summaryrefslogtreecommitdiffstats
path: root/xlators
diff options
context:
space:
mode:
authorJoseph Fernandes <josferna@redhat.com>2015-09-11 09:36:06 -0400
committerDan Lambright <dlambrig@redhat.com>2015-09-11 13:46:50 -0700
commit1d02d4bd9b5f7d730ab08961c17ef58204c8e8fd (patch)
treeb37cf8fffe901013185e8011ea32b680640871d1 /xlators
parent46674c5d5caaa183f8ee99efb64ef268eded91ab (diff)
tier/ctr: Solving DB Lock issue due to write contention from db connections
This is a backport of 12031. > Problem: The DB on the brick is been accessed by CTR, for write and > tier migrator, for read and write. The write from tier migrator is reseting > the heat counters after a cycle. Since we are using sqlite, two connections > trying to write would cause a db lock contention. As a result CTR used to fail > to update the db. > Solution: Using the same db connection of CTR for reseting the heat counters. > 1) Introducted a new IPC FOP for CTR > 2) After the query do a ipc syncop to the underlying client xlator associated > to the brick. > 3) CTR in brick will catch the IPC FOP and cleat the heat counters. > Change-Id: I53306bfc08dcdba479deb4ccc154896521336150 > BUG: 1260730 > Signed-off-by: Joseph Fernandes <josferna@redhat.com> > Reviewed-on: http://review.gluster.org/12031 > Tested-by: NetBSD Build System <jenkins@build.gluster.org> > Tested-by: Gluster Build System <jenkins@build.gluster.com> Signed-off-by: Dan Lambright <dlambrig@redhat.com> Conflicts: xlators/cluster/dht/src/tier.c Change-Id: I88aa289cdf21e216b42c3d8ccfb4e7e828b43772 BUG: 1262341 Reviewed-on: http://review.gluster.org/12161 Reviewed-by: Joseph Fernandes Tested-by: NetBSD Build System <jenkins@build.gluster.org> Tested-by: Gluster Build System <jenkins@build.gluster.com> Tested-by: Dan Lambright <dlambrig@redhat.com>
Diffstat (limited to 'xlators')
-rw-r--r--xlators/cluster/dht/src/dht-mem-types.h1
-rw-r--r--xlators/cluster/dht/src/tier.c135
-rw-r--r--xlators/cluster/dht/src/tier.h8
-rw-r--r--xlators/features/changetimerecorder/src/changetimerecorder.c34
-rw-r--r--xlators/features/changetimerecorder/src/ctr-helper.h1
5 files changed, 137 insertions, 42 deletions
diff --git a/xlators/cluster/dht/src/dht-mem-types.h b/xlators/cluster/dht/src/dht-mem-types.h
index 85e5baed62c..e3a38ed7e03 100644
--- a/xlators/cluster/dht/src/dht-mem-types.h
+++ b/xlators/cluster/dht/src/dht-mem-types.h
@@ -34,6 +34,7 @@ enum gf_dht_mem_types_ {
gf_dht_mt_container_t,
gf_dht_mt_octx_t,
gf_dht_mt_miginfo_t,
+ gf_tier_mt_bricklist_t,
gf_dht_mt_end
};
#endif
diff --git a/xlators/cluster/dht/src/tier.c b/xlators/cluster/dht/src/tier.c
index 3e3688861cc..1afbbd86634 100644
--- a/xlators/cluster/dht/src/tier.c
+++ b/xlators/cluster/dht/src/tier.c
@@ -31,7 +31,7 @@ static void *libhandle;
static gfdb_methods_t gfdb_methods;
#define DB_QUERY_RECORD_SIZE 4096
-
+#define PROMOTION_CYCLE_CNT 4
static int
@@ -440,8 +440,7 @@ out:
* It picks up each bricks db and queries for eligible files for migration.
* The list of eligible files are populated in appropriate query files*/
static int
-tier_process_brick_cbk (dict_t *brick_dict, char *key, data_t *value,
- void *args) {
+tier_process_brick_cbk (brick_list_t *local_brick, void *args) {
int ret = -1;
char *db_path = NULL;
query_cbk_args_t *query_cbk_args = NULL;
@@ -460,8 +459,12 @@ tier_process_brick_cbk (dict_t *brick_dict, char *key, data_t *value,
GF_VALIDATE_OR_GOTO (this->name,
gfdb_brick_dict_info->_query_cbk_args, out);
- GF_VALIDATE_OR_GOTO (this->name, value, out);
- db_path = data_to_str(value);
+ GF_VALIDATE_OR_GOTO (this->name, local_brick, out);
+
+ GF_VALIDATE_OR_GOTO (this->name, local_brick->xlator, out);
+
+ GF_VALIDATE_OR_GOTO (this->name, local_brick->brick_db_path, out);
+ db_path = local_brick->brick_db_path;
/*Preparing DB parameters before init_db i.e getting db connection*/
params_dict = dict_new ();
@@ -471,7 +474,7 @@ tier_process_brick_cbk (dict_t *brick_dict, char *key, data_t *value,
"DB Params cannot initialized!");
goto out;
}
- SET_DB_PARAM_TO_DICT(this->name, params_dict, gfdb_methods.dbpath,
+ SET_DB_PARAM_TO_DICT(this->name, params_dict, (char *) gfdb_methods.get_db_path(),
db_path, ret, out);
/*Get the db connection*/
@@ -539,7 +542,17 @@ tier_process_brick_cbk (dict_t *brick_dict, char *key, data_t *value,
DHT_MSG_LOG_TIER_ERROR,
"FATAL: query from db failed");
goto out;
- }
+ }
+
+ /*Clear the heat on the DB entries*/
+ ret = syncop_ipc (local_brick->xlator, GF_IPC_TARGET_CTR, NULL, NULL);
+ if (ret) {
+ gf_msg (this->name, GF_LOG_ERROR, 0,
+ DHT_MSG_LOG_TIER_ERROR, "Failed clearing the heat "
+ "on db %s", local_brick->brick_db_path);
+ goto out;
+ }
+
ret = 0;
out:
if (query_cbk_args->queryFILE) {
@@ -559,6 +572,7 @@ tier_build_migration_qfile (demotion_args_t *args,
_gfdb_brick_dict_info_t gfdb_brick_dict_info;
gfdb_time_t time_in_past;
int ret = -1;
+ brick_list_t *local_brick = NULL;
/*
* The first time this function is called, query file will
@@ -589,14 +603,18 @@ tier_build_migration_qfile (demotion_args_t *args,
gfdb_brick_dict_info.time_stamp = &time_in_past;
gfdb_brick_dict_info._gfdb_promote = is_promotion;
gfdb_brick_dict_info._query_cbk_args = query_cbk_args;
- ret = dict_foreach (args->brick_list, tier_process_brick_cbk,
- &gfdb_brick_dict_info);
- if (ret) {
- gf_msg (args->this->name, GF_LOG_ERROR, 0,
- DHT_MSG_BRICK_QUERY_FAILED,
- "Brick query failed\n");
- goto out;
+
+ list_for_each_entry (local_brick, args->brick_list, list) {
+ ret = tier_process_brick_cbk (local_brick,
+ &gfdb_brick_dict_info);
+ if (ret) {
+ gf_msg (args->this->name, GF_LOG_ERROR, 0,
+ DHT_MSG_BRICK_QUERY_FAILED,
+ "Brick query failed\n");
+ goto out;
+ }
}
+ ret = 0;
out:
return ret;
}
@@ -701,19 +719,19 @@ out:
}
static int
-tier_get_bricklist (xlator_t *xl, dict_t *bricklist)
+tier_get_bricklist (xlator_t *xl, struct list_head *local_bricklist_head)
{
xlator_list_t *child = NULL;
char *rv = NULL;
char *rh = NULL;
char localhost[256] = {0};
- char *db_path = "";
char *brickname = NULL;
char db_name[PATH_MAX] = "";
int ret = 0;
+ brick_list_t *local_brick = NULL;
GF_VALIDATE_OR_GOTO ("tier", xl, out);
- GF_VALIDATE_OR_GOTO ("tier", bricklist, out);
+ GF_VALIDATE_OR_GOTO ("tier", local_bricklist_head, out);
gethostname (localhost, sizeof (localhost));
@@ -728,27 +746,38 @@ tier_get_bricklist (xlator_t *xl, dict_t *bricklist)
if (gf_is_local_addr (rh)) {
+ local_brick = GF_CALLOC (1, sizeof(brick_list_t),
+ gf_tier_mt_bricklist_t);
+ if (!local_brick) {
+ goto out;
+ }
+
ret = dict_get_str(xl->options, "remote-subvolume",
&rv);
if (ret < 0)
goto out;
+
brickname = strrchr(rv, '/') + 1;
snprintf(db_name, sizeof(db_name), "%s.db",
brickname);
- db_path = GF_CALLOC (PATH_MAX, 1, gf_common_mt_char);
- if (!db_path) {
+
+ local_brick->brick_db_path =
+ GF_CALLOC (PATH_MAX, 1, gf_common_mt_char);
+ if (!local_brick->brick_db_path) {
gf_msg ("tier", GF_LOG_ERROR, 0,
DHT_MSG_LOG_TIER_STATUS,
"Failed. to allocate memory for bricklist");
goto out;
}
- sprintf(db_path, "%s/%s/%s", rv,
+ sprintf(local_brick->brick_db_path, "%s/%s/%s", rv,
GF_HIDDEN_PATH,
db_name);
- if (dict_add_dynstr_with_alloc(bricklist, "brick",
- db_path))
- goto out;
+
+ local_brick->xlator = xl;
+
+ list_add_tail (&(local_brick->list),
+ local_bricklist_head);
ret = 0;
goto out;
@@ -756,17 +785,48 @@ tier_get_bricklist (xlator_t *xl, dict_t *bricklist)
}
for (child = xl->children; child; child = child->next) {
- ret = tier_get_bricklist(child->xlator, bricklist);
+ ret = tier_get_bricklist(child->xlator, local_bricklist_head);
+ if (ret) {
+ goto out;
+ }
}
+
+ ret = 0;
out:
+
+ if (ret) {
+ if (local_brick) {
+ GF_FREE (local_brick->brick_db_path);
+ }
+ GF_FREE (local_brick);
+ }
+
return ret;
}
+void
+clear_bricklist (struct list_head *brick_list)
+{
+ brick_list_t *local_brick = NULL;
+ brick_list_t *temp = NULL;
+
+ if (list_empty(brick_list)) {
+ return;
+ }
+
+ list_for_each_entry_safe (local_brick, temp, brick_list, list) {
+ list_del (&local_brick->list);
+ GF_FREE (local_brick->brick_db_path);
+ GF_FREE (local_brick);
+ }
+}
+
+
int
tier_start (xlator_t *this, gf_defrag_info_t *defrag)
{
- dict_t *bricklist_cold = NULL;
- dict_t *bricklist_hot = NULL;
+ struct list_head bricklist_hot = { 0 };
+ struct list_head bricklist_cold = { 0 };
dht_conf_t *conf = NULL;
gfdb_time_t current_time;
int freq_promote = 0;
@@ -783,16 +843,11 @@ tier_start (xlator_t *this, gf_defrag_info_t *defrag)
conf = this->private;
- bricklist_cold = dict_new();
- if (!bricklist_cold)
- return -1;
-
- bricklist_hot = dict_new();
- if (!bricklist_hot)
- return -1;
+ INIT_LIST_HEAD ((&bricklist_hot));
+ INIT_LIST_HEAD ((&bricklist_cold));
- tier_get_bricklist (conf->subvolumes[0], bricklist_cold);
- tier_get_bricklist (conf->subvolumes[1], bricklist_hot);
+ tier_get_bricklist (conf->subvolumes[0], &bricklist_cold);
+ tier_get_bricklist (conf->subvolumes[1], &bricklist_hot);
gf_msg (this->name, GF_LOG_INFO, 0,
DHT_MSG_LOG_TIER_STATUS, "Begin run tier promote %d"
@@ -860,7 +915,7 @@ tier_start (xlator_t *this, gf_defrag_info_t *defrag)
if (is_demotion_triggered) {
demotion_args.this = this;
- demotion_args.brick_list = bricklist_hot;
+ demotion_args.brick_list = &bricklist_hot;
demotion_args.defrag = defrag;
demotion_args.freq_time = freq_demote;
ret_demotion = pthread_create (&demote_thread,
@@ -876,9 +931,9 @@ tier_start (xlator_t *this, gf_defrag_info_t *defrag)
if (is_promotion_triggered) {
promotion_args.this = this;
- promotion_args.brick_list = bricklist_cold;
+ promotion_args.brick_list = &bricklist_cold;
promotion_args.defrag = defrag;
- promotion_args.freq_time = freq_promote;
+ promotion_args.freq_time = freq_promote * PROMOTION_CYCLE_CNT;
ret_promotion = pthread_create (&promote_thread,
NULL, &tier_promote,
&promotion_args);
@@ -927,8 +982,8 @@ tier_start (xlator_t *this, gf_defrag_info_t *defrag)
ret = 0;
out:
- dict_unref(bricklist_cold);
- dict_unref(bricklist_hot);
+ clear_bricklist (&bricklist_cold);
+ clear_bricklist (&bricklist_hot);
return ret;
}
diff --git a/xlators/cluster/dht/src/tier.h b/xlators/cluster/dht/src/tier.h
index 34631b0a841..d5fbba655e5 100644
--- a/xlators/cluster/dht/src/tier.h
+++ b/xlators/cluster/dht/src/tier.h
@@ -61,10 +61,16 @@ typedef struct _gfdb_brick_dict_info {
query_cbk_args_t *_query_cbk_args;
} _gfdb_brick_dict_info_t;
+typedef struct brick_list {
+ xlator_t *xlator;
+ char *brick_db_path;
+ struct list_head list;
+} brick_list_t;
+
typedef struct _dm_thread_args {
xlator_t *this;
gf_defrag_info_t *defrag;
- dict_t *brick_list;
+ struct list_head *brick_list;
int freq_time;
int return_value;
} promotion_args_t, demotion_args_t;
diff --git a/xlators/features/changetimerecorder/src/changetimerecorder.c b/xlators/features/changetimerecorder/src/changetimerecorder.c
index 7fd62584752..8acd811a0cb 100644
--- a/xlators/features/changetimerecorder/src/changetimerecorder.c
+++ b/xlators/features/changetimerecorder/src/changetimerecorder.c
@@ -1379,6 +1379,36 @@ out:
fd, size, off, flags, xdata);
return 0;
}
+
+/*******************************ctr_ipc****************************************/
+
+/* IPC Call from tier migrator to clear the heat on the DB */
+int32_t
+ctr_ipc (call_frame_t *frame, xlator_t *this, int32_t op, dict_t *xdata)
+{
+ int ret = -1;
+ gf_ctr_private_t *_priv = NULL;
+
+ GF_ASSERT(this);
+ _priv = this->private;
+ GF_ASSERT (_priv);
+ GF_ASSERT(_priv->_db_conn);
+
+ if (op != GF_IPC_TARGET_CTR)
+ goto wind;
+
+ ret = clear_files_heat (_priv->_db_conn);
+
+ STACK_UNWIND_STRICT (ipc, frame, ret, 0, NULL);
+ return 0;
+
+ wind:
+ STACK_WIND (frame, default_ipc_cbk, FIRST_CHILD (this),
+ FIRST_CHILD (this)->fops->ipc, op, xdata);
+ return 0;
+}
+
+
/******************************************************************************/
int
@@ -1596,7 +1626,9 @@ struct xlator_fops fops = {
.writev = ctr_writev,
.setattr = ctr_setattr,
/*read fops*/
- .readv = ctr_readv
+ .readv = ctr_readv,
+ /* IPC call*/
+ .ipc = ctr_ipc
};
struct xlator_cbks cbks = {
diff --git a/xlators/features/changetimerecorder/src/ctr-helper.h b/xlators/features/changetimerecorder/src/ctr-helper.h
index 2a1f5203d9b..d6521d26b3d 100644
--- a/xlators/features/changetimerecorder/src/ctr-helper.h
+++ b/xlators/features/changetimerecorder/src/ctr-helper.h
@@ -22,6 +22,7 @@
#include "iatt.h"
#include "glusterfs.h"
#include "xlator.h"
+#include "defaults.h"
#include "logging.h"
#include "common-utils.h"
#include <time.h>