MDEV-15158 On commit, do not write to the TRX_SYS page

This is based on a prototype by Thirunarayanan Balathandayuthapani <thiru@mariadb.com>. Binlog and Galera write-set replication information was written into TRX_SYS page on each commit. Instead of writing to the TRX_SYS during normal operation, InnoDB can make use of rollback segment header pages, which are already being written to during a commit. The following list of fields in rollback segment header page are added: TRX_RSEG_BINLOG_OFFSET TRX_RSEG_BINLOG_NAME (NUL-terminated; empty name = not present) TRX_RSEG_WSREP_XID_FORMAT (0=not present; 1=present) TRX_RSEG_WSREP_XID_GTRID TRX_RSEG_WSREP_XID_BQUAL TRX_RSEG_WSREP_XID_DATA trx_sys_t: Introduce the fields recovered_binlog_filename, recovered_binlog_offset, recovered_wsrep_xid. To facilitate upgrade from older mysql or mariaDB versions, we will read the information in TRX_SYS page. It will be overridden by the information that we find in rollback segment header pages. Mariabackup --prepare will read the metadata from the rollback segment header pages via trx_rseg_array_init(). It will still not read any undo log pages or recover any transactions.

Author: dr-m

extra/mariabackup/wsrep.cc

@@ -43,7 +43,7 @@ permission notice:
 #include <my_global.h>
 #include <my_base.h>
 #include <handler.h>
-#include <trx0sys.h>
+#include <trx0rseg.h>
 
 #include "common.h"
 #ifdef WITH_WSREP
@@ -181,7 +181,7 @@ xb_write_galera_info(bool incremental_prepare)
 memset(&xid, 0, sizeof(xid));
 xid.formatID = -1;
 
-if (!trx_sys_read_wsrep_checkpoint(&xid)) {
+if (!trx_rseg_read_wsrep_checkpoint(xid)) {
 
 return;
 }

extra/mariabackup/xtrabackup.cc

@@ -4843,37 +4843,23 @@ xtrabackup_prepare_func(char** argv)
 }
 
 if (ok) {
-mtr_tmtr;
-mtr.start();
-const buf_block_t*	sys_header = trx_sysf_get(&mtr, false);
-
-if (mach_read_from_4(TRX_SYS_MYSQL_LOG_INFO
- + TRX_SYS_MYSQL_LOG_MAGIC_N_FLD
- + TRX_SYS + sys_header->frame)
- == TRX_SYS_MYSQL_LOG_MAGIC_N) {
-ulonglong pos = mach_read_from_8(
-TRX_SYS_MYSQL_LOG_INFO
-+ TRX_SYS_MYSQL_LOG_OFFSET
-+ TRX_SYS + sys_header->frame);
-const char* name = reinterpret_cast<const char*>(
-TRX_SYS_MYSQL_LOG_INFO + TRX_SYS_MYSQL_LOG_NAME
-+ TRX_SYS + sys_header->frame);
-msg("Last binlog file %s, position %llu\n", name, pos);
-
-/* output to xtrabackup_binlog_pos_innodb and
-(if backup_safe_binlog_info was available on
-the server) to xtrabackup_binlog_info. In the
-latter case xtrabackup_binlog_pos_innodb
-becomes redundant and is created only for
-compatibility. */
-ok = store_binlog_info(
-"xtrabackup_binlog_pos_innodb", name, pos)
-&& (!recover_binlog_info || store_binlog_info(
- XTRABACKUP_BINLOG_INFO,
- name, pos));
-}
-
-mtr.commit();
+msg("Last binlog file %s, position %lld\n",
+ trx_sys.recovered_binlog_filename,
+ longlong(trx_sys.recovered_binlog_offset));
+
+/* output to xtrabackup_binlog_pos_innodb and
+ (if backup_safe_binlog_info was available on
+ the server) to xtrabackup_binlog_info. In the
+ latter case xtrabackup_binlog_pos_innodb
+ becomes redundant and is created only for
+ compatibility. */
+ok = store_binlog_info("xtrabackup_binlog_pos_innodb",
+ trx_sys.recovered_binlog_filename,
+ trx_sys.recovered_binlog_offset)
+ && (!recover_binlog_info
+ || store_binlog_info(XTRABACKUP_BINLOG_INFO,
+ trx_sys.recovered_binlog_filename,
+ trx_sys.recovered_binlog_offset));
 }
 
 /* Check whether the log is applied enough or not. */

mysql-test/suite/mariabackup/binlog.result

@@ -0,0 +1,8 @@
+CREATE TABLE t(a varchar(60)) ENGINE INNODB;
+INSERT INTO t VALUES(1);
+SHOW VARIABLES like 'log_bin';
+Variable_name	Value
+log_bin	ON
+FOUND 1 /Last binlog file .*, position .*/ in current_test
+# expect FOUND
+DROP TABLE t;

mysql-test/suite/mariabackup/binlog.test

@@ -0,0 +1,25 @@
+--source include/have_innodb.inc
+--source include/have_log_bin.inc
+
+let $basedir=$MYSQLTEST_VARDIR/tmp/backup;
+
+CREATE TABLE t(a varchar(60)) ENGINE INNODB;
+INSERT INTO t VALUES(1);
+
+SHOW VARIABLES like 'log_bin';
+
+--disable_result_log
+exec $XTRABACKUP --defaults-file=$MYSQLTEST_VARDIR/my.cnf --backup --target-dir=$basedir;
+--enable_result_log
+
+exec $XTRABACKUP --prepare --binlog-info=1 --apply-log-only --target-dir=$basedir ;
+
+let SEARCH_FILE=$MYSQLTEST_VARDIR/log/current_test;
+--let SEARCH_PATTERN= Last binlog file .*, position .*
+--source include/search_pattern_in_file.inc
+--echo # expect FOUND
+
+DROP TABLE t;
+
+# Cleanup
+rmdir $basedir;

storage/innobase/handler/ha_innodb.cc

@@ -111,10 +111,9 @@ this program; if not, write to the Free Software Foundation, Inc.,
 #include "trx0purge.h"
 #endif /* UNIV_DEBUG */
 #include "trx0roll.h"
-#include "trx0sys.h"
+#include "trx0rseg.h"
 #include "trx0trx.h"
 #include "fil0pagecompress.h"
-#include "trx0xa.h"
 #include "ut0mem.h"
 #include "row0ext.h"
 
@@ -19679,12 +19678,8 @@ innobase_wsrep_set_checkpoint(
 DBUG_ASSERT(hton == innodb_hton_ptr);
 
 if (wsrep_is_wsrep_xid(xid)) {
-mtr_t mtr;
-mtr_start(&mtr);
-if (buf_block_t* sys_header = trx_sysf_get(&mtr)) {
-trx_sys_update_wsrep_checkpoint(xid, sys_header, &mtr);
-}
-mtr_commit(&mtr);
+
+trx_rseg_update_wsrep_checkpoint(xid);
 innobase_flush_logs(hton, false);
 return 0;
 } else {
@@ -19700,7 +19695,7 @@ innobase_wsrep_get_checkpoint(
 XID* xid)
 {
 DBUG_ASSERT(hton == innodb_hton_ptr);
- trx_sys_read_wsrep_checkpoint(xid);
+ trx_rseg_read_wsrep_checkpoint(*xid);
  return 0;
 }
 

storage/innobase/include/trx0rseg.h

@@ -27,10 +27,8 @@ Created 3/26/1996 Heikki Tuuri
 #ifndef trx0rseg_h
 #define trx0rseg_h
 
-#include "trx0types.h"
 #include "trx0sys.h"
 #include "fut0lst.h"
-#include <vector>
 
 /** Gets a rollback segment header.
 @param[in]	space	space where placed
@@ -226,6 +224,30 @@ struct trx_rseg_t {
 /** Maximum transaction ID (valid only if TRX_RSEG_FORMAT is 0) */
 #define TRX_RSEG_MAX_TRX_ID(TRX_RSEG_UNDO_SLOTS + TRX_RSEG_N_SLOTS	\
  * TRX_RSEG_SLOT_SIZE)
+
+/** 8 bytes offset within the binlog file */
+#define TRX_RSEG_BINLOG_OFFSETTRX_RSEG_MAX_TRX_ID + 8
+/** MySQL log file name, 512 bytes, including terminating NUL
+(valid only if TRX_RSEG_FORMAT is 0).
+If no binlog information is present, the first byte is NUL. */
+#define TRX_RSEG_BINLOG_NAMETRX_RSEG_MAX_TRX_ID + 16
+/** Maximum length of binlog file name, including terminating NUL, in bytes */
+#define TRX_RSEG_BINLOG_NAME_LEN512
+
+#ifdef WITH_WSREP
+/** The offset to WSREP XID headers */
+#defineTRX_RSEG_WSREP_XID_INFOTRX_RSEG_MAX_TRX_ID + 16 + 512
+
+/** WSREP XID format (1 if present and valid, 0 if not present) */
+#define TRX_RSEG_WSREP_XID_FORMATTRX_RSEG_WSREP_XID_INFO
+/** WSREP XID GTRID length */
+#define TRX_RSEG_WSREP_XID_GTRID_LENTRX_RSEG_WSREP_XID_INFO + 4
+/** WSREP XID bqual length */
+#define TRX_RSEG_WSREP_XID_BQUAL_LENTRX_RSEG_WSREP_XID_INFO + 8
+/** WSREP XID data (XIDDATASIZE bytes) */
+#define TRX_RSEG_WSREP_XID_DATATRX_RSEG_WSREP_XID_INFO + 12
+#endif /* WITH_WSREP*/
+
 /*-------------------------------------------------------------*/
 
 /** Read the page number of an undo log slot.
@@ -240,6 +262,48 @@ trx_rsegf_get_nth_undo(const trx_rsegf_t* rsegf, ulint n)
 + n * TRX_RSEG_SLOT_SIZE);
 }
 
+#ifdef WITH_WSREP
+/** Update the WSREP XID information in rollback segment header.
+@param[in,out]	rseg_header	rollback segment header
+@param[in]	xid	WSREP XID
+@param[in,out]	mtr	mini-transaction */
+void
+trx_rseg_update_wsrep_checkpoint(
+trx_rsegf_t*rseg_header,
+const XID*xid,
+mtr_t*mtr);
+
+/** Update WSREP checkpoint XID in first rollback segment header.
+@param[in]	xid	WSREP XID */
+void trx_rseg_update_wsrep_checkpoint(const XID* xid);
+
+/** Read the WSREP XID information in rollback segment header.
+@param[in]	rseg_header	Rollback segment header
+@param[out]	xid	Transaction XID
+@return	whether the WSREP XID was present */
+bool trx_rseg_read_wsrep_checkpoint(const trx_rsegf_t* rseg_header, XID& xid);
+
+/** Recover the latest WSREP checkpoint XID.
+@param[out]	xid	WSREP XID
+@return	whether the WSREP XID was found */
+bool trx_rseg_read_wsrep_checkpoint(XID& xid);
+#endif /* WITH_WSREP */
+
+/** Upgrade a rollback segment header page to MariaDB 10.3 format.
+@param[in,out]	rseg_header	rollback segment header page
+@param[in,out]	mtr	mini-transaction */
+void trx_rseg_format_upgrade(trx_rsegf_t* rseg_header, mtr_t* mtr);
+
+/** Update the offset information about the end of the binlog entry
+which corresponds to the transaction just being committed.
+In a replication slave, this updates the master binlog position
+up to which replication has proceeded.
+@param[in,out]	rseg_header	rollback segment header
+@param[in]	trx	committing transaction
+@param[in,out]	mtr	mini-transaction */
+void
+trx_rseg_update_binlog_offset(byte* rseg_header, const trx_t* trx, mtr_t* mtr);
+
 #include "trx0rseg.ic"
 
 #endif

storage/innobase/include/trx0sys.h

@@ -141,26 +141,6 @@ trx_sys_update_mysql_binlog_offset(
 system header. */
 void
 trx_sys_print_mysql_binlog_offset();
-#ifdef WITH_WSREP
-
-/** Update WSREP XID info in the TRX_SYS page.
-@param[in]	xid	Transaction XID
-@param[in,out]	sys_header	TRX_SYS page
-@param[in,out]	mtr	mini-transaction */
-UNIV_INTERN
-void
-trx_sys_update_wsrep_checkpoint(
-const XID*	xid,
-buf_block_t*	sys_header,
-mtr_t*	mtr);
-
-/** Read WSREP checkpoint XID from sys header.
-@param[out]	xid	WSREP XID
-@return	whether the checkpoint was present */
-UNIV_INTERN
-bool
-trx_sys_read_wsrep_checkpoint(XID* xid);
-#endif /* WITH_WSREP */
 
 /** Create the rollback segments.
 @return	whether the creation succeeded */
@@ -235,7 +215,8 @@ trx_sysf_rseg_get_page_no(const buf_block_t* sys_header, ulint rseg_id)
 + sys_header->frame);
 }
 
-/** Maximum length of MySQL binlog file name, in bytes. */
+/** Maximum length of MySQL binlog file name, in bytes.
+(Used before MariaDB 10.3.5.) */
 #define TRX_SYS_MYSQL_LOG_NAME_LEN512
 /** Contents of TRX_SYS_MYSQL_LOG_MAGIC_N_FLD */
 #define TRX_SYS_MYSQL_LOG_MAGIC_N873422344
@@ -312,7 +293,7 @@ FIXED WSREP XID info offsets for 4k page size 10.0.32-galera
 
 */
 #ifdef WITH_WSREP
-/** The offset to WSREP XID headers */
+/** The offset to WSREP XID headers (used before MariaDB 10.3.5) */
 #define TRX_SYS_WSREP_XID_INFO std::max(srv_page_size - 3500, 1596UL)
 #define TRX_SYS_WSREP_XID_MAGIC_N_FLD 0
 #define TRX_SYS_WSREP_XID_MAGIC_N 0x77737265
@@ -856,7 +837,6 @@ class trx_sys_t
 by any mutex, because it is read-only
 during multi-threaded operation */
 
-
  /**
  Lock-free hash of in memory read-write transactions.
  Works faster when it is on it's own cache line (tested).
@@ -865,6 +845,16 @@ class trx_sys_t
  MY_ALIGNED(CACHE_LINE_SIZE) rw_trx_hash_t rw_trx_hash;
 
 
+#ifdef WITH_WSREP
+ /** Latest recovered XID during startup */
+ XID recovered_wsrep_xid;
+#endif
+ /** Latest recovered binlog offset */
+ int64_t recovered_binlog_offset;
+ /** Latest recovred binlog file name */
+ char recovered_binlog_filename[TRX_SYS_MYSQL_LOG_NAME_LEN];
+
+
  /**
  Constructor.
 

storage/innobase/srv/srv0start.cc

@@ -2230,19 +2230,15 @@ innobase_start_or_create_for_mysql()
 if (err != DB_SUCCESS) {
 return(srv_init_abort(err));
 }
+/* fall through */
+case SRV_OPERATION_RESTORE:
 /* This must precede
 recv_apply_hashed_log_recs(true). */
 trx_lists_init_at_db_start();
 break;
 case SRV_OPERATION_RESTORE_DELTA:
 case SRV_OPERATION_BACKUP:
 ut_ad(!"wrong mariabackup mode");
-/* fall through */
-case SRV_OPERATION_RESTORE:
-/* mariabackup --prepare only deals with
-the redo log and the data files, not with
-transactions or the data dictionary. */
-break;
 }
 
 if (srv_force_recovery < SRV_FORCE_NO_LOG_REDO) {