diff --git a/src/box/memtx_engine.cc b/src/box/memtx_engine.cc
index dedf211c46ec983128843037850270abcd39bd55..e24f5c67379fbfb441ef551089da5b3162dbceb5 100644
--- a/src/box/memtx_engine.cc
+++ b/src/box/memtx_engine.cc
@@ -541,6 +541,7 @@ memtx_engine_prepare(struct engine *engine, struct txn *txn)
assert(stmt->space->engine == engine);
memtx_tx_history_prepare_stmt(stmt);
}
+ memtx_tx_prepare_finalize(txn);
}
if (txn->is_schema_changed)
memtx_tx_abort_all_for_ddl(txn);
diff --git a/src/box/memtx_tx.c b/src/box/memtx_tx.c
index c69dc0d2ef16b0ed57e39a24ca760847ef8d3314..ef79c3fcfe3cbf55ccd404d0507347c8efb32e88 100644
--- a/src/box/memtx_tx.c
+++ b/src/box/memtx_tx.c
@@ -38,6 +38,100 @@
#include "schema_def.h"
#include "small/mempool.h"
+/**
+ * Link that connects a memtx_story with older and newer stories of the same
+ * key in index.
+ */
+struct memtx_story_link {
+ /** Story that was happened after that story was ended. */
+ struct memtx_story *newer_story;
+ /** Story that was happened before that story was started. */
+ struct memtx_story *older_story;
+ /** List of interval items @sa gap_item. */
+ struct rlist nearby_gaps;
+ /**
+ * If the tuple of story is physically in index, here the pointer
+ * to that index is stored.
+ */
+ struct index *in_index;
+};
+
+/**
+ * A part of a history of a value in space.
+ * It's a story about a tuple, from the point it was added to space to the
+ * point when it was deleted from a space.
+ * All stories are linked into a list of stories of the same key of each index.
+ */
+struct memtx_story {
+ /** The story is about this tuple. The tuple is referenced. */
+ struct tuple *tuple;
+ /**
+ * Statement that introduced this story. Is set to NULL when the
+ * statement's transaction becomes committed. Can also be NULL if we
+ * don't know who introduced that story, the tuple was added by a
+ * transaction that was completed and destroyed some time ago.
+ */
+ struct txn_stmt *add_stmt;
+ /**
+ * Prepare sequence number of add_stmt's transaction. Is set when
+ * the transaction is prepared. Can be 0 if the transaction is
+ * in progress or we don't know who introduced that story.
+ */
+ int64_t add_psn;
+ /**
+ * Statement that ended this story. Is set to NULL when the statement's
+ * transaction becomes committed. Can also be NULL if the tuple has not
+ * been deleted yet.
+ */
+ struct txn_stmt *del_stmt;
+ /**
+ * Prepare sequence number of del_stmt's transaction. Is set when
+ * the transaction is prepared. Can be 0 if the transaction is
+ * in progress or if nobody has deleted the tuple.
+ */
+ int64_t del_psn;
+ /**
+ * List of trackers - transactions that has read this tuple.
+ */
+ struct rlist reader_list;
+ /**
+ * Link in tx_manager::all_stories
+ */
+ struct rlist in_all_stories;
+ /**
+ * Link in space::memtx_tx_stories.
+ */
+ struct rlist in_space_stories;
+ /**
+ * Number of indexes in this space - and the count of link[].
+ */
+ uint32_t index_count;
+ /**
+ * Status of story, describes the reason why story cannot be deleted.
+ * It is initialized in memtx_story constructor and is changed only in
+ * memtx_tx_story_gc.
+ */
+ enum memtx_tx_story_status status;
+ /**
+ * Flag is set when @a tuple is not placed in primary key and
+ * the story is the only reason why @a tuple cannot be deleted.
+ */
+ bool tuple_is_retained;
+ /*
+ * Transaction that added this story was rollbacked: this story is
+ * absolutely invisible — its only purpose is to retain the reader list.
+ * It is present at the end of some history chains and completely
+ * unlinked from others, which also implies it is not present in the
+ * corresponding indexes.
+ */
+ bool rollbacked;
+ /**
+ * Link with older and newer stories (and just tuples) for each
+ * index respectively.
+ */
+ struct memtx_story_link link[];
+};
+
static uint32_t
memtx_tx_story_key_hash(const struct tuple *a)
{
@@ -59,6 +153,37 @@ memtx_tx_story_key_hash(const struct tuple *a)
#define MH_SOURCE
#include "salad/mhash.h"
+/**
+ * Record that links two transactions, breaker and victim.
+ * See memtx_tx_cause_conflict for details.
+ */
+struct tx_conflict_tracker {
+ /** TX that aborts victim on commit. */
+ struct txn *breaker;
+ /** TX that will be aborted on breaker's commit. */
+ struct txn *victim;
+ /** Link in breaker->conflict_list. */
+ struct rlist in_conflict_list;
+ /** Link in victim->conflicted_by_list. */
+ struct rlist in_conflicted_by_list;
+};
+
+/**
+ * Record that links transaction and a story that the transaction have read.
+ */
+struct tx_read_tracker {
+ /** The TX that read story. */
+ struct txn *reader;
+ /** The story that was read by reader. */
+ struct memtx_story *story;
+ /** Link in story->reader_list. */
+ struct rlist in_reader_list;
+ /** Link in reader->read_set. */
+ struct rlist in_read_set;
+ /** Bit field of indexes in which the data was tread by reader. */
+ uint64_t index_mask;
+};
+
/**
* An element that stores the fact that some transaction have read
* a full key and found nothing.
@@ -540,7 +665,21 @@ memtx_tx_abort_all_for_ddl(struct txn *ddl_owner)
}
}
-int
+/**
+ * Notify TX manager that if transaction @a breaker is committed then the
+ * transaction @a victim must be aborted due to conflict. It is achieved
+ * by adding corresponding entry (of tx_conflict_tracker type) to @a breaker
+ * conflict list. In case there's already such entry, then move it to the head
+ * of the list in order to optimize next invocations of this function.
+ * For example: there's two rw transaction in progress, one have read
+ * some value while the second is about to overwrite it. If the second
+ * is committed first, the first must be aborted.
+ *
+ * NB: can trigger story garbage collection.
+ *
+ * @return 0 on success, -1 on memory error.
+ */
+static int
memtx_tx_cause_conflict(struct txn *breaker, struct txn *victim)
{
assert(breaker != victim);
@@ -618,7 +757,22 @@ memtx_tx_adjust_position_in_read_view_list(struct txn *txn)
rlist_add_tail(&prev_txn->in_read_view_txs, &txn->in_read_view_txs);
}
-void
+/**
+ * Handle conflict when @a victim has read and @a breaker has written the same
+ * key, and @a breaker is prepared. The functions must be called in two cases:
+ * 1. @a breaker becomes prepared for every victim with non-empty intersection
+ * of victim read set / breaker write set.
+ * 2. @a victim has to read confirmed value and skips the value that prepared
+ * @a breaker wrote.
+ * If @a victim is read-only or hasn't made any changes, it should be sent
+ * to read view, in which is will not see @a breaker's changes. If @a victim
+ * is already in a read view - a read view that does not see every breaker
+ * changes is chosen.
+ * Otherwise @a victim must be marked as conflicted and aborted on occasion.
+ *
+ * NB: can trigger story garbage collection.
+ */
+static void
memtx_tx_handle_conflict(struct txn *breaker, struct txn *victim)
{
assert(breaker != victim);
@@ -1703,6 +1857,16 @@ check_hole(struct space *space, uint32_t index,
return 0;
}
+/**
+ * Record in TX manager that a transaction @txn have read a @tuple in @space.
+ *
+ * NB: can trigger story garbage collection.
+ *
+ * @return 0 on success, -1 on memory error.
+ */
+static int
+memtx_tx_track_read(struct txn *txn, struct space *space, struct tuple *tuple);
+
/**
* Check that replaced tuples in space's indexes does not violate common
* replace rules. See memtx_space_replace_all_keys comment.
@@ -2605,6 +2769,29 @@ memtx_tx_history_prepare_stmt(struct txn_stmt *stmt)
memtx_tx_story_gc();
}
+void
+memtx_tx_prepare_finalize(struct txn *txn)
+{
+ struct tx_conflict_tracker *entry, *next;
+ /* Handle conflicts. */
+ rlist_foreach_entry_safe(entry, &txn->conflict_list,
+ in_conflict_list, next) {
+ assert(entry->breaker == txn);
+ memtx_tx_handle_conflict(txn, entry->victim);
+ rlist_del(&entry->in_conflict_list);
+ rlist_del(&entry->in_conflicted_by_list);
+ }
+ /* Just free conflict list - we don't need it anymore. */
+ rlist_foreach_entry_safe(entry, &txn->conflicted_by_list,
+ in_conflicted_by_list, next) {
+ assert(entry->victim == txn);
+ rlist_del(&entry->in_conflict_list);
+ rlist_del(&entry->in_conflicted_by_list);
+ }
+ assert(rlist_empty(&txn->conflict_list));
+ assert(rlist_empty(&txn->conflicted_by_list));
+}
+
void
memtx_tx_history_commit_stmt(struct txn_stmt *stmt, size_t *bsize)
{
@@ -2713,6 +2900,10 @@ memtx_tx_tuple_clarify_slow(struct txn *txn, struct space *space,
struct tuple *tuple, struct index *index,
uint32_t mk_index)
{
+ if (!tuple_has_flag(tuple, TUPLE_IS_DIRTY)) {
+ memtx_tx_track_read(txn, space, tuple);
+ return tuple;
+ }
bool is_prepared_ok = detect_whether_prepared_ok(txn);
struct tuple *res =
memtx_tx_tuple_clarify_impl(txn, space, tuple, index, mk_index,
@@ -2891,9 +3082,11 @@ memtx_tx_track_read_story(struct txn *txn, struct space *space,
/**
* Record in TX manager that a transaction @txn have read a @tuple in @space.
*
+ * NB: can trigger story garbage collection.
+ *
* @return 0 on success, -1 on memory error.
*/
-int
+static int
memtx_tx_track_read(struct txn *txn, struct space *space, struct tuple *tuple)
{
if (tuple == NULL)
@@ -3211,6 +3404,32 @@ memtx_tx_clean_txn(struct txn *txn)
in_full_scan_list);
memtx_tx_full_scan_item_delete(item);
}
+
+ struct tx_read_tracker *tracker, *tmp;
+ rlist_foreach_entry_safe(tracker, &txn->read_set,
+ in_read_set, tmp) {
+ rlist_del(&tracker->in_reader_list);
+ rlist_del(&tracker->in_read_set);
+ }
+ assert(rlist_empty(&txn->read_set));
+
+ struct tx_conflict_tracker *entry, *next;
+ rlist_foreach_entry_safe(entry, &txn->conflict_list,
+ in_conflict_list, next) {
+ rlist_del(&entry->in_conflict_list);
+ rlist_del(&entry->in_conflicted_by_list);
+ }
+ rlist_foreach_entry_safe(entry, &txn->conflicted_by_list,
+ in_conflicted_by_list, next) {
+ rlist_del(&entry->in_conflict_list);
+ rlist_del(&entry->in_conflicted_by_list);
+ }
+ assert(rlist_empty(&txn->conflict_list));
+ assert(rlist_empty(&txn->conflicted_by_list));
+
+ rlist_del(&txn->in_read_view_txs);
+ rlist_del(&txn->in_all_txs);
+
memtx_tx_story_gc();
}
diff --git a/src/box/memtx_tx.h b/src/box/memtx_tx.h
index ae71fe5bf7f3f8673a39b9d4e0ee9462021f9090..6a1067f0acb4bacef6615d31508d7e7ad6d10ea3 100644
--- a/src/box/memtx_tx.h
+++ b/src/box/memtx_tx.h
@@ -97,131 +97,6 @@ enum memtx_tx_story_status {
extern const char *memtx_tx_story_status_strs[];
-/**
- * Record that links two transactions, breaker and victim.
- * See memtx_tx_cause_conflict for details.
- */
-struct tx_conflict_tracker {
- /** TX that aborts victim on commit. */
- struct txn *breaker;
- /** TX that will be aborted on breaker's commit. */
- struct txn *victim;
- /** Link in breaker->conflict_list. */
- struct rlist in_conflict_list;
- /** Link in victim->conflicted_by_list. */
- struct rlist in_conflicted_by_list;
-};
-
-/**
- * Record that links transaction and a story that the transaction have read.
- */
-struct tx_read_tracker {
- /** The TX that read story. */
- struct txn *reader;
- /** The story that was read by reader. */
- struct memtx_story *story;
- /** Link in story->reader_list. */
- struct rlist in_reader_list;
- /** Link in reader->read_set. */
- struct rlist in_read_set;
- /** Bit field of indexes in which the data was tread by reader. */
- uint64_t index_mask;
-};
-
-/**
- * Link that connects a memtx_story with older and newer stories of the same
- * key in index.
- */
-struct memtx_story_link {
- /** Story that was happened after that story was ended. */
- struct memtx_story *newer_story;
- /** Story that was happened before that story was started. */
- struct memtx_story *older_story;
- /** List of interval items @sa gap_item. */
- struct rlist nearby_gaps;
- /**
- * If the tuple of story is physically in index, here the pointer
- * to that index is stored.
- */
- struct index *in_index;
-};
-
-/**
- * A part of a history of a value in space.
- * It's a story about a tuple, from the point it was added to space to the
- * point when it was deleted from a space.
- * All stories are linked into a list of stories of the same key of each index.
- */
-struct memtx_story {
- /** The story is about this tuple. The tuple is referenced. */
- struct tuple *tuple;
- /**
- * Statement that introduced this story. Is set to NULL when the
- * statement's transaction becomes committed. Can also be NULL if we
- * don't know who introduced that story, the tuple was added by a
- * transaction that was completed and destroyed some time ago.
- */
- struct txn_stmt *add_stmt;
- /**
- * Prepare sequence number of add_stmt's transaction. Is set when
- * the transaction is prepared. Can be 0 if the transaction is
- * in progress or we don't know who introduced that story.
- */
- int64_t add_psn;
- /**
- * Statement that ended this story. Is set to NULL when the statement's
- * transaction becomes committed. Can also be NULL if the tuple has not
- * been deleted yet.
- */
- struct txn_stmt *del_stmt;
- /**
- * Prepare sequence number of del_stmt's transaction. Is set when
- * the transaction is prepared. Can be 0 if the transaction is
- * in progress or if nobody has deleted the tuple.
- */
- int64_t del_psn;
- /**
- * List of trackers - transactions that has read this tuple.
- */
- struct rlist reader_list;
- /**
- * Link in tx_manager::all_stories
- */
- struct rlist in_all_stories;
- /**
- * Link in space::memtx_tx_stories.
- */
- struct rlist in_space_stories;
- /**
- * Number of indexes in this space - and the count of link[].
- */
- uint32_t index_count;
- /**
- * Status of story, describes the reason why story cannot be deleted.
- * It is initialized in memtx_story constructor and is changed only in
- * memtx_tx_story_gc.
- */
- enum memtx_tx_story_status status;
- /**
- * Flag is set when @a tuple is not placed in primary key and
- * the story is the only reason why @a tuple cannot be deleted.
- */
- bool tuple_is_retained;
- /*
- * Transaction that added this story was rollbacked: this story is
- * absolutely invisible — its only purpose is to retain the reader list.
- * It is present at the end of some history chains and completely
- * unlinked from others, which also implies it is not present in the
- * corresponding indexes.
- */
- bool rollbacked;
- /**
- * Link with older and newer stories (and just tuples) for each
- * index respectively.
- */
- struct memtx_story_link link[];
-};
-
/**
* Snapshot cleaner is a short part of history that is supposed to clarify
* tuples in a index snapshot. It's also supposed to be used in another
@@ -299,41 +174,6 @@ memtx_tx_acquire_ddl(struct txn *tx);
void
memtx_tx_abort_all_for_ddl(struct txn *ddl_owner);
-/**
- * Notify TX manager that if transaction @a breaker is committed then the
- * transaction @a victim must be aborted due to conflict. It is achieved
- * by adding corresponding entry (of tx_conflict_tracker type) to @a breaker
- * conflict list. In case there's already such entry, then move it to the head
- * of the list in order to optimize next invocations of this function.
- * For example: there's two rw transaction in progress, one have read
- * some value while the second is about to overwrite it. If the second
- * is committed first, the first must be aborted.
- *
- * NB: can trigger story garbage collection.
- *
- * @return 0 on success, -1 on memory error.
- */
-int
-memtx_tx_cause_conflict(struct txn *breaker, struct txn *victim);
-
-/**
- * Handle conflict when @a victim has read and @a breaker has written the same
- * key, and @a breaker is prepared. The functions must be called in two cases:
- * 1. @a breaker becomes prepared for every victim with non-empty intersection
- * of victim read set / breaker write set.
- * 2. @a victim has to read confirmed value and skips the value that prepared
- * @a breaker wrote.
- * If @a victim is read-only or hasn't made any changes, it should be sent
- * to read view, in which is will not see @a breaker's changes. If @a victim
- * is already in a read view - a read view that does not see every breaker
- * changes is chosen.
- * Otherwise @a victim must be marked as conflicted and aborted on occasion.
- *
- * NB: can trigger story garbage collection.
- */
-void
-memtx_tx_handle_conflict(struct txn *breaker, struct txn *victim);
-
/**
* @brief Add a statement to transaction manager's history.
* Until unlinking or releasing the space could internally contain
@@ -390,6 +230,16 @@ memtx_tx_history_rollback_stmt(struct txn_stmt *stmt);
void
memtx_tx_history_prepare_stmt(struct txn_stmt *stmt);
+/**
+ * Finish preparing of a transaction.
+ * Must be called for entire transaction after `memtx_tx_history_rollback_stmt`
+ * was called for each transaction statement.
+ *
+ * NB: can trigger story garbage collection.
+ */
+void
+memtx_tx_prepare_finalize(struct txn *txn);
+
/**
* @brief Commit statement in history.
* Make the statement's changes permanent. It becomes visible to all.
@@ -408,17 +258,6 @@ memtx_tx_tuple_clarify_slow(struct txn *txn, struct space *space,
struct tuple *tuples, struct index *index,
uint32_t mk_index);
-/**
- * Record in TX manager that a transaction @txn have read a @tuple in @space.
- *
- * NB: can trigger story garbage collection.
- *
- * @return 0 on success, -1 on memory error.
- */
-int
-memtx_tx_track_read(struct txn *txn, struct space *space, struct tuple *tuple);
-
-
/** Helper of memtx_tx_track_point */
int
memtx_tx_track_point_slow(struct txn *txn, struct index *index,
@@ -538,10 +377,6 @@ memtx_tx_tuple_clarify(struct txn *txn, struct space *space,
{
if (!memtx_tx_manager_use_mvcc_engine)
return tuple;
- if (!tuple_has_flag(tuple, TUPLE_IS_DIRTY)) {
- memtx_tx_track_read(txn, space, tuple);
- return tuple;
- }
return memtx_tx_tuple_clarify_slow(txn, space, tuple, index, mk_index);
}
diff --git a/src/box/txn.c b/src/box/txn.c
index 823c7f6628ad267b3046395d4df3424bb0c643d5..2ba83505891729612894b80217cba30b30caefcd 100644
--- a/src/box/txn.c
+++ b/src/box/txn.c
@@ -451,31 +451,6 @@ txn_free(struct txn *txn)
if (txn->rollback_timer != NULL)
ev_timer_stop(loop(), txn->rollback_timer);
memtx_tx_clean_txn(txn);
- struct tx_read_tracker *tracker, *tmp;
- rlist_foreach_entry_safe(tracker, &txn->read_set,
- in_read_set, tmp) {
- rlist_del(&tracker->in_reader_list);
- rlist_del(&tracker->in_read_set);
- }
- assert(rlist_empty(&txn->read_set));
-
- struct tx_conflict_tracker *entry, *next;
- rlist_foreach_entry_safe(entry, &txn->conflict_list,
- in_conflict_list, next) {
- rlist_del(&entry->in_conflict_list);
- rlist_del(&entry->in_conflicted_by_list);
- }
- rlist_foreach_entry_safe(entry, &txn->conflicted_by_list,
- in_conflicted_by_list, next) {
- rlist_del(&entry->in_conflict_list);
- rlist_del(&entry->in_conflicted_by_list);
- }
- assert(rlist_empty(&txn->conflict_list));
- assert(rlist_empty(&txn->conflicted_by_list));
-
- rlist_del(&txn->in_read_view_txs);
- rlist_del(&txn->in_all_txs);
-
struct txn_stmt *stmt;
stailq_foreach_entry(stmt, &txn->stmts, next)
txn_stmt_destroy(stmt);
@@ -1017,25 +992,6 @@ txn_prepare(struct txn *txn)
return -1;
}
- struct tx_conflict_tracker *entry, *next;
- /* Handle conflicts. */
- rlist_foreach_entry_safe(entry, &txn->conflict_list,
- in_conflict_list, next) {
- assert(entry->breaker == txn);
- memtx_tx_handle_conflict(txn, entry->victim);
- rlist_del(&entry->in_conflict_list);
- rlist_del(&entry->in_conflicted_by_list);
- }
- /* Just free conflict list - we don't need it anymore. */
- rlist_foreach_entry_safe(entry, &txn->conflicted_by_list,
- in_conflicted_by_list, next) {
- assert(entry->victim == txn);
- rlist_del(&entry->in_conflict_list);
- rlist_del(&entry->in_conflicted_by_list);
- }
- assert(rlist_empty(&txn->conflict_list));
- assert(rlist_empty(&txn->conflicted_by_list));
-
trigger_clear(&txn->fiber_on_stop);
trigger_clear(&txn->fiber_on_yield);