trx0trx.c

这是linux下运行的mysql软件包,可用于linux 下安装 php + mysql + apach 的网络配置

	
	node = thr->run_node;

	ut_ad(que_node_get_type(node) == QUE_NODE_COMMIT);

	if (thr->prev_node == que_node_get_parent(node)) {
		node->state = COMMIT_NODE_SEND;
	}

	if (node->state == COMMIT_NODE_SEND) {
		mutex_enter(&kernel_mutex);

		node->state = COMMIT_NODE_WAIT;

		next_thr = NULL;
		
		thr->state = QUE_THR_SIG_REPLY_WAIT;

		/* Send the commit signal to the transaction */
		
		success = trx_sig_send(thr_get_trx(thr), TRX_SIG_COMMIT,
					TRX_SIG_SELF, thr, NULL, &next_thr);
		
		mutex_exit(&kernel_mutex);

		if (!success) {
			/* Error in delivering the commit signal */
			que_thr_handle_error(thr, DB_ERROR, NULL, 0);
		}

		return(next_thr);
	}

	ut_ad(node->state == COMMIT_NODE_WAIT);
		
	node->state = COMMIT_NODE_SEND;
	
	thr->run_node = que_node_get_parent(node);

	return(thr);
}

/**************************************************************************
Does the transaction commit for MySQL. */

ulint
trx_commit_for_mysql(
/*=================*/
			/* out: 0 or error number */
	trx_t*	trx)	/* in: trx handle */
{
	/* Because we do not do the commit by sending an Innobase
	sig to the transaction, we must here make sure that trx has been
	started. */

	ut_a(trx);

	trx->op_info = "committing";
	
	trx_start_if_not_started(trx);

	mutex_enter(&kernel_mutex);

	trx_commit_off_kernel(trx);

	mutex_exit(&kernel_mutex);

	trx->op_info = "";
	
	return(0);
}

/**************************************************************************
If required, flushes the log to disk if we called trx_commit_for_mysql()
with trx->flush_log_later == TRUE. */

ulint
trx_commit_complete_for_mysql(
/*==========================*/
			/* out: 0 or error number */
	trx_t*	trx)	/* in: trx handle */
{
        dulint  lsn     = trx->commit_lsn;

        ut_a(trx);
	
	trx->op_info = "flushing log";

	if (!trx->must_flush_log_later) {
                /* Do nothing */
        } 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->must_flush_log_later = FALSE;

	trx->op_info = "";

        return(0);
}

/**************************************************************************
Marks the latest SQL statement ended. */

void
trx_mark_sql_stat_end(
/*==================*/
	trx_t*	trx)	/* in: trx handle */
{
	ut_a(trx);

	if (trx->conc_state == TRX_NOT_STARTED) {
		trx->undo_no = ut_dulint_zero;
	}

	trx->last_sql_stat_start.least_undo_no = trx->undo_no;
}

/**************************************************************************
Prints info about a transaction to the given file. The caller must own the
kernel mutex and must have called
innobase_mysql_prepare_print_arbitrary_thd(), unless he knows that MySQL
or InnoDB cannot meanwhile change the info printed here. */

void
trx_print(
/*======*/
	FILE*	f,		/* in: output stream */
	trx_t*	trx,		/* in: transaction */
	uint	max_query_len)	/* in: max query length to print, or 0 to
				   use the default max length */
{
	ibool	newline;

	fprintf(f, "TRANSACTION %lu %lu",
		(ulong) ut_dulint_get_high(trx->id),
		 (ulong) ut_dulint_get_low(trx->id));

        switch (trx->conc_state) {
		case TRX_NOT_STARTED:
			fputs(", not started", f);
			break;
		case TRX_ACTIVE:
			fprintf(f, ", ACTIVE %lu sec",
				(ulong)difftime(time(NULL), trx->start_time));
                        break;
		case TRX_PREPARED:
			fprintf(f, ", ACTIVE (PREPARED) %lu sec",
				(ulong)difftime(time(NULL), trx->start_time));
                        break;
		case TRX_COMMITTED_IN_MEMORY:
			fputs(", COMMITTED IN MEMORY", f);
			break;
		default:
			fprintf(f, " state %lu", (ulong) trx->conc_state);
        }

#ifdef UNIV_LINUX
	fprintf(f, ", process no %lu", trx->mysql_process_no);
#endif
	fprintf(f, ", OS thread id %lu",
		       (ulong) os_thread_pf(trx->mysql_thread_id));

	if (*trx->op_info) {
		putc(' ', f);
		fputs(trx->op_info, f);
	}
	
  	if (trx->type != TRX_USER) {
		fputs(" purge trx", f);
  	}

	if (trx->declared_to_be_inside_innodb) {
		fprintf(f, ", thread declared inside InnoDB %lu",
			       (ulong) trx->n_tickets_to_enter_innodb);
	}

	putc('\n', f);
  	
        if (trx->n_mysql_tables_in_use > 0 || trx->mysql_n_tables_locked > 0) {
		fprintf(f, "mysql tables in use %lu, locked %lu\n",
					(ulong) trx->n_mysql_tables_in_use,
					(ulong) trx->mysql_n_tables_locked);
	}

	newline = TRUE;

  	switch (trx->que_state) {
		case TRX_QUE_RUNNING:
			newline = FALSE; break;
		case TRX_QUE_LOCK_WAIT:
			fputs("LOCK WAIT ", f); break;
		case TRX_QUE_ROLLING_BACK:
			fputs("ROLLING BACK ", f); break;
		case TRX_QUE_COMMITTING:
			fputs("COMMITTING ", f); break;
		default:
			fprintf(f, "que state %lu ", (ulong) trx->que_state);
  	}

  	if (0 < UT_LIST_GET_LEN(trx->trx_locks) ||
	    mem_heap_get_size(trx->lock_heap) > 400) {
		newline = TRUE;

		fprintf(f, "%lu lock struct(s), heap size %lu",
			       (ulong) UT_LIST_GET_LEN(trx->trx_locks),
			       (ulong) mem_heap_get_size(trx->lock_heap));
	}

  	if (trx->has_search_latch) {
		newline = TRUE;
		fputs(", holds adaptive hash latch", f);
  	}

	if (ut_dulint_cmp(trx->undo_no, ut_dulint_zero) != 0) {
		newline = TRUE;
		fprintf(f, ", undo log entries %lu",
			(ulong) ut_dulint_get_low(trx->undo_no));
	}
	
	if (newline) {
		putc('\n', f);
	}

  	if (trx->mysql_thd != NULL) {
		innobase_mysql_print_thd(f, trx->mysql_thd, max_query_len);
  	}  
}

/********************************************************************
Prepares a transaction. */

void
trx_prepare_off_kernel(
/*===================*/
	trx_t*	trx)	/* in: transaction */
{
	page_t*		update_hdr_page;
	trx_rseg_t*	rseg;
	ibool		must_flush_log	= FALSE;
	dulint		lsn;
	mtr_t		mtr;
	
#ifdef UNIV_SYNC_DEBUG
	ut_ad(mutex_own(&kernel_mutex));
#endif /* UNIV_SYNC_DEBUG */

	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 TRX_UNDO_PREPARED: these modifications to the file data
		structure define the transaction as prepared in the
		file-based world, at the serialization point of lsn. */

		mutex_enter(&(rseg->mutex));
			
		if (trx->insert_undo != NULL) {

			/* It is not necessary to obtain trx->undo_mutex here
			because only a single OS thread is allowed to do the
			transaction prepare for this transaction. */

			trx_undo_set_state_at_prepare(trx, trx->insert_undo,
							  		&mtr);
		}

		if (trx->update_undo) {
			update_hdr_page = trx_undo_set_state_at_prepare(trx,
						trx->update_undo, &mtr);
		}

		mutex_exit(&(rseg->mutex));

		/*--------------*/
		mtr_commit(&mtr);	/* This mtr commit makes the
					transaction prepared in the file-based
					world */
		/*--------------*/
 		lsn = mtr.end_lsn;

		mutex_enter(&kernel_mutex);
	}

#ifdef UNIV_SYNC_DEBUG
	ut_ad(mutex_own(&kernel_mutex));
#endif /* UNIV_SYNC_DEBUG */

	/*--------------------------------------*/
	trx->conc_state = TRX_PREPARED;
	/*--------------------------------------*/

	if (must_flush_log) {
                /* Depending on the my.cnf options, we may now write the log
                buffer to the log files, making the prepared state of the
		transaction durable if the OS does not crash. We may also
		flush the log files to disk, making the prepared state of the
		transaction durable also at an OS crash or a power outage.

                The idea in InnoDB's group prepare 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 prepares the whole
                group. Note that this group prepare 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.

		TODO: find out if MySQL holds some mutex when calling this.
		That would spoil our group prepare algorithm. */

		mutex_exit(&kernel_mutex);

                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;
                }

		mutex_enter(&kernel_mutex);
	}
}

/**************************************************************************
Does the transaction prepare for MySQL. */

ulint
trx_prepare_for_mysql(
/*====-=============*/
			/* out: 0 or error number */
	trx_t*	trx)	/* in: trx handle */
{
	/* Because we do not do the prepare by sending an Innobase
	sig to the transaction, we must here make sure that trx has been
	started. */

	ut_a(trx);

	trx->op_info = "preparing";
	
	trx_start_if_not_started(trx);

	mutex_enter(&kernel_mutex);

	trx_prepare_off_kernel(trx);

	mutex_exit(&kernel_mutex);

	trx->op_info = "";
	
	return(0);
}

/**************************************************************************
This function is used to find number of prepared transactions and
their transaction objects for a recovery. */

int
trx_recover_for_mysql(
/*==================*/
				/* out: number of prepared transactions 
				stored in xid_list */
	XID*    xid_list, 	/* in/out: prepared transactions */
	ulint	len)		/* in: number of slots in xid_list */
{
	trx_t*	trx;
	int	count = 0;

	ut_ad(xid_list);
	ut_ad(len);

	/* We should set those transactions which are in the prepared state
	to the xid_list */

	mutex_enter(&kernel_mutex);

	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);

	while (trx) {
		if (trx->conc_state == TRX_PREPARED) {
			xid_list[count] = trx->xid;

			if (count == 0) {
				ut_print_timestamp(stderr);
				fprintf(stderr,
"  InnoDB: Starting recovery for XA transactions...\n");
			}

			ut_print_timestamp(stderr);
			fprintf(stderr,
"  InnoDB: Transaction %lu %lu in prepared state after recovery\n",
				(ulong) ut_dulint_get_high(trx->id),
				(ulong) ut_dulint_get_low(trx->id));

			ut_print_timestamp(stderr);
			fprintf(stderr,
"  InnoDB: Transaction contains changes to %lu rows\n",
			(ulong)ut_conv_dulint_to_longlong(trx->undo_no));

			count++;
		
			if ((uint)count == len ) {
				break;
			}
		}

		trx = UT_LIST_GET_NEXT(trx_list, trx);
	}

	mutex_exit(&kernel_mutex);

	if (count > 0){
		ut_print_timestamp(stderr);
		fprintf(stderr,
"  InnoDB: %d transactions in prepared state after recovery\n",
			count);
	}

	return (count);			
}

/***********************************************************************
This function is used to find one X/Open XA distributed transaction
which is in the prepared state */

trx_t*
trx_get_trx_by_xid(
/*===============*/
			/* out: trx or NULL */
	XID*	xid)	/* in: X/Open XA transaction identification */
{
	trx_t*	trx;

	if (xid == NULL) {

		return (NULL);
	}
 
	mutex_enter(&kernel_mutex);

	trx = UT_LIST_GET_FIRST(trx_sys->trx_list);

	while (trx) {
		/* Compare two X/Open XA transaction id's: their
		length should be the same and binary comparison
		of gtrid_lenght+bqual_length bytes should be
		the same */

		if (xid->gtrid_length == trx->xid.gtrid_length &&
		    xid->bqual_length == trx->xid.bqual_length &&
		    memcmp(xid->data, trx->xid.data, 
				xid->gtrid_length + 
				xid->bqual_length) == 0) {
			break;
		}

		trx = UT_LIST_GET_NEXT(trx_list, trx);
	}

	mutex_exit(&kernel_mutex);

	if (trx) {
		if (trx->conc_state != TRX_PREPARED) {

			return(NULL);
		}

		return(trx);
	} else {
		return(NULL);
	}
}