MDEV-17262: mysql crashed on galera while node rejoined cluster (#895)

This patch contains a fix for the MDEV-17262/17243 issues and
new mtr test.

These issues (MDEV-17262/17243) have two reasons:

1) After an intermediate commit, a transaction loses its status
of "transaction that registered in the MySQL for 2pc coordinator"
(in the InnoDB) due to the fact that since version 10.2 the
write_row() function (which located in the ha_innodb.cc) does
not call trx_register_for_2pc(m_prebuilt->trx) during the processing
of split transactions. It is necessary to restore this call inside
the write_row() when an intermediate commit was made (for a split
transaction).

Similarly, we need to set the flag of the started transaction
(m_prebuilt->sql_stat_start) after intermediate commit.

The table->file->extra(HA_EXTRA_FAKE_START_STMT) called from the
wsrep_load_data_split() function (which located in sql_load.cc)
will also do this, but it will be too late. As a result, the call
to the wsrep_append_keys() function from the InnoDB engine may be
lost or function may be called with invalid transaction identifier.

2) If a transaction with the LOAD DATA statement is divided into
logical mini-transactions (of the 10K rows) and binlog is rotated,
then in rare cases due to the wsrep handler re-registration at the
boundary of the split, the last portion of data may be lost. Since
splitting of the LOAD DATA into mini-transactions is technical,
I believe that we should not allow these mini-transactions to fall
into separate binlogs. Therefore, it is necessary to prohibit the
rotation of binlog in the middle of processing LOAD DATA statement.

https://jira.mariadb.org/browse/MDEV-17262 and
https://jira.mariadb.org/browse/MDEV-17243

1 files changed, 35 insertions, 20 deletions

diff --git a/sql/sql_load.cc b/sql/sql_load.cc
index 8c2f17dac3f..8e0bdcb32b8 100644
--- a/sql/sql_load.cc
+++ b/sql/sql_load.cc
@@ -41,6 +41,7 @@
 #include "sql_trigger.h"
 #include "sql_derived.h"
 #include "sql_show.h"
+#include "debug_sync.h"
 
 extern "C" int _my_b_net_read(IO_CACHE *info, uchar *Buffer, size_t Count);
 
@@ -119,21 +120,43 @@ static bool wsrep_load_data_split(THD *thd, const TABLE *table,
     if (hton->db_type != DB_TYPE_INNODB)
       DBUG_RETURN(false);
     WSREP_DEBUG("intermediate transaction commit in LOAD DATA");
+    wsrep_set_load_multi_commit(thd, true);
     if (wsrep_run_wsrep_commit(thd, true) != WSREP_TRX_OK) DBUG_RETURN(true);
     if (binlog_hton->commit(binlog_hton, thd, true)) DBUG_RETURN(true);
     wsrep_post_commit(thd, true);
     hton->commit(hton, thd, true);
+    wsrep_set_load_multi_commit(thd, false);
+    DEBUG_SYNC(thd, "intermediate_transaction_commit");
     table->file->extra(HA_EXTRA_FAKE_START_STMT);
   }
 
   DBUG_RETURN(false);
 }
-# define WSREP_LOAD_DATA_SPLIT(thd,table,info)		\
-  if (wsrep_load_data_split(thd,table,info)) DBUG_RETURN(1)
+/*
+  If the commit fails, then an early return from
+  the function occurs there and therefore we need
+  to reset the table->auto_increment_field_not_null
+  flag, which is usually reset after calling
+  the write_record():
+*/
+#define WSREP_LOAD_DATA_SPLIT(thd,table,info)		\
+  if (wsrep_load_data_split(thd,table,info))		\
+  {							\
+    table->auto_increment_field_not_null= FALSE;	\
+    DBUG_RETURN(1);					\
+  }
 #else /* WITH_WSREP */
 #define WSREP_LOAD_DATA_SPLIT(thd,table,info) /* empty */
 #endif /* WITH_WSREP */
 
+#define WRITE_RECORD(thd,table,info)			\
+  do {							\
+    int err_= write_record(thd, table, &info);		\
+    table->auto_increment_field_not_null= FALSE;	\
+    if (err_)						\
+      DBUG_RETURN(1);					\
+  } while (0)
+
 class READ_INFO: public Load_data_param
 {
   File	file;
@@ -911,7 +934,7 @@ read_fixed_length(THD *thd, COPY_INFO &info, TABLE_LIST *table_list,
   List_iterator_fast<Item> it(fields_vars);
   Item *item;
   TABLE *table= table_list->table;
-  bool err, progress_reports;
+  bool progress_reports;
   ulonglong counter, time_to_report_progress;
   DBUG_ENTER("read_fixed_length");
 
@@ -1003,11 +1026,8 @@ read_fixed_length(THD *thd, COPY_INFO &info, TABLE_LIST *table_list,
     }
 
     WSREP_LOAD_DATA_SPLIT(thd, table, info);
-    err= write_record(thd, table, &info);
-    table->auto_increment_field_not_null= FALSE;
-    if (err)
-      DBUG_RETURN(1);
-   
+    WRITE_RECORD(thd, table, info);
+
     /*
       We don't need to reset auto-increment field since we are restoring
       its default value at the beginning of each loop iteration.
@@ -1040,7 +1060,7 @@ read_sep_field(THD *thd, COPY_INFO &info, TABLE_LIST *table_list,
   Item *item;
   TABLE *table= table_list->table;
   uint enclosed_length;
-  bool err, progress_reports;
+  bool progress_reports;
   ulonglong counter, time_to_report_progress;
   DBUG_ENTER("read_sep_field");
 
@@ -1124,7 +1144,7 @@ read_sep_field(THD *thd, COPY_INFO &info, TABLE_LIST *table_list,
       {
         Load_data_outvar *dst= item->get_load_data_outvar_or_error();
         DBUG_ASSERT(dst);
-        if (dst->load_data_set_no_data(thd, &read_info))
+        if (unlikely(dst->load_data_set_no_data(thd, &read_info)))
           DBUG_RETURN(1);
       }
     }
@@ -1146,10 +1166,8 @@ read_sep_field(THD *thd, COPY_INFO &info, TABLE_LIST *table_list,
     }
 
     WSREP_LOAD_DATA_SPLIT(thd, table, info);
-    err= write_record(thd, table, &info);
-    table->auto_increment_field_not_null= FALSE;
-    if (err)
-      DBUG_RETURN(1);
+    WRITE_RECORD(thd, table, info);
+
     /*
       We don't need to reset auto-increment field since we are restoring
       its default value at the beginning of each loop iteration.
@@ -1267,13 +1285,10 @@ read_xml_field(THD *thd, COPY_INFO &info, TABLE_LIST *table_list,
     case VIEW_CHECK_ERROR:
       DBUG_RETURN(-1);
     }
-    
+
     WSREP_LOAD_DATA_SPLIT(thd, table, info);
-    err= write_record(thd, table, &info);
-    table->auto_increment_field_not_null= false;
-    if (err)
-      DBUG_RETURN(1);
-    
+    WRITE_RECORD(thd, table, info);
+
     /*
       We don't need to reset auto-increment field since we are restoring
       its default value at the beginning of each loop iteration.