cli : check for bad brick order while creating replicate volume

gluster cli now checks the brick order when creating
a replicate or distributed-replicate volume.
If a non-optimal order is found the user is given a
warning and asked if the volume creation can proceed.

Change-Id: I38c4cb65bffb40ccf95319cf3f4f3423a4cdebe9
BUG: 2407
Reviewed-on: http://review.gluster.com/151
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Vijay Bellur <vijay@gluster.com>

3 files changed, 200 insertions, 2 deletions

diff --git a/cli/src/cli-cmd-volume.c b/cli/src/cli-cmd-volume.c
index 16dc32328d2..ec578417606 100644
--- a/cli/src/cli-cmd-volume.c
+++ b/cli/src/cli-cmd-volume.c
@@ -23,6 +23,11 @@
 #include <stdint.h>
 #include <pthread.h>
 
+#include <sys/socket.h>
+#include <netdb.h>
+#include <sys/types.h>
+#include <netinet/in.h>
+
 #ifndef _CONFIG_H
 #define _CONFIG_H
 #include "config.h"
@@ -150,6 +155,156 @@ out:
         return ret;
 }
 
+gf_ai_compare_t
+cli_cmd_compare_addrinfo (struct addrinfo *first, struct addrinfo *next)
+{
+        int             ret = -1;
+        struct addrinfo *tmp1 = NULL;
+        struct addrinfo *tmp2 = NULL;
+        char            firstip[NI_MAXHOST] = {0.};
+        char            nextip[NI_MAXHOST] = {0,};
+
+        for (tmp1 = first; tmp1 != NULL; tmp1 = tmp1->ai_next) {
+                ret = getnameinfo (tmp1->ai_addr, tmp1->ai_addrlen, firstip,
+                                   NI_MAXHOST, NULL, 0, NI_NUMERICHOST);
+                if (ret)
+                        return GF_AI_COMPARE_ERROR;
+                for (tmp2 = next; tmp2 != NULL; tmp2 = tmp2->ai_next) {
+                        ret = getnameinfo (tmp2->ai_addr, tmp2->ai_addrlen, nextip,
+                                           NI_MAXHOST, NULL, 0, NI_NUMERICHOST);
+                        if (ret)
+                                return GF_AI_COMPARE_ERROR;
+                        if (!strcmp (firstip, nextip)) {
+                                return GF_AI_COMPARE_MATCH;
+                        }
+                }
+        }
+        return GF_AI_COMPARE_NO_MATCH;
+}
+
+/* Check for non optimal brick order for replicate :
+ * Checks if bricks belonging to a replicate volume
+ * are present on the same server
+ */
+int32_t
+cli_cmd_check_brick_order (struct cli_state *state, const char *bricks,
+                           int brick_count, int sub_count)
+{
+        int             ret = -1;
+        int             i = 0;
+        int             j = 0;
+        int             k = 0;
+        addrinfo_list_t *ai_list = NULL;
+        addrinfo_list_t *ai_list_tmp1 = NULL;
+        addrinfo_list_t *ai_list_tmp2 = NULL;
+        char            *brick = NULL;
+        char            *brick_list = NULL;
+        char            *brick_list_dup = NULL;
+        char            *tmpptr = NULL;
+        struct addrinfo *ai_info = NULL;
+        gf_answer_t     answer = GF_ANSWER_NO;
+        const char      *failed_question = NULL;
+        const char      *found_question = NULL;
+        failed_question = "Failed to perform brick order check. "
+                          "Do you want to continue creating the volume? ";
+        found_question =  "Multiple bricks of a replicate volume are present"
+                          " on the same server. This setup is not optimal.\n"
+                          "Do you still want to continue creating the volume? ";
+
+        GF_ASSERT (bricks);
+        GF_ASSERT (brick_count > 0);
+        GF_ASSERT (sub_count > 0);
+
+        ai_list = malloc (sizeof (addrinfo_list_t));
+        ai_list->info = NULL;
+        INIT_LIST_HEAD (&ai_list->list);
+        brick_list = gf_strdup (bricks);
+        if (brick_list == NULL) {
+                gf_log ("cli", GF_LOG_DEBUG, "failed to allocate memory");
+                goto check_failed;
+        }
+        brick_list_dup = brick_list;
+        /* Resolve hostnames and get addrinfo */
+        while (i < brick_count) {
+                ++i;
+                brick = strtok_r (brick_list, " \n", &tmpptr);
+                brick_list = tmpptr;
+                if (brick == NULL)
+                        goto check_failed;
+                brick = strtok_r (brick, ":", &tmpptr);
+                if (brick == NULL)
+                        goto check_failed;
+                ret = getaddrinfo (brick, NULL, NULL, &ai_info);
+                if (ret)
+                        goto check_failed;
+                ai_list_tmp1 = malloc (sizeof (addrinfo_list_t));
+                if (ai_list_tmp1 == NULL)
+                        goto check_failed;
+                ai_list_tmp1->info = ai_info;
+                list_add_tail (&ai_list_tmp1->list, &ai_list->list);
+                ai_list_tmp1 = NULL;
+        }
+
+        i = 0;
+        ai_list_tmp1 = list_entry (ai_list->list.next, addrinfo_list_t, list);
+
+        /* Check for bad brick order */
+        while (i < brick_count) {
+                ++i;
+                ai_info = ai_list_tmp1->info;
+                ai_list_tmp1 = list_entry (ai_list_tmp1->list.next,
+                                           addrinfo_list_t, list);
+                if ( 0 == i % sub_count) {
+                        j = 0;
+                        continue;
+                }
+                ai_list_tmp2 = ai_list_tmp1;
+                k = j;
+                while (k < sub_count - 1) {
+                        ++k;
+                        ret = cli_cmd_compare_addrinfo (ai_info,
+                                                        ai_list_tmp2->info);
+                        if (GF_AI_COMPARE_ERROR == ret)
+                                goto check_failed;
+                        if (GF_AI_COMPARE_MATCH == ret)
+                                goto found_bad_brick_order;
+                        ai_list_tmp2 = list_entry (ai_list_tmp2->list.next,
+                                                   addrinfo_list_t, list);
+                }
+                ++j;
+        }
+        gf_log ("cli", GF_LOG_INFO, "Brick order okay");
+        ret = 0;
+        goto out;
+
+check_failed:
+        gf_log ("cli", GF_LOG_INFO, "Failed bad brick order check");
+        answer = cli_cmd_get_confirmation(state, failed_question);
+        if (GF_ANSWER_YES == answer)
+                ret = 0;
+        goto out;
+
+found_bad_brick_order:
+        gf_log ("cli", GF_LOG_INFO, "Bad brick order found");
+        answer = cli_cmd_get_confirmation (state, found_question);
+        if (GF_ANSWER_YES == answer)
+                ret = 0;
+out:
+        ai_list_tmp2 = NULL;
+        i = 0;
+        if (brick_list_dup)
+                GF_FREE (brick_list_dup);
+        list_for_each_entry (ai_list_tmp1, &ai_list->list, list) {
+                if (ai_list_tmp1->info)
+                        freeaddrinfo (ai_list_tmp1->info);
+                if (ai_list_tmp2)
+                        free (ai_list_tmp2);
+                ai_list_tmp2 = ai_list_tmp1;
+        }
+        free (ai_list_tmp2);
+        return ret;
+}
+
 int
 cli_cmd_volume_create_cbk (struct cli_state *state, struct cli_cmd_word *word,
                            const char **words, int wordcount)
@@ -160,6 +315,11 @@ cli_cmd_volume_create_cbk (struct cli_state *state, struct cli_cmd_word *word,
         dict_t                  *options = NULL;
         int                     sent = 0;
         int                     parse_error = 0;
+        char                    *brick_list = NULL;
+        int32_t                 brick_count = 0;
+        int32_t                 sub_count = 0;
+        int32_t                 type = GF_CLUSTER_TYPE_NONE;
+
 
         proc = &cli_rpc_prog->proctable[GLUSTER_CLI_CREATE_VOLUME];
 
@@ -174,7 +334,31 @@ cli_cmd_volume_create_cbk (struct cli_state *state, struct cli_cmd_word *word,
                 parse_error = 1;
                 goto out;
         }
-
+        /*Check brick order if type is replicate*/
+        if (dict_get_int32 (options, "type", &type) == 0 && type == GF_CLUSTER_TYPE_REPLICATE) {
+                if ((ret = dict_get_str (options, "bricks", &brick_list)) != 0) {
+                        gf_log ("cli", GF_LOG_ERROR, "Replica bricks check : "
+                                                     "Could not retrieve bricks list");
+                        goto out;
+                }
+                if ((ret = dict_get_int32 (options, "count", &brick_count)) != 0) {
+                        gf_log ("cli", GF_LOG_ERROR, "Replica bricks check : "
+                                                     "Could not retrieve brick count");
+                        goto out;
+                }
+                if ((ret = dict_get_int32 (options, "replica-count", &sub_count)) != 0) {
+                        gf_log ("cli", GF_LOG_ERROR, "Replica bricks check : "
+                                                    "Could not retrieve replica count");
+                        goto out;
+                }
+                gf_log ("cli", GF_LOG_INFO, "Replicate cluster type found."
+                                            " Checking brick order.");
+                ret = cli_cmd_check_brick_order (state, brick_list, brick_count, sub_count);
+                if (ret) {
+                        gf_log("cli", GF_LOG_INFO, "Not creating volume because of bad brick order");
+                        goto out;
+                }
+        }
         if (proc->fn) {
                 ret = proc->fn (frame, THIS, options);
         }
diff --git a/cli/src/cli-cmd.h b/cli/src/cli-cmd.h
index f6abfa4645f..1878571fa34 100644
--- a/cli/src/cli-cmd.h
+++ b/cli/src/cli-cmd.h
@@ -25,7 +25,10 @@
 #include "config.h"
 #endif
 
+#include <netdb.h>
+
 #include "cli.h"
+#include "list.h"
 
 typedef enum {
         GF_ANSWER_YES = 1,
@@ -55,6 +58,17 @@ typedef struct cli_profile_info_ {
         double percentage_avg_latency;
 } cli_profile_info_t;
 
+typedef struct addrinfo_list {
+        struct list_head list;
+        struct addrinfo *info;
+} addrinfo_list_t;
+
+typedef enum {
+        GF_AI_COMPARE_NO_MATCH     = 0,
+        GF_AI_COMPARE_MATCH        = 1,
+        GF_AI_COMPARE_ERROR        = 2
+} gf_ai_compare_t;
+
 typedef struct cli_cmd_volume_get_ctx_ cli_cmd_volume_get_ctx_t;
 
 int cli_cmd_volume_register (struct cli_state *state);
diff --git a/smoke.sh b/smoke.sh
index 3226b74f172..612f6400851 100755
--- a/smoke.sh
+++ b/smoke.sh
@@ -23,7 +23,7 @@ function start_fs()
     chmod 0755 $P/export;
 
     glusterd;
-    gluster volume create $V replica 2 $H:$P/export/export{1,2,3,4};
+    gluster --mode=script volume create $V replica 2 $H:$P/export/export{1,2,3,4};
     gluster volume start $V;
     glusterfs -s $H --volfile-id $V $M;
 #    mount -t glusterfs $H:/$V $M;