/*******************************************************************//**
Rolls back a transaction back to a named savepoint. Modifications after the
savepoint are undone but InnoDB does NOT release the corresponding locks
which are stored in memory. If a lock is 'implicit', that is, a new inserted
row holds a lock where the lock information is carried by the trx id stored in
the row, these locks are naturally released in the rollback. Savepoints which
were set after this savepoint are deleted.
@return if no savepoint of the name found then DB_NO_SAVEPOINT,
otherwise DB_SUCCESS */
UNIV_INTERN
ulint
trx_rollback_to_savepoint_for_mysql(
/*================================*/
trx_t* trx, /*!< in: transaction handle */
const char* savepoint_name, /*!< in: savepoint name */
ib_int64_t* mysql_binlog_cache_pos) /*!< out: the MySQL binlog cache
position corresponding to this
savepoint; MySQL needs this
information to remove the
binlog entries of the queries
executed after the savepoint */
{
trx_named_savept_t* savep;
ulint err;
savep = UT_LIST_GET_FIRST(trx->trx_savepoints);
while (savep != NULL) {
if (0 == ut_strcmp(savep->name, savepoint_name)) {
/* Found */
break;
}
savep = UT_LIST_GET_NEXT(trx_savepoints, savep);
}
if (savep == NULL) {
return(DB_NO_SAVEPOINT);
}
if (trx->conc_state == TRX_NOT_STARTED) {
ut_print_timestamp(stderr);
fputs(" InnoDB: Error: transaction has a savepoint ", stderr);
ut_print_name(stderr, trx, FALSE, savep->name);
fputs(" though it is not started\n", stderr);
return(DB_ERROR);
}
/* We can now free all savepoints strictly later than this one */
trx_roll_savepoints_free(trx, savep);
*mysql_binlog_cache_pos = savep->mysql_binlog_cache_pos;
trx->op_info = "rollback to a savepoint";
err = trx_general_rollback_for_mysql(trx, &savep->savept);
/* Store the current undo_no of the transaction so that we know where
to roll back if we have to roll back the next SQL statement: */
trx_mark_sql_stat_end(trx);
trx->op_info = "";
return(err);
}
void
trx_commit_off_kernel(
/*==================*/
trx_t* trx) /* in: transaction */
{
page_t* update_hdr_page;
dulint lsn;
trx_rseg_t* rseg;
trx_undo_t* undo;
ibool must_flush_log = FALSE;
mtr_t mtr;
ut_ad(mutex_own(&kernel_mutex));
trx->must_flush_log_later = FALSE;
rseg = trx->rseg;
if (trx->insert_undo != NULL || trx->update_undo != NULL) {
mutex_exit(&kernel_mutex);
mtr_start(&mtr);
must_flush_log = TRUE;
/* Change the undo log segment states from TRX_UNDO_ACTIVE
to some other state: these modifications to the file data
structure define the transaction as committed in the file
based world, at the serialization point of the log sequence
number lsn obtained below. */
mutex_enter(&(rseg->mutex));
if (trx->insert_undo != NULL) {
trx_undo_set_state_at_finish(
rseg, trx, trx->insert_undo, &mtr);
}
undo = trx->update_undo;
if (undo) {
mutex_enter(&kernel_mutex);
trx->no = trx_sys_get_new_trx_no();
mutex_exit(&kernel_mutex);
/* It is not necessary to obtain trx->undo_mutex here
because only a single OS thread is allowed to do the
transaction commit for this transaction. */
update_hdr_page = trx_undo_set_state_at_finish(
rseg, trx, undo, &mtr);
/* We have to do the cleanup for the update log while
holding the rseg mutex because update log headers
have to be put to the history list in the order of
the trx number. */
trx_undo_update_cleanup(trx, update_hdr_page, &mtr);
}
mutex_exit(&(rseg->mutex));
/* Update the latest MySQL binlog name and offset info
in trx sys header if MySQL binlogging is on or the database
server is a MySQL replication slave */
if (trx->mysql_log_file_name
&& trx->mysql_log_file_name[0] != '\0') {
trx_sys_update_mysql_binlog_offset(
trx->mysql_log_file_name,
trx->mysql_log_offset,
TRX_SYS_MYSQL_LOG_INFO, &mtr);
trx->mysql_log_file_name = NULL;
}
/* The following call commits the mini-transaction, making the
whole transaction committed in the file-based world, at this
log sequence number. The transaction becomes 'durable' when
we write the log to disk, but in the logical sense the commit
in the file-based data structures (undo logs etc.) happens
here.
NOTE that transaction numbers, which are assigned only to
transactions with an update undo log, do not necessarily come
in exactly the same order as commit lsn's, if the transactions
have different rollback segments. To get exactly the same
order we should hold the kernel mutex up to this point,
adding to to the contention of the kernel mutex. However, if
a transaction T2 is able to see modifications made by
a transaction T1, T2 will always get a bigger transaction
number and a bigger commit lsn than T1. */
/*--------------*/
mtr_commit(&mtr);
/*--------------*/
lsn = mtr.end_lsn;
mutex_enter(&kernel_mutex);
}
ut_ad(trx->conc_state == TRX_ACTIVE
|| trx->conc_state == TRX_PREPARED);
ut_ad(mutex_own(&kernel_mutex));
/* The following assignment makes the transaction committed in memory
and makes its changes to data visible to other transactions.
NOTE that there is a small discrepancy from the strict formal
visibility rules here: a human user of the database can see
modifications made by another transaction T even before the necessary
log segment has been flushed to the disk. If the database happens to
crash before the flush, the user has seen modifications from T which
will never be a committed transaction. However, any transaction T2
which sees the modifications of the committing transaction T, and
which also itself makes modifications to the database, will get an lsn
larger than the committing transaction T. In the case where the log
flush fails, and T never gets committed, also T2 will never get
committed. */
/*--------------------------------------*/
trx->conc_state = TRX_COMMITTED_IN_MEMORY;
/*--------------------------------------*/
lock_release_off_kernel(trx);
if (trx->global_read_view) {
read_view_close(trx->global_read_view);
mem_heap_empty(trx->global_read_view_heap);
trx->global_read_view = NULL;
}
trx->read_view = NULL;
if (must_flush_log) {
mutex_exit(&kernel_mutex);
if (trx->insert_undo != NULL) {
trx_undo_insert_cleanup(trx);
}
/* NOTE that we could possibly make a group commit more
efficient here: call os_thread_yield here to allow also other
trxs to come to commit! */
/*-------------------------------------*/
/* Depending on the my.cnf options, we may now write the log
buffer to the log files, making the transaction durable if
the OS does not crash. We may also flush the log files to
disk, making the transaction durable also at an OS crash or a
power outage.
The idea in InnoDB's group commit is that a group of
transactions gather behind a trx doing a physical disk write
to log files, and when that physical write has been completed,
one of those transactions does a write which commits the whole
group. Note that this group commit will only bring benefit if
there are > 2 users in the database. Then at least 2 users can
gather behind one doing the physical log write to disk.
If we are calling trx_commit() under MySQL's binlog mutex, we
will delay possible log write and flush to a separate function
trx_commit_complete_for_mysql(), which is only called when the
thread has released the binlog mutex. This is to make the
group commit algorithm to work. Otherwise, the MySQL binlog
mutex would serialize all commits and prevent a group of
transactions from gathering. */
if (trx->flush_log_later) {
/* Do nothing yet */
trx->must_flush_log_later = TRUE;
} else if (srv_flush_log_at_trx_commit == 0) {
/* Do nothing */
} else if (srv_flush_log_at_trx_commit == 1) {
if (srv_unix_file_flush_method == SRV_UNIX_NOSYNC) {
/* Write the log but do not flush it to disk */
log_write_up_to(lsn, LOG_WAIT_ONE_GROUP,
FALSE);
} else {
/* Write the log to the log files AND flush
them to disk */
log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, TRUE);
}
} else if (srv_flush_log_at_trx_commit == 2) {
/* Write the log but do not flush it to disk */
log_write_up_to(lsn, LOG_WAIT_ONE_GROUP, FALSE);
} else {
ut_error;
}
trx->commit_lsn = lsn;
/*-------------------------------------*/
mutex_enter(&kernel_mutex);
}
/* Free savepoints */
trx_roll_savepoints_free(trx, NULL);
trx->conc_state = TRX_NOT_STARTED;
trx->rseg = NULL;
trx->undo_no = ut_dulint_zero;
trx->last_sql_stat_start.least_undo_no = ut_dulint_zero;
trx->mysql_query_str = NULL;
ut_ad(UT_LIST_GET_LEN(trx->wait_thrs) == 0);
ut_ad(UT_LIST_GET_LEN(trx->trx_locks) == 0);
UT_LIST_REMOVE(trx_list, trx_sys->trx_list, trx);
}
