Changeset: 0d45b03e2bff for MonetDB
URL: https://dev.monetdb.org/hg/MonetDB/rev/0d45b03e2bff
Modified Files:
sql/storage/bat/bat_storage.c
sql/storage/objectset.c
sql/storage/sql_storage.h
sql/storage/store.c
Branch: default
Log Message:
Reapply changeset e9a9d667c7f9
diffs (248 lines):
diff --git a/sql/storage/bat/bat_storage.c b/sql/storage/bat/bat_storage.c
--- a/sql/storage/bat/bat_storage.c
+++ b/sql/storage/bat/bat_storage.c
@@ -4147,6 +4147,9 @@ commit_update_col( sql_trans *tr, sql_ch
ATOMIC_PTR_TYPE* data = &c->data;
int type = c->type.type->localtype;
+ if (isDeleted(c->t))
+ return LOG_OK;
+
return commit_update_delta(tr, change, t, base, data, type, commit_ts,
oldest);
}
@@ -4173,6 +4176,9 @@ commit_update_idx( sql_trans *tr, sql_ch
ATOMIC_PTR_TYPE* data = &i->data;
int type = (oid_index(i->type))?TYPE_oid:TYPE_lng;
+ if (isDeleted(i->t))
+ return LOG_OK;
+
return commit_update_delta(tr, change, t, base, data, type, commit_ts,
oldest);
}
@@ -4213,6 +4219,9 @@ commit_update_del( sql_trans *tr, sql_ch
sql_table *t = (sql_table*)change->obj;
storage *dbat = ATOMIC_PTR_GET(&t->data);
+ if (isDeleted(t))
+ return ok;
+
if (t->commit_action == CA_DELETE || t->commit_action == CA_DROP) {
assert(isTempTable(t));
if ((ok = clear_storage(tr, t, dbat)) == LOG_OK)
@@ -4410,7 +4419,7 @@ claim_segmentsV2(sql_trans *tr, sql_tabl
{
int in_transaction = segments_in_transaction(tr, t), ok = LOG_OK;
assert(s->segs);
- ulng oldest = store_oldest(tr->store);
+ ulng oldest = store_oldest(tr->store, NULL);
BUN slot = 0;
size_t total = cnt;
@@ -4470,7 +4479,7 @@ claim_segmentsV2(sql_trans *tr, sql_tabl
if (!locked)
unlock_table(tr->store, t->base.id);
- if (ok == LOG_OK) {
+ if (ok == LOG_OK) {
/* hard to only add this once per transaction (probably want to
change to once per new segment) */
if (!in_transaction) {
trans_add(tr, &t->base, s, &tc_gc_del,
&commit_update_del, NOT_TO_BE_LOGGED(t) ? NULL : &log_update_del);
@@ -4499,7 +4508,7 @@ claim_segments(sql_trans *tr, sql_table
return claim_segmentsV2(tr, t, s, cnt, offset, offsets, locked);
int in_transaction = segments_in_transaction(tr, t), ok = LOG_OK;
assert(s->segs);
- ulng oldest = store_oldest(tr->store);
+ ulng oldest = store_oldest(tr->store, NULL);
BUN slot = 0;
int reused = 0;
diff --git a/sql/storage/objectset.c b/sql/storage/objectset.c
--- a/sql/storage/objectset.c
+++ b/sql/storage/objectset.c
@@ -401,7 +401,7 @@ objectversion_destroy(sqlstore *store, o
node_destroy(ov->os, store, ov->id_based_head);
}
- if (os->destroy)
+ if (os->destroy && ov->b)
os->destroy(store, ov->b);
if (os->temporary && (state & deleted || state & under_destruction ||
state & rollbacked))
@@ -500,6 +500,18 @@ os_rollback(objectversion *ov, sqlstore
return LOG_OK;
}
+static void
+ov_destroy_obj_recursive(sqlstore* store, objectversion *ov)
+{
+ if (ov->id_based_older && ov->id_based_older == ov->name_based_older) {
+ ov_destroy_obj_recursive(store, ov->id_based_older);
+ }
+ if (ov->os->destroy && ov->b) {
+ ov->os->destroy(store, ov->b);
+ ov->b = NULL;
+ }
+}
+
static inline void
try_to_mark_deleted_for_destruction(sqlstore* store, objectversion *ov)
{
@@ -525,6 +537,7 @@ try_to_mark_deleted_for_destruction(sqls
}
ov->ts = store_get_timestamp(store)+1;
+ ov_destroy_obj_recursive(store, ov);
}
}
@@ -577,6 +590,7 @@ os_cleanup(sqlstore* store, objectversio
if (ov->ts <= oldest) {
// the oldest relevant state is deleted so lets try to
mark it as destroyed
try_to_mark_deleted_for_destruction(store, ov);
+ return LOG_OK+1;
}
// Keep it inplace on the cleanup list, either because it is
now marked for destruction or
@@ -603,14 +617,14 @@ os_cleanup(sqlstore* store, objectversio
static int
tc_gc_objectversion(sql_store store, sql_change *change, ulng oldest)
{
- assert(!change->handled);
+// assert(!change->handled);
objectversion *ov = (objectversion*)change->data;
if (oldest >= TRANSACTION_ID_BASE)
return 0;
int res = os_cleanup( (sqlstore*) store, ov, oldest);
change->handled = (res)?true:false;
- return res;
+ return res>=0?LOG_OK:LOG_ERR;
}
static int
diff --git a/sql/storage/sql_storage.h b/sql/storage/sql_storage.h
--- a/sql/storage/sql_storage.h
+++ b/sql/storage/sql_storage.h
@@ -345,7 +345,7 @@ extern lng store_hot_snapshot_to_stream(
extern ulng store_function_counter(struct sqlstore *store);
-extern ulng store_oldest(struct sqlstore *store);
+extern ulng store_oldest(struct sqlstore *store, sql_trans *tr);
extern ulng store_get_timestamp(struct sqlstore *store);
extern void store_manager(struct sqlstore *store);
diff --git a/sql/storage/store.c b/sql/storage/store.c
--- a/sql/storage/store.c
+++ b/sql/storage/store.c
@@ -52,8 +52,15 @@ store_transaction_id(sqlstore *store)
}
ulng
-store_oldest(sqlstore *store)
-{
+store_oldest(sqlstore *store, sql_trans *tr)
+{
+ if (tr && tr->ts == store->oldest) {
+ sql_session *s = store->active->h->data;
+ if (s->tr == tr && store->active->h->next) {
+ s = store->active->h->next->data;
+ return s->tr->ts;
+ }
+ }
return store->oldest;
}
@@ -2334,7 +2341,7 @@ id_hash_clear_older(sql_hash *h, ulng ol
}
static void
-store_pending_changes(sqlstore *store, ulng oldest)
+store_pending_changes(sqlstore *store, ulng oldest, sql_trans *tr)
{
ulng oldest_changes = store_get_timestamp(store);
if (!list_empty(store->changes)) { /* lets first cleanup old stuff */
@@ -2346,7 +2353,7 @@ store_pending_changes(sqlstore *store, u
if (c->cleanup(store, c, oldest)) {
list_remove_node(store->changes, store, n);
_DELETE(c);
- } else if (c->ts < oldest_changes) {
+ } else if (!c->handled && c->ts < oldest_changes) {
oldest_changes = c->ts;
}
n = next;
@@ -2356,7 +2363,7 @@ store_pending_changes(sqlstore *store, u
dep_hash_clear(store->dependencies);
dep_hash_clear(store->depchanges);
} else {
- ulng stoldest = store_oldest(store);
+ ulng stoldest = store_oldest(store, tr);
id_hash_clear_older(store->dependencies, stoldest);
id_hash_clear_older(store->depchanges, stoldest);
}
@@ -2380,7 +2387,7 @@ store_manager(sqlstore *store)
store_lock(store);
if (ATOMIC_GET(&store->nr_active) == 0) {
ulng oldest = store_timestamp(store)+1;
- store_pending_changes(store, oldest);
+ store_pending_changes(store, oldest, NULL);
}
store_unlock(store);
MT_lock_set(&store->flush);
@@ -3664,7 +3671,7 @@ sql_trans_rollback(sql_trans *tr, bool c
if (!commit_lock)
MT_lock_set(&store->commit);
store_lock(store);
- ulng oldest = store_oldest(store);
+ ulng oldest = store_oldest(store, tr);
ulng commit_ts = store_get_timestamp(store); /* use most recent
timestamp such that we can cleanup savely */
for(node *n=nl->h; n; n = n->next) {
sql_change *c = n->data;
@@ -3673,7 +3680,7 @@ sql_trans_rollback(sql_trans *tr, bool c
c->commit(tr, c, 0 /* ie rollback */, oldest);
c->ts = commit_ts;
}
- store_pending_changes(store, oldest);
+ store_pending_changes(store, oldest, tr);
for(node *n=nl->h; n; n = n->next) {
sql_change *c = n->data;
@@ -3695,8 +3702,8 @@ sql_trans_rollback(sql_trans *tr, bool c
if (!commit_lock)
MT_lock_set(&store->commit);
store_lock(store);
- ulng oldest = store_oldest(store);
- store_pending_changes(store, oldest);
+ ulng oldest = store_oldest(store, tr);
+ store_pending_changes(store, oldest, tr);
store_unlock(store);
if (!commit_lock)
MT_lock_unset(&store->commit);
@@ -4006,14 +4013,14 @@ sql_trans_commit(sql_trans *tr)
}
else {
commit_ts = store_timestamp(store);
- oldest = store_oldest(store);
+ oldest = store_oldest(store, tr);
}
tr->logchanges = 0;
TRC_DEBUG(SQL_STORE, "Forwarding changes (" ULLFMT ", " ULLFMT
") -> " ULLFMT "\n", tr->tid, tr->ts, commit_ts);
/* apply committed changes */
if (ATOMIC_GET(&store->nr_active) == 1 && !tr->parent)
oldest = commit_ts;
- store_pending_changes(store, oldest);
+ store_pending_changes(store, oldest, tr);
for(node *n=tr->changes->h; n && ok == LOG_OK; n = n->next) {
sql_change *c = n->data;
@@ -4073,7 +4080,7 @@ sql_trans_commit(sql_trans *tr)
MT_lock_set(&store->commit);
store_lock(store);
ulng oldest = store_timestamp(store);
- store_pending_changes(store, oldest);
+ store_pending_changes(store, oldest, tr);
store_unlock(store);
MT_lock_unset(&store->commit);
}
_______________________________________________
checkin-list mailing list -- [email protected]
To unsubscribe send an email to [email protected]
