glusterd : initiating gsyncd restart during add-brick

During add-brick, when a new brick is added in one of the
nodes that was already a part of the existing volume, and
gsyncd was already running on that node, then all gsyncd
processes running on that node, for that particular master
and any slave sessions will be restarted

If a new brick is added in a new node, then after adding the
brick, the user has to perform the following steps:

1. gluster system:: execute gsec_create
2. gluster volume geo-replication <master-vol> <slave-vol> create push-pem force
3. gluster volume geo-replication <master-vol> <slave-vol> start force

Change-Id: I4b9633e176c80e4a7cf33f42ebfa47ab8fc283f1
BUG: 989532
Signed-off-by: Avra Sengupta <asengupt@redhat.com>
Reviewed-on: http://review.gluster.org/5416
Tested-by: Gluster Build System <jenkins@build.gluster.com>
Reviewed-by: Amar Tumballi <amarts@redhat.com>
Reviewed-by: Vijay Bellur <vbellur@redhat.com>

3 files changed, 150 insertions, 25 deletions

diff --git a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
index 854d8c728f1..45927e23583 100644
--- a/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
+++ b/xlators/mgmt/glusterd/src/glusterd-brick-ops.c
@@ -871,23 +871,115 @@ glusterd_handle_remove_brick (rpcsvc_request_t *req)
                                             __glusterd_handle_remove_brick);
 }
 
+static int
+_glusterd_restart_gsync_session (dict_t *this, char *key,
+                                 data_t *value, void *data)
+{
+        char                          *slave      = NULL;
+        char                          *slave_buf  = NULL;
+        char                          *path_list  = NULL;
+        char                          *slave_vol  = NULL;
+        char                          *slave_ip   = NULL;
+        char                          *conf_path  = NULL;
+        int                            ret        = -1;
+        glusterd_gsync_status_temp_t  *param      = NULL;
+        gf_boolean_t                   is_running = _gf_false;
+
+        param = (glusterd_gsync_status_temp_t *)data;
+
+        GF_ASSERT (param);
+        GF_ASSERT (param->volinfo);
+
+        slave = strchr(value->data, ':');
+        if (slave) {
+                slave++;
+                slave_buf = gf_strdup (slave);
+                if (!slave_buf) {
+                        gf_log ("", GF_LOG_ERROR,
+                                "Failed to gf_strdup");
+                        ret = -1;
+                        goto out;
+                }
+        }
+        else
+                return 0;
+
+        ret = dict_set_dynstr (param->rsp_dict, "slave", slave_buf);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to store slave");
+                if (slave_buf)
+                        GF_FREE(slave_buf);
+                goto out;
+        }
+
+        ret = glusterd_get_slave_details_confpath (param->volinfo,
+                                                   param->rsp_dict,
+                                                   &slave_ip, &slave_vol,
+                                                   &conf_path);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to fetch slave or confpath details.");
+                goto out;
+        }
+
+        /* In cases that gsyncd is not running, we will not invoke it
+         * because of add-brick. */
+        ret = glusterd_check_gsync_running_local (param->volinfo->volname,
+                                                  slave, conf_path,
+                                                  &is_running);
+        if (ret) {
+                gf_log ("", GF_LOG_ERROR, "gsync running validation failed.");
+                goto out;
+        }
+        if (_gf_false == is_running) {
+                gf_log ("", GF_LOG_DEBUG, "gsync session for %s and %s is"
+                        " not running on this node. Hence not restarting.",
+                        param->volinfo->volname, slave);
+                ret = 0;
+                goto out;
+        }
+
+        ret = glusterd_get_local_brickpaths (param->volinfo, &path_list);
+        if (!path_list) {
+                gf_log ("", GF_LOG_DEBUG, "This node not being part of"
+                        " volume should not be running gsyncd. Hence"
+                        " no gsyncd process to restart.");
+                ret = 0;
+                goto out;
+        }
+
+        ret = glusterd_check_restart_gsync_session (param->volinfo, slave,
+                                                    param->rsp_dict, path_list,
+                                                    conf_path, 0);
+        if (ret)
+                gf_log ("", GF_LOG_ERROR,
+                        "Unable to restart gsync session.");
+
+out:
+        gf_log ("", GF_LOG_DEBUG, "Returning %d.", ret);
+        return ret;
+}
+
 /* op-sm */
 
 int
 glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
                                 char  *bricks, dict_t *dict)
 {
-        glusterd_brickinfo_t *brickinfo     = NULL;
-        char                 *brick         = NULL;
-        int32_t               i             = 1;
-        char                 *brick_list    = NULL;
-        char                 *free_ptr1     = NULL;
-        char                 *free_ptr2     = NULL;
-        char                 *saveptr       = NULL;
-        int32_t               ret           = -1;
-        int32_t               stripe_count  = 0;
-        int32_t               replica_count = 0;
-        int32_t               type          = 0;
+        char                         *brick          = NULL;
+        int32_t                       i              = 1;
+        char                         *brick_list     = NULL;
+        char                         *free_ptr1      = NULL;
+        char                         *free_ptr2      = NULL;
+        char                         *saveptr        = NULL;
+        int32_t                       ret            = -1;
+        int32_t                       stripe_count   = 0;
+        int32_t                       replica_count  = 0;
+        int32_t                       type           = 0;
+        glusterd_brickinfo_t         *brickinfo      = NULL;
+        glusterd_gsync_status_temp_t  param          = {0, };
+        gf_boolean_t                  restart_needed = 0;
 
         GF_ASSERT (volinfo);
 
@@ -970,18 +1062,45 @@ glusterd_op_perform_add_bricks (glusterd_volinfo_t *volinfo, int32_t count,
                 brick = strtok_r (brick_list+1, " \n", &saveptr);
 
         while (i <= count) {
-
                 ret = glusterd_volume_brickinfo_get_by_brick (brick, volinfo,
                                                               &brickinfo);
                 if (ret)
                         goto out;
 
+                if (uuid_is_null (brickinfo->uuid)) {
+                        ret = glusterd_resolve_brick (brickinfo);
+                        if (ret) {
+                                gf_log ("", GF_LOG_ERROR, FMTSTR_RESOLVE_BRICK,
+                                        brickinfo->hostname, brickinfo->path);
+                                goto out;
+                        }
+                }
+
                 ret = glusterd_brick_start (volinfo, brickinfo,
                                             _gf_true);
                 if (ret)
                         goto out;
                 i++;
                 brick = strtok_r (NULL, " \n", &saveptr);
+
+                /* Check if the brick is added in this node, and set
+                 * the restart_needed flag. */
+                if ((!uuid_compare (brickinfo->uuid, MY_UUID)) &&
+                    !restart_needed) {
+                        restart_needed = 1;
+                        gf_log ("", GF_LOG_DEBUG,
+                                "Restart gsyncd session, if it's already "
+                                "running.");
+                }
+        }
+
+        /* If the restart_needed flag is set, restart gsyncd sessions for that
+         * particular master with all the slaves. */
+        if (restart_needed) {
+                param.rsp_dict = dict;
+                param.volinfo = volinfo;
+                dict_foreach (volinfo->gsync_slaves,
+                              _glusterd_restart_gsync_session, &param);
         }
 
 out:
diff --git a/xlators/mgmt/glusterd/src/glusterd-geo-rep.c b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
index 67748daa3fc..d785cb05bfa 100644
--- a/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
+++ b/xlators/mgmt/glusterd/src/glusterd-geo-rep.c
@@ -33,11 +33,6 @@ glusterd_get_statefile_name (glusterd_volinfo_t *volinfo, char *slave,
                              char *conf_path, char **statefile);
 
 static int
-glusterd_get_slave_details_confpath (glusterd_volinfo_t *volinfo, dict_t *dict,
-                                     char **slave_ip, char **slave_vol,
-                                     char **conf_path);
-
-static int
 glusterd_get_slave_info (dict_t *dict, char **slave_ip, char **slave_vol);
 
 static int
@@ -1031,7 +1026,7 @@ glusterd_gsync_get_uuid (char *slave, glusterd_volinfo_t *vol,
         return ret;
 }
 
-static int
+int
 glusterd_check_gsync_running_local (char *master, char *slave,
                                     char *conf_path,
                                     gf_boolean_t *is_run)
@@ -2159,11 +2154,6 @@ out:
 }
 
 static int
-glusterd_check_restart_gsync_session (glusterd_volinfo_t *volinfo, char *slave,
-                                      dict_t *resp_dict, char *path_list,
-                                      char *conf_path, gf_boolean_t is_force);
-
-static int
 glusterd_gsync_configure (glusterd_volinfo_t *volinfo, char *slave,
                           char *path_list, dict_t *dict,
                           dict_t *resp_dict, char **op_errstr)
@@ -2569,7 +2559,7 @@ glusterd_read_status_file (glusterd_volinfo_t *volinfo, char *slave,
         return 0;
 }
 
-static int
+int
 glusterd_check_restart_gsync_session (glusterd_volinfo_t *volinfo, char *slave,
                                       dict_t *resp_dict, char *path_list,
                                       char *conf_path, gf_boolean_t is_force)
@@ -3394,7 +3384,7 @@ out:
         return ret;
 }
 
-static int
+int
 glusterd_get_slave_details_confpath (glusterd_volinfo_t *volinfo, dict_t *dict,
                                      char **slave_ip, char **slave_vol,
                                      char **conf_path)
diff --git a/xlators/mgmt/glusterd/src/glusterd-utils.h b/xlators/mgmt/glusterd/src/glusterd-utils.h
index 818b3f4faed..4d1dde7c8ee 100644
--- a/xlators/mgmt/glusterd/src/glusterd-utils.h
+++ b/xlators/mgmt/glusterd/src/glusterd-utils.h
@@ -549,4 +549,20 @@ gf_boolean_t
 glusterd_are_vol_all_peers_up (glusterd_volinfo_t *volinfo,
                                struct list_head *peers,
                                char **down_peerstr);
+
+int
+glusterd_get_slave_details_confpath (glusterd_volinfo_t *volinfo, dict_t *dict,
+                                     char **slave_ip, char **slave_vol,
+                                     char **conf_path);
+
+int
+glusterd_check_restart_gsync_session (glusterd_volinfo_t *volinfo, char *slave,
+                                      dict_t *resp_dict, char *path_list,
+                                      char *conf_path, gf_boolean_t is_force);
+
+int
+glusterd_check_gsync_running_local (char *master, char *slave,
+                                    char *conf_path,
+                                    gf_boolean_t *is_run);
+
 #endif