Changeset: 5c6184a59d69 for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB?cmd=changeset;node=5c6184a59d69
Modified Files:
sql/storage/objectset.c
Branch: default
Log Message:
Clean up remaining known synchronicity issues in objectset.
diffs (truncated from 499 to 300 lines):
diff --git a/sql/storage/objectset.c b/sql/storage/objectset.c
--- a/sql/storage/objectset.c
+++ b/sql/storage/objectset.c
@@ -14,11 +14,17 @@
struct versionhead ;
-#define active (0)
-#define under_destruction (1<<1)
-#define block_destruction (1<<2)
-#define deleted (1<<3)
-#define rollbacked (1<<4)
+#define active (0)
+#define under_destruction (1<<1)
+#define block_destruction (1<<2)
+#define deleted (1<<3)
+#define rollbacked (1<<4)
+
+/* This objectversion owns its associated versionhead.
+ * When this objectversion gets destroyed,
+ * the cleanup procedure should also destroy the associated (name|id) based
versionhead.*/
+#define name_based_versionhead_owner (1<<5)
+#define id_based_versionhead_owner (1<<6)
typedef struct objectversion {
ulng ts;
@@ -40,7 +46,6 @@ typedef struct versionhead {
objectversion* ov;
} versionhead ;
-// TODO: this might be moved to the objectversion struct itself.
typedef struct RW_lock { //readers-writer lock
int reader_cnt;
MT_Lock readers_lock;
@@ -102,28 +107,20 @@ unlock_writer(objectset* os)
MT_lock_unset(&os->rw_lock.general_lock);
}
+static bte os_atmc_get_state(objectversion *ov) {
+ bte state = (bte) ATOMIC_GET(&ov->state);
+ return state;
+}
+
+static void os_atmc_set_state(objectversion *ov, bte state) {
+ ATOMIC_SET(&ov->state, state);
+}
+
static versionhead *
find_id(objectset *os, sqlid id)
{
if (os) {
- lock_writer(os);
- if ((!os->id_map || os->id_map->size*16 < os->id_based_cnt) &&
os->id_based_cnt > HASH_MIN_SIZE && os->sa) {
- hash_destroy(os->id_map);
- os->id_map = hash_new(os->sa, os->id_based_cnt,
(fkeyvalue)&os_id_key);
- if (os->id_map == NULL) {
- unlock_writer(os);
- return NULL;
- }
-
- for (versionhead *n = os->id_based_h; n; n = n->next )
{
- int key = os_id_key(n);
-
- if (hash_add(os->id_map, key, n) == NULL) {
- unlock_writer(os);
- return NULL;
- }
- }
- }
+ lock_reader(os);
if (os->id_map) {
int key = (int) BATatoms[TYPE_int].atomHash(&id);
sql_hash_e *he =
os->id_map->buckets[key&(os->id_map->size-1)];
@@ -132,23 +129,26 @@ find_id(objectset *os, sqlid id)
versionhead *n = he->value;
if (n && n->ov->b->id == id) {
- unlock_writer(os);
+ unlock_reader(os);
return n;
}
}
- unlock_writer(os);
+ unlock_reader(os);
return NULL;
}
- unlock_writer(os);
+
for (versionhead *n = os->id_based_h; n; n = n->next) {
objectversion *ov = n->ov;
/* check if ids match */
if (id == ov->b->id) {
+ unlock_reader(os);
return n;
}
}
}
+
+ unlock_reader(os);
return NULL;
}
@@ -179,11 +179,11 @@ node_destroy(objectset *os, sqlstore *st
}
static versionhead *
-os_remove_name_based_chain(objectset *os, sqlstore *store, versionhead *n)
+os_remove_name_based_chain(objectset *os, objectversion* ov)
{
- assert(n);
+ lock_writer(os);
+ versionhead *n = ov->name_based_head;
versionhead *p = os->name_based_h;
-
if (p != n)
while (p && p->next != n)
p = p->next;
@@ -201,19 +201,23 @@ os_remove_name_based_chain(objectset *os
if (n == os->name_based_t)
os->name_based_t = p;
- lock_writer(os);
if (os->name_map && n)
hash_delete(os->name_map, n);
+
+ os->name_based_cnt--;
unlock_writer(os);
- node_destroy(os, store, n);
+ bte state = os_atmc_get_state(ov);
+ state |= name_based_versionhead_owner;
+ os_atmc_set_state(ov, state);
return p;
}
static versionhead *
-os_remove_id_based_chain(objectset *os, sqlstore *store, versionhead *n)
+os_remove_id_based_chain(objectset *os, objectversion* ov)
{
- assert(n);
+ lock_writer(os);
+ versionhead *n = ov->id_based_head;
versionhead *p = os->id_based_h;
if (p != n)
@@ -233,12 +237,15 @@ os_remove_id_based_chain(objectset *os,
if (n == os->id_based_t)
os->id_based_t = p;
- lock_writer(os);
if (os->id_map && n)
hash_delete(os->id_map, n);
+
+ os->name_based_cnt--;
unlock_writer(os);
- node_destroy(os, store, n);
+ bte state = os_atmc_get_state(ov);
+ state |= id_based_versionhead_owner;
+ os_atmc_set_state(ov, state);
return p;
}
@@ -255,11 +262,34 @@ node_create(sql_allocator *sa, objectver
return n;
}
+static int
+os_name_key(versionhead *n)
+{
+ return hash_key(n->ov->b->name);
+}
+
static objectset *
-
os_append_node_name(objectset *os, versionhead *n)
{
lock_writer(os);
+ if ((!os->name_map || os->name_map->size*16 < os->name_based_cnt) &&
os->name_based_cnt > HASH_MIN_SIZE && os->sa) {
+ hash_destroy(os->name_map);
+ os->name_map = hash_new(os->sa, os->name_based_cnt,
(fkeyvalue)& os_name_key);
+ if (os->name_map == NULL) {
+ unlock_writer(os);
+ return NULL;
+ }
+
+ for (versionhead *n = os->name_based_h; n; n = n->next ) {
+ int key = os_name_key(n);
+
+ if (hash_add(os->name_map, key, n) == NULL) {
+ unlock_writer(os);
+ return NULL;
+ }
+ }
+ }
+
if (os->name_map) {
int key = os->name_map->key(n);
@@ -268,6 +298,7 @@ os_append_node_name(objectset *os, versi
return NULL;
}
}
+
if (os->name_based_t) {
os->name_based_t->next = n;
} else {
@@ -275,8 +306,8 @@ os_append_node_name(objectset *os, versi
}
n->prev = os->name_based_t; // aka the double linked list.
os->name_based_t = n;
+ os->name_based_cnt++;
unlock_writer(os);
- os->name_based_cnt++;
return os;
}
@@ -301,14 +332,31 @@ static objectset *
os_append_node_id(objectset *os, versionhead *n)
{
lock_writer(os);
+ if ((!os->id_map || (os->id_map->size*16 < os->id_based_cnt &&
os->id_based_cnt > HASH_MIN_SIZE)) && os->sa) {
+ hash_destroy(os->id_map);
+ os->id_map = hash_new(os->sa, os->id_based_cnt,
(fkeyvalue)&os_id_key);
+ if (os->id_map == NULL) {
+ unlock_writer(os);
+ return NULL;
+ }
+ for (versionhead *n = os->id_based_h; n; n = n->next ) {
+ int key = os_id_key(n);
+
+ if (hash_add(os->id_map, key, n) == NULL) {
+ unlock_writer(os);
+ return NULL;
+ }
+ }
+ }
+
if (os->id_map) {
int key = os->id_map->key(n);
-
if (hash_add(os->id_map, key, n) == NULL) {
unlock_writer(os);
return NULL;
}
}
+
if (os->id_based_t) {
os->id_based_t->next = n;
} else {
@@ -316,8 +364,8 @@ os_append_node_id(objectset *os, version
}
n->prev = os->id_based_t; // aka the double linked list.
os->id_based_t = n;
+ os->id_based_cnt++;
unlock_writer(os);
- os->id_based_cnt++;
return os;
}
@@ -342,21 +390,23 @@ static versionhead * find_name(objectset
static void
objectversion_destroy(sqlstore *store, objectset* os, objectversion *ov)
{
+
+ bte state = os_atmc_get_state(ov);
+
+ if (state & name_based_versionhead_owner) {
+ node_destroy(ov->os, store, ov->name_based_head);
+ }
+
+ if (state & id_based_versionhead_owner) {
+ node_destroy(ov->os, store, ov->id_based_head);
+ }
+
if (os->destroy)
os->destroy(store, ov->b);
_DELETE(ov);
}
-static bte os_atmc_get_state(objectversion *ov) {
- bte state = (bte) ATOMIC_GET(&ov->state);
- return state;
-}
-
-static void os_atmc_set_state(objectversion *ov, bte state) {
- ATOMIC_SET(&ov->state, state);
-}
-
static inline objectversion*
get_name_based_older_locked(objectversion* ov) {
lock_reader(ov->os);
@@ -414,7 +464,7 @@ static void
else if (!name_based_older) {
// this is a terminal node. i.e. this objectversion does not
have name based committed history
if (ov->name_based_head) // The oposite can happen during an
early conflict in os_add or os_del.
- os_remove_name_based_chain(ov->os, store,
ov->name_based_head);
+ os_remove_name_based_chain(ov->os, ov);
_______________________________________________
checkin-list mailing list
checkin-list@monetdb.org
https://www.monetdb.org/mailman/listinfo/checkin-list
