bzr merge -r3997..4010 codership-mysql/5.5

sql/event_data_objects.cc

@@ -1468,12 +1468,16 @@ Event_job_data::execute(THD *thd, bool drop)
       thd->security_ctx->master_access |= SUPER_ACL;
 
 #ifdef WITH_WSREP
+      if (WSREP(thd)) {
           // sql_print_information("sizeof(LEX) = %d", sizeof(struct LEX));
           // sizeof(LEX) = 4512, so it's relatively safe to allocate it on stack.
           LEX lex;
           lex.sql_command = SQLCOM_DROP_EVENT;
+          LEX* saved = thd->lex;
           thd->lex = &lex;
           WSREP_TO_ISOLATION_BEGIN(WSREP_MYSQL_DB, NULL, NULL);
+          thd->lex = saved;
+      }
 #endif
      
       ret= Events::drop_event(thd, dbname, name, FALSE);

sql/sql_parse.cc

@@ -993,6 +993,9 @@ static void wsrep_copy_query(THD *thd)
 {
   thd->wsrep_retry_command   = thd->command;
   thd->wsrep_retry_query_len = thd->query_length();
+  if (thd->wsrep_retry_query) {
+      my_free(thd->wsrep_retry_query);
+  }
   thd->wsrep_retry_query     = (char *)my_malloc(
                                  thd->wsrep_retry_query_len + 1, MYF(0));
   strncpy(thd->wsrep_retry_query, thd->query(), thd->wsrep_retry_query_len);

sql/sys_vars.cc

@@ -3918,6 +3918,18 @@ static Sys_var_mybool Sys_wsrep_load_data_splitting(
        GLOBAL_VAR(wsrep_load_data_splitting), 
        CMD_LINE(OPT_ARG), DEFAULT(TRUE));
 
+static Sys_var_mybool Sys_wsrep_slave_FK_checks(
+       "wsrep_slave_FK_checks", "Should slave thread do "
+       "foreign key constraint checks",
+       GLOBAL_VAR(wsrep_slave_FK_checks), 
+       CMD_LINE(OPT_ARG), DEFAULT(TRUE));
+
+static Sys_var_mybool Sys_wsrep_slave_UK_checks(
+       "wsrep_slave_UK_checks", "Should slave thread do "
+       "secondary index uniqueness chesks",
+       GLOBAL_VAR(wsrep_slave_UK_checks), 
+       CMD_LINE(OPT_ARG), DEFAULT(FALSE));
+
 static Sys_var_mybool Sys_wsrep_restart_slave(
        "wsrep_restart_slave", "Should MySQL slave be restarted automatically, when node joins back to cluster",
        GLOBAL_VAR(wsrep_restart_slave), CMD_LINE(OPT_ARG), DEFAULT(FALSE));

sql/wsrep_applier.cc

@@ -207,6 +207,17 @@ wsrep_cb_status_t wsrep_apply_cb(void* const             ctx,
   thd_proc_info(thd, "applying write set");
 #endif /* WSREP_PROC_INFO */
 
+  /* tune FK and UK checking policy */
+  if (wsrep_slave_UK_checks == FALSE) 
+    thd->variables.option_bits|= OPTION_RELAXED_UNIQUE_CHECKS;
+  else
+    thd->variables.option_bits&= ~OPTION_RELAXED_UNIQUE_CHECKS;
+
+  if (wsrep_slave_FK_checks == FALSE) 
+    thd->variables.option_bits|= OPTION_NO_FOREIGN_KEY_CHECKS;
+  else
+    thd->variables.option_bits&= ~OPTION_NO_FOREIGN_KEY_CHECKS;
+
   if (flags & WSREP_FLAG_ISOLATION)
   {
     thd->wsrep_apply_toi= true;

sql/wsrep_mysqld.cc

@@ -51,7 +51,7 @@ ulong   wsrep_max_ws_size              = 1073741824UL;//max ws (RBR buffer) size
 ulong   wsrep_max_ws_rows              = 65536; // max number of rows in ws
 int     wsrep_to_isolation             = 0; // # of active TO isolation threads
 my_bool wsrep_certify_nonPK            = 1; // certify, even when no primary key
-long    wsrep_max_protocol_version     = 2; // maximum protocol version to use
+long    wsrep_max_protocol_version     = 3; // maximum protocol version to use
 ulong   wsrep_forced_binlog_format     = BINLOG_FORMAT_UNSPEC;
 my_bool wsrep_recovery                 = 0; // recovery
 my_bool wsrep_replicate_myisam         = 0; // enable myisam replication
@@ -64,6 +64,8 @@ my_bool wsrep_restart_slave            = 0; // should mysql slave thread be
                                             // restarted, if node joins back
 my_bool wsrep_restart_slave_activated  = 0; // node has dropped, and slave
                                             // restart will be needed
+my_bool wsrep_slave_UK_checks          = 0; // slave thread does UK checks
+my_bool wsrep_slave_FK_checks          = 0; // slave thread does FK checks
 /*
  * End configuration options
  */
@@ -109,7 +111,7 @@ const char* wsrep_provider_vendor    = provider_vendor;
 wsrep_uuid_t     local_uuid   = WSREP_UUID_UNDEFINED;
 wsrep_seqno_t    local_seqno  = WSREP_SEQNO_UNDEFINED;
 wsp::node_status local_status;
-long             wsrep_protocol_version = 2;
+long             wsrep_protocol_version = 3;
 
 // Boolean denoting if server is in initial startup phase. This is needed
 // to make sure that main thread waiting in wsrep_sst_wait() is signaled
@@ -130,7 +132,7 @@ static void wsrep_log_cb(wsrep_log_level_t level, const char *msg) {
     sql_print_error("WSREP: %s", msg);
     break;
   case WSREP_LOG_DEBUG:
-    sql_print_information ("[Debug] WSREP: %s", msg);
+    if (wsrep_debug) sql_print_information ("[Debug] WSREP: %s", msg);
   default:
     break;
   }
@@ -252,6 +254,7 @@ wsrep_view_handler_cb (void*                    app_ctx,
   case 0:
   case 1:
   case 2:
+  case 3:
       // version change
       if (view->proto_ver != wsrep_protocol_version)
       {
@@ -356,7 +359,7 @@ wsrep_view_handler_cb (void*                    app_ctx,
   }
 
 out:
-  wsrep_startup= FALSE;
+  if (view->status == WSREP_VIEW_PRIMARY) wsrep_startup= FALSE;
   local_status.set(new_status, view);
 
   return WSREP_CB_SUCCESS;
@@ -948,6 +951,7 @@ static bool wsrep_prepare_key_for_isolation(const char* db,
         break;
     case 1:
     case 2:
+    case 3:
     {
         *key_len= 0;
         if (db)
@@ -1079,6 +1083,7 @@ bool wsrep_prepare_key_for_innodb(const uchar* cache_key,
     }
     case 1:
     case 2:
+    case 3:
     {
         key[0].ptr = cache_key;
         key[0].len = strlen( (char*)cache_key );
@@ -1279,6 +1284,14 @@ static void wsrep_TOI_end(THD *thd) {
 
   WSREP_DEBUG("TO END: %lld, %d : %s", (long long)wsrep_thd_trx_seqno(thd),
               thd->wsrep_exec_mode, (thd->query()) ? thd->query() : "void");
+  
+  XID xid;
+  wsrep_xid_init(&xid, &thd->wsrep_trx_meta.gtid.uuid,
+                 thd->wsrep_trx_meta.gtid.seqno);
+  wsrep_set_SE_checkpoint(&xid);
+  WSREP_DEBUG("TO END: %lld, update seqno",
+              (long long)wsrep_thd_trx_seqno(thd));
+  
   if (WSREP_OK == (ret = wsrep->to_execute_end(wsrep, thd->thread_id))) {
     WSREP_DEBUG("TO END: %lld", (long long)wsrep_thd_trx_seqno(thd));
   }

sql/wsrep_mysqld.h

@@ -97,6 +97,8 @@ extern ulong       wsrep_mysql_replication_bundle;
 extern my_bool     wsrep_load_data_splitting;
 extern my_bool     wsrep_restart_slave;
 extern my_bool     wsrep_restart_slave_activated;
+extern my_bool     wsrep_slave_FK_checks;
+extern my_bool     wsrep_slave_UK_checks;
 
 enum enum_wsrep_OSU_method { WSREP_OSU_TOI, WSREP_OSU_RSU };
 

sql/wsrep_thd.cc

@@ -414,6 +414,17 @@ void wsrep_create_rollbacker()
   }
 }
 
+
+extern "C"
+void wsrep_thd_set_PA_safe(void *thd_ptr, my_bool safe)
+{ 
+  if (thd_ptr) 
+  {
+    THD* thd = (THD*)thd_ptr;
+    thd->wsrep_PA_safe = safe;
+  }
+}
+
 extern "C"
 my_bool wsrep_thd_is_BF(void *thd_ptr, my_bool sync)
 { 

sql/wsrep_thd.h

@@ -25,6 +25,7 @@ void wsrep_create_appliers(long threads);
 void wsrep_create_rollbacker();
 
 extern "C" my_bool  wsrep_thd_is_BF(void *thd_ptr, my_bool sync);
+extern "C" void  wsrep_thd_set_PA_safe(void *thd_ptr, my_bool safe);
 extern "C" my_bool  wsrep_thd_is_BF_or_commit(void *thd_ptr, my_bool sync);
 extern "C" my_bool  wsrep_thd_is_local(void *thd_ptr, my_bool sync);
 extern "C" int  wsrep_abort_thd(void *bf_thd_ptr, void *victim_thd_ptr,

storage/innobase/handler/ha_innodb.cc

@@ -4572,18 +4572,21 @@ innobase_mysql_cmp(
 }
 #ifdef WITH_WSREP
 extern "C" UNIV_INTERN
-void
+int
 wsrep_innobase_mysql_sort(
 /*===============*/
 					/* out: str contains sort string */
 	int		mysql_type,	/* in: MySQL type */
 	uint		charset_number,	/* in: number of the charset */
 	unsigned char*	str,		/* in: data field */
-	unsigned int	str_length)	/* in: data field length,
+	unsigned int	str_length,	/* in: data field length,
 					not UNIV_SQL_NULL */
+	unsigned int	buf_length)	/* in: total str buffer length */
+
 {
 	CHARSET_INFO*		charset;
 	enum_field_types	mysql_tp;
+	int ret_length =	str_length;
 
 	DBUG_ASSERT(str_length != UNIV_SQL_NULL);
 
@@ -4627,9 +4630,29 @@ wsrep_innobase_mysql_sort(
 		ut_a(str_length <= tmp_length);
 		memcpy(tmp_str, str, str_length);
 
-		tmp_length = charset->coll->strnxfrm(charset, str, str_length,
+		if (wsrep_protocol_version < 3) {
+			tmp_length = charset->coll->strnxfrm(
+				charset, str, str_length,
 				tmp_str, str_length);
 			DBUG_ASSERT(tmp_length <= str_length);
+		} else {
+			/* strnxfrm will expand the destination string,
+			   protocols < 3 truncated the sorted sring
+			   protocols > 3 gets full sorted sring
+			*/
+		  	/* 5.5 strnxfrm pads the tail with spaces and
+			   always returns the full destination buffer lenght
+			   we cannot know how many characters were converted
+			   using 2 * str length here as best guess
+			*/
+			uint dst_length = (str_length * 2 < tmp_length) ? 
+				(str_length * 2) : tmp_length; 
+			tmp_length = charset->coll->strnxfrm(
+				charset, str, dst_length,
+				tmp_str, str_length);
+			DBUG_ASSERT(tmp_length <= buf_length);
+			ret_length = tmp_length;
+		}
  
 		break;
 	}
@@ -4657,7 +4680,7 @@ wsrep_innobase_mysql_sort(
 		break;
 	}
 
-	return;
+	return ret_length;
 }
 #endif // WITH_WSREP
 /**************************************************************//**
@@ -4871,8 +4894,9 @@ wsrep_store_key_val_for_row(
 			}
 
 			memcpy(sorted, data, true_len);
-			wsrep_innobase_mysql_sort(
-			       mysql_type, cs->number, sorted, true_len);
+			true_len = wsrep_innobase_mysql_sort(
+				mysql_type, cs->number, sorted, true_len, 
+				REC_VERSION_56_MAX_INDEX_COL_LEN);
 
 			if (wsrep_protocol_version > 1) {
 				memcpy(buff, sorted, true_len);
@@ -4944,8 +4968,9 @@ wsrep_store_key_val_for_row(
 			}
 
 			memcpy(sorted, blob_data, true_len);
-			wsrep_innobase_mysql_sort(
-			       mysql_type, cs->number, sorted, true_len);
+			true_len = wsrep_innobase_mysql_sort(
+				mysql_type, cs->number, sorted, true_len,
+				REC_VERSION_56_MAX_INDEX_COL_LEN);
 
 			memcpy(buff, sorted, true_len);
 
@@ -5008,8 +5033,10 @@ wsrep_store_key_val_for_row(
 							&error);
 				}
 				memcpy(sorted, src_start, true_len);
-				wsrep_innobase_mysql_sort(
-					mysql_type, cs->number, sorted, true_len);
+				true_len = wsrep_innobase_mysql_sort(
+					mysql_type, cs->number, sorted, true_len,
+					REC_VERSION_56_MAX_INDEX_COL_LEN);
+
 				memcpy(buff, sorted, true_len);
 			} else {
 				memcpy(buff, src_start, true_len);
@@ -7269,7 +7296,7 @@ wsrep_append_foreign_key(
 			   wsrep_thd_query(thd) : "void");
 		return DB_ERROR;
 	}
-	byte  key[WSREP_MAX_SUPPORTED_KEY_LENGTH+1];
+	byte  key[WSREP_MAX_SUPPORTED_KEY_LENGTH+1] = {'\0'};
 	ulint len = WSREP_MAX_SUPPORTED_KEY_LENGTH;
 
 	dict_index_t *idx_target = (referenced) ? 
@@ -7439,11 +7466,11 @@ ha_innobase::wsrep_append_keys(
 		uint	len;
 		char 	keyval[WSREP_MAX_SUPPORTED_KEY_LENGTH+1] = {'\0'};
 		char 	*key 		= &keyval[0];
-		KEY	*key_info	= table->key_info;
 		ibool    is_null;
 
 		len = wsrep_store_key_val_for_row(
-			table, 0, key, key_info->key_length, record0, &is_null);
+			table, 0, key, WSREP_MAX_SUPPORTED_KEY_LENGTH, 
+			record0, &is_null);
 
 		if (!is_null) {
 			int rcode = wsrep_append_key(
@@ -7461,9 +7488,14 @@ ha_innobase::wsrep_append_keys(
 		uint i;
 		bool hasPK= false;
 
-		for (i=0; i<table->s->keys && !hasPK; ++i) {
+		for (i=0; i<table->s->keys; ++i) {
 			KEY*  key_info	= table->key_info + i;
-			if (key_info->flags & HA_NOSAME) hasPK = true;
+			if (key_info->flags & HA_NOSAME) {
+				hasPK = true;
+				if (i != table->s->primary_key) {
+					wsrep_thd_set_PA_safe(thd, FALSE);
+				}
+			}
 		}
 
 		for (i=0; i<table->s->keys; ++i) {
@@ -7486,13 +7518,15 @@ ha_innobase::wsrep_append_keys(
 					   table->s->table_name.str, 
 					   key_info->name);
 			}
+			/* !hasPK == table with no PK, must append all non-unique keys */
 			if (!hasPK || key_info->flags & HA_NOSAME ||
 			    ((tab &&
 			      dict_table_get_referenced_constraint(tab, idx)) ||
 			     (!tab && referenced_by_foreign_key()))) {
 
 				len = wsrep_store_key_val_for_row(
-					table, i, key0, key_info->key_length, 
+					table, i, key0, 
+					WSREP_MAX_SUPPORTED_KEY_LENGTH,
 					record0, &is_null);
 				if (!is_null) {
 					int rcode = wsrep_append_key(
@@ -7510,7 +7544,8 @@ ha_innobase::wsrep_append_keys(
 				}
 				if (record1) {
 					len = wsrep_store_key_val_for_row(
-						table, i, key1, key_info->key_length, 
+						table, i, key1, 
+						WSREP_MAX_SUPPORTED_KEY_LENGTH, 
 						record1, &is_null);
 					if (!is_null && memcmp(key0, key1, len)) {
 						int rcode = wsrep_append_key(
@@ -12846,6 +12881,7 @@ static int innobase_wsrep_set_checkpoint(handlerton* hton, const XID* xid)
                 trx_sysf_t* sys_header = trx_sysf_get(&mtr);
                 trx_sys_update_wsrep_checkpoint(xid, sys_header, &mtr);
                 mtr_commit(&mtr);
+                innobase_flush_logs(hton);
                 return 0;
         } else {
                 return 1;

storage/innobase/include/ha_prototypes.h

@@ -291,11 +291,12 @@ UNIV_INTERN int wsrep_innobase_kill_one_trx(void *thd, trx_t *bf_trx, trx_t *vic
 my_bool wsrep_thd_is_BF(void *thd_ptr, my_bool sync);
 //int64_t wsrep_thd_trx_seqno(THD *thd);
 int wsrep_trx_order_before(void *thd1, void *thd2);
-void wsrep_innobase_mysql_sort(int mysql_type, uint charset_number,
-			       unsigned char* str, unsigned int str_length);
-int 
-wsrep_on(void *thd_ptr);
+int wsrep_innobase_mysql_sort(int mysql_type, uint charset_number,
+			      unsigned char* str, unsigned int str_length,
+			      unsigned int buf_length);
+int wsrep_on(void *thd_ptr);
 int wsrep_is_wsrep_xid(const void*);
+my_bool wsrep_thd_set_PA_safe(void *thd_ptr, my_bool safe);
 #endif /* WITH_WSREP */
 /**********************************************************************//**
 Get the current setting of the lower_case_table_names global parameter from

storage/innobase/lock/lock0lock.c

@@ -1820,12 +1820,7 @@ lock_rec_create(
 	automatically of the gap type */
 
 	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
-#ifdef WITH_WSREP
-		ut_ad(!(type_mode & LOCK_REC_NOT_GAP) || 
-		      wsrep_thd_is_BF(trx->mysql_thd, FALSE));
-#else
 		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));
-#endif /* WITH_WSREP */
 
 		type_mode = type_mode & ~(LOCK_GAP | LOCK_REC_NOT_GAP);
 	}
@@ -2109,12 +2104,7 @@ lock_rec_add_to_queue(
 	struct for a gap type lock */
 
 	if (UNIV_UNLIKELY(heap_no == PAGE_HEAP_NO_SUPREMUM)) {
-#ifdef WITH_WSREP
-		ut_ad(!(type_mode & LOCK_REC_NOT_GAP) || 
-		      wsrep_thd_is_BF(trx->mysql_thd, FALSE));
-#else
 		ut_ad(!(type_mode & LOCK_REC_NOT_GAP));
-#endif /* WITH_WSREP */
 
 		/* There should never be LOCK_REC_NOT_GAP on a supremum
 		record, but let us play safe */

storage/innobase/rem/rem0rec.c

@@ -1961,10 +1961,10 @@ wsrep_rec_get_foreign_key(
 				key_len++;
 			}
 			memcpy(buf, data, len);
-			wsrep_innobase_mysql_sort(
+			*buf_len = wsrep_innobase_mysql_sort(
 				(int)(col_f->prtype & DATA_MYSQL_TYPE_MASK),
 				(uint)dtype_get_charset_coll(col_f->prtype),
-				buf, len);
+				buf, len, *buf_len);
 		} else { /* new protocol */
 			if (!(col_r->prtype & DATA_NOT_NULL)) {
 				*buf++ = 0;
@@ -1994,12 +1994,12 @@ wsrep_rec_get_foreign_key(
 			case DATA_MYSQL:
 				/* Copy the actual data */
 				ut_memcpy(buf, data, len);
-				wsrep_innobase_mysql_sort(
+				len = wsrep_innobase_mysql_sort(
 					(int)
 					(col_f->prtype & DATA_MYSQL_TYPE_MASK),
 					(uint)
 					dtype_get_charset_coll(col_f->prtype),
-					buf, len);
+					buf, len, *buf_len);
 				break;
 			case DATA_BLOB:
 			case DATA_BINARY:

storage/innobase/row/row0ins.c

@@ -1760,13 +1760,6 @@ row_ins_scan_sec_index_for_duplicate(
 				lock_type, block, rec, index, offsets, thr);
 		} else {
 
-#ifdef WITH_WSREP
-			if (wsrep_thd_is_BF(thr_get_trx(thr)->mysql_thd, 0)) {
-				if (!(lock_type & LOCK_REC_NOT_GAP)) {
-					lock_type |= LOCK_REC_NOT_GAP;
-				}
-			}
-#endif /* WITH_WSREP */
 			err = row_ins_set_shared_rec_lock(
 				lock_type, block, rec, index, offsets, thr);
 		}

support-files/mysql.server.sh

@@ -364,7 +364,10 @@ case "$mode" in
     # Stop the service and regardless of whether it was
     # running or not, start it again.
     if $0 stop  $other_args; then
-      $0 start $other_args
+      if ! $0 start $other_args; then
+        log_failure_msg "Failed to restart server."
+        exit 1
+      fi
     else
       log_failure_msg "Failed to stop running server, so refusing to try to start."
       exit 1