ctr/libgfdb: Performance enhancer for unlink/create/rename/link fops

1) ctr_link_consistency option for ctr xaltor is provided so that
   the user can choose to switch it on or off.
   /* For link consistency we do a double update i.e mark the link
   * during the wind and during the unwind we update/delete the link.
   * This has a performance hit. We give a choice here whether we need
   * link consistency to be spoton or not using link_consistency flag.
   * This will have only one link update */

2) In delete the wind time recording is moved to unwind path.
   /* Special performance case:
   * Updating wind time in unwind for delete. This is done here
   * as in the wind path we will not know whether its the last
   * link or not. For a last link there is not use to update any
   * wind or unwind time!*/

Change-Id: I209472fb816f939db4a868b97ba053b028f17ea6
BUG: 1217786
Signed-off-by: Joseph Fernandes <josferna@redhat.com>
Reviewed-on: http://review.gluster.org/10170
Reviewed-by: Dan Lambright <dlambrig@redhat.com>
Tested-by: Gluster Build System <jenkins@build.gluster.com>

7 files changed, 88 insertions, 31 deletions

diff --git a/libglusterfs/src/gfdb/gfdb_data_store_types.h b/libglusterfs/src/gfdb/gfdb_data_store_types.h
index 5bd95241f06..e460b15d063 100644
--- a/libglusterfs/src/gfdb/gfdb_data_store_types.h
+++ b/libglusterfs/src/gfdb/gfdb_data_store_types.h
@@ -288,6 +288,12 @@ typedef struct gfdb_db_record {
         gfdb_time_t                     gfdb_unwind_change_time;
         /* For crash consistancy while inserting/updating hard links */
         gf_boolean_t                    islinkupdate;
+        /* For link consistency we do a double update i.e mark the link
+         * during the wind and during the unwind we update/delete the link.
+         * This has a performance hit. We give a choice here whether we need
+         * link consistency to be spoton or not using link_consistency flag.
+         * This will have only one link update */
+         gf_boolean_t                   link_consistency;
         /* For dentry fops we can choose to ignore recording of unwind time */
         /* For inode fops "record_exit" volume option does the trick,       */
         /* but for dentry fops we update the LINK_UPDATE, so an extra       */
diff --git a/libglusterfs/src/gfdb/gfdb_sqlite3_helper.c b/libglusterfs/src/gfdb/gfdb_sqlite3_helper.c
index 4d925e95ef7..4b70b49419d 100644
--- a/libglusterfs/src/gfdb/gfdb_sqlite3_helper.c
+++ b/libglusterfs/src/gfdb/gfdb_sqlite3_helper.c
@@ -10,6 +10,8 @@
 
 #include "gfdb_sqlite3_helper.h"
 
+#define GFDB_SQL_STMT_SIZE 256
+
 /*****************************************************************************
  *
  *                 Helper function to execute actual sql queries
@@ -297,15 +299,19 @@ gf_sql_insert_link (gf_sql_connection_t  *sql_conn,
                    char                 *gfid,
                    char                 *pargfid,
                    char                 *basename,
-                   char                 *basepath)
+                   char                 *basepath,
+                   gf_boolean_t         link_consistency)
 {
         int ret = -1;
         sqlite3_stmt *insert_stmt = NULL;
-        char *insert_str = "INSERT INTO "
+        char insert_str[GFDB_SQL_STMT_SIZE] = "";
+
+        sprintf (insert_str, "INSERT INTO "
                            GF_FILE_LINK_TABLE
                            " (GF_ID, GF_PID, FNAME, FPATH,"
                            " W_DEL_FLAG, LINK_UPDATE) "
-                           " VALUES (?, ?, ?, ?, 0, 1);";
+                           " VALUES (?, ?, ?, ?, 0, %d);",
+                           link_consistency);
 
         CHECK_SQL_CONN (sql_conn, out);
         GF_VALIDATE_OR_GOTO (GFDB_STR_SQLITE3, gfid, out);
@@ -389,15 +395,19 @@ gf_sql_update_link (gf_sql_connection_t  *sql_conn,
                    char                 *basename,
                    char                 *basepath,
                    char                 *old_pargfid,
-                   char                 *old_basename)
+                   char                 *old_basename,
+                   gf_boolean_t         link_consistency)
 {
         int ret = -1;
         sqlite3_stmt *insert_stmt = NULL;
-        char *insert_str =  "INSERT INTO "
+        char insert_str[GFDB_SQL_STMT_SIZE] = "";
+
+        sprintf (insert_str, "INSERT INTO "
                             GF_FILE_LINK_TABLE
                             " (GF_ID, GF_PID, FNAME, FPATH,"
                             " W_DEL_FLAG, LINK_UPDATE) "
-                            " VALUES (? , ?, ?, ?, 0, 1);";
+                            " VALUES (? , ?, ?, ?, 0, %d);",
+                            link_consistency);
 
         CHECK_SQL_CONN (sql_conn, out);
         GF_VALIDATE_OR_GOTO (GFDB_STR_SQLITE3, gfid, out);
@@ -746,13 +756,15 @@ gf_sql_insert_wind (gf_sql_connection_t  *sql_conn,
                         ret = gf_sql_insert_link(sql_conn,
                                         gfid_str, pargfid_str,
                                         gfdb_db_record->file_name,
-                                        gfdb_db_record->file_path);
+                                        gfdb_db_record->file_path,
+                                        gfdb_db_record->link_consistency);
                         if (ret) {
                                 gf_log (GFDB_STR_SQLITE3, GF_LOG_ERROR,
                                 "Failed inserting link in DB");
                                 goto out;
                         }
-                        gfdb_db_record->islinkupdate = _gf_true;
+                        gfdb_db_record->islinkupdate = gfdb_db_record->
+                                                        link_consistency;
 
                         /*
                          * Only for create/mknod insert wind time
@@ -783,26 +795,31 @@ gf_sql_insert_wind (gf_sql_connection_t  *sql_conn,
                                                 gfdb_db_record->file_name,
                                                 gfdb_db_record->file_path,
                                                 old_pargfid_str,
-                                                gfdb_db_record->old_file_name);
+                                                gfdb_db_record->old_file_name,
+                                                gfdb_db_record->
+                                                        link_consistency);
                                 if (ret) {
                                         gf_log (GFDB_STR_SQLITE3, GF_LOG_ERROR,
                                                 "Failed updating link");
                                         goto out;
                                 }
-                                gfdb_db_record->islinkupdate = _gf_true;
+                                gfdb_db_record->islinkupdate = gfdb_db_record->
+                                                        link_consistency;
                         }
                         /*link*/
                         else {
                                 ret = gf_sql_insert_link (sql_conn,
                                         gfid_str, pargfid_str,
                                         gfdb_db_record->file_name,
-                                        gfdb_db_record->file_path);
+                                        gfdb_db_record->file_path,
+                                        gfdb_db_record->link_consistency);
                                 if (ret) {
                                         gf_log (GFDB_STR_SQLITE3, GF_LOG_ERROR,
                                         "Failed inserting link in DB");
                                         goto out;
                                 }
-                                gfdb_db_record->islinkupdate = _gf_true;
+                                gfdb_db_record->islinkupdate = gfdb_db_record->
+                                                        link_consistency;
                         }
                 }
         }
@@ -901,7 +918,6 @@ gf_sql_update_delete_wind (gf_sql_connection_t  *sql_conn,
                           gfdb_db_record_t     *gfdb_db_record)
 {
         int ret = -1;
-        gfdb_time_t *modtime    = NULL;
         char *gfid_str          = NULL;
         char *pargfid_str       = NULL;
 
@@ -922,29 +938,17 @@ gf_sql_update_delete_wind (gf_sql_connection_t  *sql_conn,
                         goto out;
         }
 
-        if (gfdb_db_record->do_record_times) {
-                /*Update the wind write times*/
-                modtime = &gfdb_db_record->gfdb_wind_change_time;
-                ret = gf_update_time (sql_conn, gfid_str, modtime,
-                        gfdb_db_record->do_record_counters,
-                        _gf_true,
-                        isreadfop (gfdb_db_record->gfdb_fop_type));
+        if (gfdb_db_record->link_consistency) {
+                ret = gf_sql_update_link_flags (sql_conn, gfid_str, pargfid_str,
+                                        gfdb_db_record->file_name, 1,
+                                        _gf_false);
                 if (ret) {
                         gf_log (GFDB_STR_SQLITE3, GF_LOG_ERROR,
-                                "Failed update wind time in DB");
+                                "Failed updating link flags in wind");
                         goto out;
                 }
         }
 
-        ret = gf_sql_update_link_flags (sql_conn, gfid_str, pargfid_str,
-                                        gfdb_db_record->file_name, 1,
-                                        _gf_false);
-        if (ret) {
-                gf_log (GFDB_STR_SQLITE3, GF_LOG_ERROR,
-                        "Failed updating link flags in wind");
-                goto out;
-        }
-
         ret = 0;
 out:
         GF_FREE (gfid_str);
@@ -985,6 +989,25 @@ gf_sql_delete_unwind (gf_sql_connection_t  *sql_conn,
                         goto out;
                 }
 
+                /* Special performance case:
+                 * Updating wind time in unwind for delete. This is done here
+                 * as in the wind path we will not know whether its the last
+                 * link or not. For a last link there is not use to update any
+                 * wind or unwind time!*/
+                if (gfdb_db_record->do_record_times) {
+                        /*Update the wind write times*/
+                        modtime = &gfdb_db_record->gfdb_wind_change_time;
+                        ret = gf_update_time (sql_conn, gfid_str, modtime,
+                                gfdb_db_record->do_record_counters,
+                                _gf_true,
+                                isreadfop (gfdb_db_record->gfdb_fop_type));
+                        if (ret) {
+                                gf_log (GFDB_STR_SQLITE3, GF_LOG_ERROR,
+                                        "Failed update wind time in DB");
+                                goto out;
+                        }
+                }
+
                 modtime = &gfdb_db_record->gfdb_unwind_change_time;
 
                 ret = gf_sql_delete_link(sql_conn, gfid_str, pargfid_str,
@@ -1007,6 +1030,10 @@ gf_sql_delete_unwind (gf_sql_connection_t  *sql_conn,
                                 goto out;
                         }
                 }
+        } else {
+                gf_log (GFDB_STR_SQLITE3, GF_LOG_ERROR,
+                        "Invalid unlink option");
+                goto out;
         }
         ret = 0;
 out:
diff --git a/xlators/features/changetimerecorder/src/changetimerecorder.c b/xlators/features/changetimerecorder/src/changetimerecorder.c
index 42470a92ef9..fd57a445e08 100644
--- a/xlators/features/changetimerecorder/src/changetimerecorder.c
+++ b/xlators/features/changetimerecorder/src/changetimerecorder.c
@@ -958,6 +958,9 @@ reconfigure (xlator_t *this, dict_t *options)
         GF_OPTION_RECONF ("record-counters", _priv->ctr_record_counter, options,
                           bool, out);
 
+        GF_OPTION_RECONF ("ctr_link_consistency", _priv->ctr_link_consistency,
+                        options, bool, out);
+
         GF_OPTION_RECONF ("record-exit", _priv->ctr_record_unwind, options,
                           bool, out);
 
@@ -1154,6 +1157,11 @@ struct volume_options options[] = {
           .value = {"on", "off"},
           .default_value = "off"
         },
+        { .key  = {"ctr_link_consistency"},
+          .type = GF_OPTION_TYPE_BOOL,
+          .value = {"on", "off"},
+          .default_value = "off"
+        },
         { .key  = {"hot-brick"},
           .type = GF_OPTION_TYPE_BOOL,
           .value = {"on", "off"},
diff --git a/xlators/features/changetimerecorder/src/ctr-helper.c b/xlators/features/changetimerecorder/src/ctr-helper.c
index da5a3deed96..a2e2569818d 100644
--- a/xlators/features/changetimerecorder/src/ctr-helper.c
+++ b/xlators/features/changetimerecorder/src/ctr-helper.c
@@ -97,6 +97,7 @@ fill_db_record_for_wind (xlator_t               *this,
         ctr_wtime = &CTR_DB_REC(ctr_local).gfdb_wind_change_time;
         CTR_DB_REC(ctr_local).gfdb_fop_path = ctr_inode_cx->fop_path;
         CTR_DB_REC(ctr_local).gfdb_fop_type = ctr_inode_cx->fop_type;
+        CTR_DB_REC(ctr_local).link_consistency = _priv->ctr_link_consistency;
 
         ret = gettimeofday (ctr_wtime, NULL);
         if (ret) {
@@ -265,6 +266,10 @@ int extract_ctr_options (xlator_t *this, gf_ctr_private_t *_priv) {
         GF_OPTION_INIT ("record-counters", _priv->ctr_record_counter, bool,
                         out);
 
+        /*Extract flag for link consistency*/
+        GF_OPTION_INIT ("ctr_link_consistency", _priv->ctr_link_consistency,
+                        bool, out);
+
         /*Extract flag for hot tier brick*/
         GF_OPTION_INIT ("hot-brick", _priv->ctr_hot_brick, bool, out);
 
diff --git a/xlators/features/changetimerecorder/src/ctr-helper.h b/xlators/features/changetimerecorder/src/ctr-helper.h
index 6f4ad5cbe7a..85ee872fa33 100644
--- a/xlators/features/changetimerecorder/src/ctr-helper.h
+++ b/xlators/features/changetimerecorder/src/ctr-helper.h
@@ -37,6 +37,7 @@ typedef struct gf_ctr_private {
         gf_boolean_t                    ctr_record_wind;
         gf_boolean_t                    ctr_record_unwind;
         gf_boolean_t                    ctr_record_counter;
+        gf_boolean_t                    ctr_link_consistency;
         gfdb_db_type_t                  gfdb_db_type;
         gfdb_sync_type_t                gfdb_sync_type;
         gfdb_conn_node_t                *_db_conn;
@@ -380,6 +381,7 @@ ctr_insert_wind (call_frame_t                    *frame,
                                 "WIND: Error filling  ctr local");
                         goto out;
                 }
+
                 /*Insert the db record*/
                 ret = insert_record (_priv->_db_conn,
                                         &ctr_local->gfdb_db_record);
diff --git a/xlators/mgmt/glusterd/src/glusterd-volgen.c b/xlators/mgmt/glusterd/src/glusterd-volgen.c
index b6c0e4926d3..f8050c87124 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volgen.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volgen.c
@@ -1689,6 +1689,10 @@ brick_graph_add_changetimerecorder (volgen_graph_t *graph,
         if (ret)
                 goto out;
 
+        ret = xlator_set_option (xl, "ctr_link_consistency", "off");
+        if (ret)
+                goto out;
+
         ret = xlator_set_option (xl, "record-entry", "on");
         if (ret)
                 goto out;
diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-set.c b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
index 17f34e6f86d..c57840f0fde 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volume-set.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
@@ -1730,8 +1730,13 @@ struct volopt_map_entry glusterd_volopt_map[] = {
           .option      = "record-counters",
           .op_version  = GD_OP_VERSION_3_7_0
         },
+        { .key         = "features.ctr_link_consistency",
+          .voltype     = "features/changetimerecorder",
+          .value       = "off",
+          .option      = "ctr_link_consistency",
+          .op_version  = GD_OP_VERSION_3_7_0
+        },
 #endif /* USE_GFDB */
-
         { .key         = "locks.trace",
           .voltype     = "features/locks",
           .type        = NO_DOC,