xlators/mgmt/glusterd/src/glusterd-handler.c
+++ b/xlators/mgmt/glusterd/src/glusterd-handler.c
@@ -480,6 +480,11 @@ glusterd_add_volume_detail_to_dict (glusterd_volinfo_t *volinfo,
         if (ret)
                 goto out;
 
+        snprintf (key, sizeof (key), "volume%d.arbiter_count", count);
+        ret = dict_set_int32 (volumes, key, volinfo->arbiter_count);
+        if (ret)
+                goto out;
+
         snprintf (key, 256, "volume%d.transport", count);
         ret = dict_set_int32 (volumes, key, volinfo->transport_type);
         if (ret)
diff --git a/xlators/mgmt/glusterd/src/glusterd-store.c b/xlators/mgmt/glusterd/src/glusterd-store.c
index 66cc51f5242..5d9004c496e 100644
--- a/xlators/mgmt/glusterd/src/glusterd-store.c
+++ b/xlators/mgmt/glusterd/src/glusterd-store.c
@@ -945,9 +945,20 @@ glusterd_volume_exclude_options_write (int fd, glusterd_volinfo_t *volinfo)
         if (ret)
                 goto out;
 
+        if ((conf->op_version >= GD_OP_VERSION_3_7_6) &&
+            volinfo->arbiter_count) {
+                snprintf (buf, sizeof (buf), "%d", volinfo->arbiter_count);
+                ret = gf_store_save_value (fd,
+                                           GLUSTERD_STORE_KEY_VOL_ARBITER_CNT,
+                                           buf);
+                if (ret)
+                        goto out;
+        }
+
         if (conf->op_version >= GD_OP_VERSION_3_6_0) {
                 snprintf (buf, sizeof (buf), "%d", volinfo->disperse_count);
-                ret = gf_store_save_value (fd, GLUSTERD_STORE_KEY_VOL_DISPERSE_CNT,
+                ret = gf_store_save_value (fd,
+                                           GLUSTERD_STORE_KEY_VOL_DISPERSE_CNT,
                                            buf);
                 if (ret)
                         goto out;
@@ -2585,6 +2596,8 @@ glusterd_store_update_volinfo (glusterd_volinfo_t *volinfo)
                 } else if (!strncmp (key, GLUSTERD_STORE_KEY_VOL_REPLICA_CNT,
                                      strlen (GLUSTERD_STORE_KEY_VOL_REPLICA_CNT))) {
                         volinfo->replica_count = atoi (value);
+                } else if (!strcmp (key, GLUSTERD_STORE_KEY_VOL_ARBITER_CNT)) {
+                        volinfo->arbiter_count = atoi (value);
                 } else if (!strncmp (key, GLUSTERD_STORE_KEY_VOL_DISPERSE_CNT,
                                      strlen (GLUSTERD_STORE_KEY_VOL_DISPERSE_CNT))) {
                         volinfo->disperse_count = atoi (value);
diff --git a/xlators/mgmt/glusterd/src/glusterd-store.h b/xlators/mgmt/glusterd/src/glusterd-store.h
index 7335c0a8445..0f4cc9b138b 100644
--- a/xlators/mgmt/glusterd/src/glusterd-store.h
+++ b/xlators/mgmt/glusterd/src/glusterd-store.h
@@ -41,6 +41,7 @@ typedef enum glusterd_store_ver_ac_{
 #define GLUSTERD_STORE_KEY_VOL_REPLICA_CNT      "replica_count"
 #define GLUSTERD_STORE_KEY_VOL_DISPERSE_CNT     "disperse_count"
 #define GLUSTERD_STORE_KEY_VOL_REDUNDANCY_CNT   "redundancy_count"
+#define GLUSTERD_STORE_KEY_VOL_ARBITER_CNT      "arbiter_count"
 #define GLUSTERD_STORE_KEY_VOL_BRICK            "brick"
 #define GLUSTERD_STORE_KEY_VOL_VERSION          "version"
 #define GLUSTERD_STORE_KEY_VOL_TRANSPORT        "transport-type"
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.c b/xlators/mgmt/glusterd/src/glusterd-utils.c
index 4c1b9d42936..8f8b3b36abd 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.c
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.c
@@ -543,6 +543,7 @@ glusterd_volinfo_dup (glusterd_volinfo_t *volinfo,
 
         new_volinfo->type = volinfo->type;
         new_volinfo->replica_count = volinfo->replica_count;
+        new_volinfo->arbiter_count = volinfo->arbiter_count;
         new_volinfo->stripe_count = volinfo->stripe_count;
         new_volinfo->disperse_count = volinfo->disperse_count;
         new_volinfo->redundancy_count = volinfo->redundancy_count;
@@ -2232,6 +2233,12 @@ glusterd_add_volume_to_dict (glusterd_volinfo_t *volinfo,
                 goto out;
 
         memset (key, 0, sizeof (key));
+        snprintf (key, sizeof (key), "%s%d.arbiter_count", prefix, count);
+        ret = dict_set_int32 (dict, key, volinfo->arbiter_count);
+        if (ret)
+                goto out;
+
+        memset (key, 0, sizeof (key));
         snprintf (key, sizeof (key), "%s%d.disperse_count", prefix, count);
         ret = dict_set_int32 (dict, key, volinfo->disperse_count);
         if (ret)
@@ -3308,6 +3315,16 @@ glusterd_import_volinfo (dict_t *peer_data, int count,
                         GD_MSG_DICT_GET_FAILED,
                         "peer is possibly old version");
 
+        /* not having a 'arbiter_count' key is not a error
+           (as peer may be of old version) */
+        memset (key, 0, sizeof (key));
+        snprintf (key, sizeof (key), "%s%d.arbiter_count", prefix, count);
+        ret = dict_get_int32 (peer_data, key, &new_volinfo->arbiter_count);
+        if (ret)
+                gf_msg (THIS->name, GF_LOG_INFO, 0,
+                        GD_MSG_DICT_GET_FAILED,
+                        "peer is possibly old version");
+
         /* not having a 'disperse_count' key is not a error
            (as peer may be of old version) */
         memset (key, 0, sizeof (key));