cluster/tier: handle fast demotions

Demote files on priority if hi-watermark has been breached and continue
to demote until the watermark drops below hi-watermark.

Monitor watermark more frequently.
Trigger demotion as soon as hi-watermark is breached.
Add cluster.tier-emergency-demote-query-limit option to limit number
of files returned from the database query for every iteration of
tier_migrate_using_query_file(). If watermark hasn't dropped below
hi-watermark during the first iteration, the next iteration will be
triggered approximately 1 second after tier_demote() returns to the
main tiering loop.
Update changetimerecorder xlator to handle query for emergency demote
mode.

Add tier-ctr-interface.h:
Move tier and ctr interface specific macros and struct definition from
libglusterfs/src/gfdb/gfdb_data_store.h to new header
libglusterfs/src/tier-ctr-interface.h

Change-Id: If56af78c6c81d37529b9b6e65ae606ba5c99a811
BUG: 1366648
Signed-off-by: Milind Changire <mchangir@redhat.com>
Reviewed-on: http://review.gluster.org/15158
Smoke: Gluster Build System <jenkins@build.gluster.org>
CentOS-regression: Gluster Build System <jenkins@build.gluster.org>
NetBSD-regression: NetBSD Build System <jenkins@build.gluster.org>
Reviewed-by: Dan Lambright <dlambrig@redhat.com>

6 files changed, 111 insertions, 20 deletions

diff --git a/xlators/cluster/dht/src/dht-common.h b/xlators/cluster/dht/src/dht-common.h
index 7adc849fb63..719d214f92d 100644
--- a/xlators/cluster/dht/src/dht-common.h
+++ b/xlators/cluster/dht/src/dht-common.h
@@ -396,6 +396,7 @@ typedef struct gf_tier_conf {
         int                          percent_full;
         uint64_t                     max_migrate_bytes;
         int                          max_migrate_files;
+        int                          query_limit;
         tier_mode_t                  mode;
         /* These flags are only used for tier-compact */
         gf_boolean_t                 compact_active;
diff --git a/xlators/cluster/dht/src/dht-shared.c b/xlators/cluster/dht/src/dht-shared.c
index f410f71b5a6..46bf461cf63 100644
--- a/xlators/cluster/dht/src/dht-shared.c
+++ b/xlators/cluster/dht/src/dht-shared.c
@@ -1064,6 +1064,10 @@ struct volume_options options[] = {
           .type = GF_OPTION_TYPE_INT,
           .default_value = "10000",
         },
+        { .key         = {"tier-query-limit"},
+          .type = GF_OPTION_TYPE_INT,
+          .default_value = "100",
+        },
         /* switch option */
         { .key  = {"pattern.switch.case"},
           .type = GF_OPTION_TYPE_ANY
diff --git a/xlators/cluster/dht/src/tier.c b/xlators/cluster/dht/src/tier.c
index 2f8eddbc07d..60e967a88a2 100644
--- a/xlators/cluster/dht/src/tier.c
+++ b/xlators/cluster/dht/src/tier.c
@@ -15,6 +15,7 @@
 #include "tier-common.h"
 #include "syscall.h"
 #include "events.h"
+#include "tier-ctr-interface.h"
 
 /*Hard coded DB info*/
 static gfdb_db_type_t dht_tier_db_type = GFDB_SQLITE3;
@@ -193,6 +194,7 @@ out:
 
 /* Check and update the watermark every WM_INTERVAL seconds */
 #define WM_INTERVAL            5
+#define WM_INTERVAL_EMERG      1
 
 static int
 tier_check_same_node (xlator_t *this, loc_t *loc, gf_defrag_info_t *defrag)
@@ -523,7 +525,7 @@ tier_can_promote_file (xlator_t *this, char const *file_name,
                                 defrag->tier_conf.blocks_used;
 
         /* test if the estimated block usage goes above HI watermark */
-        if (GF_PERCENTAGE (estimated_usage, defrag->tier_conf.blocks_total) >
+        if (GF_PERCENTAGE (estimated_usage, defrag->tier_conf.blocks_total) >=
                         defrag->tier_conf.watermark_hi) {
                 gf_msg (this->name, GF_LOG_INFO, 0,
                         DHT_MSG_LOG_TIER_STATUS,
@@ -575,6 +577,7 @@ tier_migrate_using_query_file (void *_args)
         gfdb_time_t  current_time               = { 0 };
         int total_time                          = 0;
         int max_time                            = 0;
+        gf_boolean_t emergency_demote_mode      = _gf_false;
 
 
         GF_VALIDATE_OR_GOTO ("tier", query_cbk_args, out);
@@ -591,6 +594,9 @@ tier_migrate_using_query_file (void *_args)
         if (!migrate_data)
                 goto out;
 
+        emergency_demote_mode = (!query_cbk_args->is_promotion &&
+                                 is_hot_tier_full(&defrag->tier_conf));
+
         xdata_request = dict_new ();
         if (!xdata_request) {
                 gf_msg (this->name, GF_LOG_ERROR, 0,
@@ -1013,6 +1019,18 @@ per_file_out:
 
                 gfdb_methods.gfdb_query_record_free (query_record);
                 query_record = NULL;
+
+                /* If we are demoting and the entry watermark was HI, then
+                 * we are done with emergency demotions if the current
+                 * watermark has fallen below hi-watermark level
+                 */
+                if (emergency_demote_mode) {
+                        if (tier_check_watermark (this) == 0) {
+                                if (!is_hot_tier_full (&defrag->tier_conf)) {
+                                        break;
+                                }
+                        }
+                }
         }
 
 out:
@@ -1126,14 +1144,23 @@ tier_process_self_query (tier_brick_list_t *local_brick, void *args)
                 goto out;
         }
         if (!gfdb_brick_info->_gfdb_promote) {
-                if (query_cbk_args->defrag->write_freq_threshold == 0 &&
-                        query_cbk_args->defrag->read_freq_threshold == 0) {
-                                ret = gfdb_methods.find_unchanged_for_time (
-                                        conn_node,
-                                        tier_gf_query_callback,
-                                        (void *)query_cbk_args,
-                                        gfdb_brick_info->time_stamp);
+                if (query_cbk_args->defrag->tier_conf.watermark_last ==
+                        TIER_WM_HI) {
+                        /* emergency demotion mode */
+                        ret = gfdb_methods.find_all (conn_node,
+                                tier_gf_query_callback,
+                                (void *)query_cbk_args,
+                                query_cbk_args->defrag->tier_conf.
+                                        query_limit);
                 } else {
+                        if (query_cbk_args->defrag->write_freq_threshold == 0 &&
+                            query_cbk_args->defrag->read_freq_threshold == 0) {
+                                ret = gfdb_methods.find_unchanged_for_time (
+                                                conn_node,
+                                                tier_gf_query_callback,
+                                                (void *)query_cbk_args,
+                                                gfdb_brick_info->time_stamp);
+                        } else {
                                 ret = gfdb_methods.find_unchanged_for_time_freq (
                                         conn_node,
                                         tier_gf_query_callback,
@@ -1144,6 +1171,7 @@ tier_process_self_query (tier_brick_list_t *local_brick, void *args)
                                         query_cbk_args->defrag->
                                                         read_freq_threshold,
                                         _gf_false);
+                        }
                 }
         } else {
                 if (query_cbk_args->defrag->write_freq_threshold == 0 &&
@@ -1159,8 +1187,7 @@ tier_process_self_query (tier_brick_list_t *local_brick, void *args)
                                 tier_gf_query_callback,
                                 (void *)query_cbk_args,
                                 gfdb_brick_info->time_stamp,
-                                query_cbk_args->defrag->
-                                write_freq_threshold,
+                                query_cbk_args->defrag->write_freq_threshold,
                                 query_cbk_args->defrag->read_freq_threshold,
                                 _gf_false);
                 }
@@ -1267,10 +1294,21 @@ tier_process_ctr_query (tier_brick_list_t *local_brick, void *args)
 
         /* set all the query params*/
         ipc_ctr_params->is_promote = gfdb_brick_info->_gfdb_promote;
-        ipc_ctr_params->write_freq_threshold = query_cbk_args->
-                                                defrag->write_freq_threshold;
-        ipc_ctr_params->read_freq_threshold = query_cbk_args->
-                                                defrag->read_freq_threshold;
+
+        ipc_ctr_params->write_freq_threshold =
+                query_cbk_args->defrag->write_freq_threshold;
+
+        ipc_ctr_params->read_freq_threshold =
+                query_cbk_args->defrag->read_freq_threshold;
+
+        ipc_ctr_params->query_limit =
+                query_cbk_args->defrag->tier_conf.query_limit;
+
+        ipc_ctr_params->emergency_demote =
+                (!gfdb_brick_info->_gfdb_promote &&
+                 query_cbk_args->defrag->tier_conf.watermark_last ==
+                        TIER_WM_HI);
+
         memcpy (&ipc_ctr_params->time_stamp,
                 gfdb_brick_info->time_stamp,
                 sizeof (gfdb_time_t));
@@ -2208,6 +2246,15 @@ out:
         return ret;
 }
 
+static int
+tier_get_wm_interval(tier_mode_t mode, tier_watermark_op_t wm)
+{
+        if (mode == TIER_MODE_WM && wm == TIER_WM_HI)
+                return WM_INTERVAL_EMERG;
+
+        return WM_INTERVAL;
+}
+
 /*
  * Main tiering loop. This is called from the promotion and the
  * demotion threads spawned in tier_start().
@@ -2316,7 +2363,10 @@ static void
 
                 check_watermark++;
 
-                if (check_watermark >= WM_INTERVAL) {
+                /* emergency demotion requires frequent watermark monitoring */
+                if (check_watermark >=
+                        tier_get_wm_interval(tier_conf->mode,
+                                             tier_conf->watermark_last)) {
                         check_watermark = 0;
                         if (tier_conf->mode == TIER_MODE_WM) {
                                 ret = tier_get_fs_stat (this, &root_loc);
@@ -2828,6 +2878,15 @@ tier_init (xlator_t *this)
 
         defrag->tier_conf.max_migrate_files = freq;
 
+
+        ret = dict_get_int32 (this->options,
+                              "tier-query-limit",
+                              &(defrag->tier_conf.query_limit));
+        if (ret) {
+                defrag->tier_conf.query_limit =
+                        DEFAULT_TIER_QUERY_LIMIT;
+        }
+
         ret = dict_get_str (this->options,
                             "tier-compact", &mode);
 
@@ -3041,6 +3100,10 @@ tier_reconfigure (xlator_t *this, dict_t *options)
                                   defrag->tier_conf.max_migrate_files, options,
                                   int32, out);
 
+                GF_OPTION_RECONF ("tier-query-limit",
+                                  defrag->tier_conf.query_limit,
+                                  options, int32, out);
+
                 GF_OPTION_RECONF ("tier-pause",
                                   req_pause, options,
                                   bool, out);
diff --git a/xlators/cluster/dht/src/tier.h b/xlators/cluster/dht/src/tier.h
index ffb04173bd5..764860e6884 100644
--- a/xlators/cluster/dht/src/tier.h
+++ b/xlators/cluster/dht/src/tier.h
@@ -98,7 +98,7 @@ typedef enum tier_watermark_op_ {
 #define DEFAULT_DEMOTE_FREQ_SEC        120
 #define DEFAULT_HOT_COMPACT_FREQ_SEC   604800
 #define DEFAULT_COLD_COMPACT_FREQ_SEC  604800
-#define DEFAULT_DEMOTE_DEGRADED        10
+#define DEFAULT_DEMOTE_DEGRADED        1
 #define DEFAULT_WRITE_FREQ_SEC         0
 #define DEFAULT_READ_FREQ_SEC          0
 #define DEFAULT_WM_LOW                 75
@@ -107,5 +107,6 @@ typedef enum tier_watermark_op_ {
 #define DEFAULT_COMP_MODE              _gf_true
 #define DEFAULT_TIER_MAX_MIGRATE_MB    1000
 #define DEFAULT_TIER_MAX_MIGRATE_FILES 5000
+#define DEFAULT_TIER_QUERY_LIMIT       100
 
 #endif
diff --git a/xlators/features/changetimerecorder/src/changetimerecorder.c b/xlators/features/changetimerecorder/src/changetimerecorder.c
index 4e4ea851a86..ffc4625626b 100644
--- a/xlators/features/changetimerecorder/src/changetimerecorder.c
+++ b/xlators/features/changetimerecorder/src/changetimerecorder.c
@@ -16,6 +16,7 @@
 #include "syscall.h"
 
 #include "changetimerecorder.h"
+#include "tier-ctr-interface.h"
 
 /*******************************inode forget***********************************/
 
@@ -1726,14 +1727,21 @@ ctr_db_query (xlator_t *this,
                 goto out;
         }
         if (!ipc_ctr_params->is_promote) {
-                if (ipc_ctr_params->write_freq_threshold == 0 &&
-                        ipc_ctr_params->read_freq_threshold == 0) {
+                if (ipc_ctr_params->emergency_demote) {
+                        /* emergency demotion mode */
+                        ret = find_all (conn_node,
+                                ctr_db_query_callback,
+                                (void *)&query_cbk_args,
+                                ipc_ctr_params->query_limit);
+                } else {
+                        if (ipc_ctr_params->write_freq_threshold == 0 &&
+                                ipc_ctr_params->read_freq_threshold == 0) {
                                 ret = find_unchanged_for_time (
                                         conn_node,
                                         ctr_db_query_callback,
                                         (void *)&query_cbk_args,
                                         &ipc_ctr_params->time_stamp);
-                } else {
+                        } else {
                                 ret = find_unchanged_for_time_freq (
                                         conn_node,
                                         ctr_db_query_callback,
@@ -1742,6 +1750,7 @@ ctr_db_query (xlator_t *this,
                                         ipc_ctr_params->write_freq_threshold,
                                         ipc_ctr_params->read_freq_threshold,
                                         _gf_false);
+                        }
                 }
         } else {
                 if (ipc_ctr_params->write_freq_threshold == 0 &&
diff --git a/xlators/mgmt/glusterd/src/glusterd-volume-set.c b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
index 0c67135db3f..931f99ba5af 100644
--- a/xlators/mgmt/glusterd/src/glusterd-volume-set.c
+++ b/xlators/mgmt/glusterd/src/glusterd-volume-set.c
@@ -462,7 +462,8 @@ validate_tier (glusterd_volinfo_t *volinfo, dict_t *dict, char *key,
                    strstr (key, "tier-max-files") ||
                    strstr (key, "tier-demote-frequency") ||
                    strstr (key, "tier-hot-compact-frequency") ||
-                   strstr (key, "tier-cold-compact-frequency")) {
+                   strstr (key, "tier-cold-compact-frequency") ||
+                   strstr (key, "tier-query-limit")) {
                 if (origin_val < 1) {
                         snprintf (errstr, sizeof (errstr), "%s is not a "
                                   " compatible value. %s expects a positive "
@@ -2739,6 +2740,18 @@ struct volopt_map_entry glusterd_volopt_map[] = {
           .description = "The maximum number of files that may be migrated"
           " in any direction in a given cycle by a single node."
         },
+        { .key         = "cluster.tier-query-limit",
+          .voltype     = "cluster/tier",
+          .option      = "tier-query-limit",
+          .value       = "100",
+          .op_version  = GD_OP_VERSION_3_9_0,
+          .flags       = OPT_FLAG_CLIENT_OPT,
+          .validate_fn = validate_tier,
+          .type        = NO_DOC,
+          .description = "The maximum number of files that may be migrated "
+                         "during an emergency demote. An emergency condition "
+                         "is flagged when writes breach the hi-watermark."
+        },
         { .key         = "cluster.tier-compact",
           .voltype     = "cluster/tier",
           .option      = "tier-compact",