Commit 1f8d86b4 authored by Arun Kuruvila's avatar Arun Kuruvila

Bug #16324629 : SERVER CRASHES ON UPDATE/JOIN FEDERATED +

                LOCAL TABLE WHEN ONLY 1 LOCAL ROW

Description: When updating a federated table with UPDATE...
JOIN, the server consistently crashes with Signal 11 when
only 1 row exists in the local table involved in the join 
and that 1 row can be joined with a row in the federated 
table.

Analysis: Interaction between the federated engine and the
optimizer results in the crash. In our scenario, ie, local
table having only one row, the program is following a 
different path because the table is treated as a constant
table by the join optimizer. So in this scenario 
"index_read()" is happening in the prepare phase,
since optimizer plan is different for constant table joins.
In this case, "index_read_idx_map()" (inside handler.cc) is
calling "index_read()" and inside "index_read()", matching 
rows are fetched and "stored_result" gets populated by 
calling "store_result()". And just after "index_read()", 
"index_end()" function is called. And in the "index_end()",
its freeing the "stored_result" by calling "free_result()".
So when it reaches the execution phase, in "position()" 
function, we are getting assertion at 
"DBUG_ASSERT(stored_result);". In all other scenarios (ie, 
table with more than 1 row), optimizer plan is different 
and "index_read()" is happening in the execution phase.

Fix: So my fix is to have a separate ha_federated member
function for "index_read_idx_map()" which will handle 
federated engine separately. So that position() will be 
called before index_end() call in constant table scenario.
parent 64b697ca
/* Copyright (c) 2004, 2011, Oracle and/or its affiliates. All rights reserved. /* Copyright (c) 2004, 2013, Oracle and/or its affiliates. All rights reserved.
This program is free software; you can redistribute it and/or modify This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by it under the terms of the GNU General Public License as published by
...@@ -2349,6 +2349,22 @@ int ha_federated::delete_row(const uchar *buf) ...@@ -2349,6 +2349,22 @@ int ha_federated::delete_row(const uchar *buf)
DBUG_RETURN(0); DBUG_RETURN(0);
} }
int ha_federated::index_read_idx_map(uchar *buf, uint index, const uchar *key,
key_part_map keypart_map,
enum ha_rkey_function find_flag)
{
int error= index_init(index, 0);
if (error)
return error;
error= index_read_map(buf, key, keypart_map, find_flag);
if(!error && stored_result)
{
uchar *dummy_arg=NULL;
position(dummy_arg);
}
int error1= index_end();
return error ? error : error1;
}
/* /*
Positions an index cursor to the index specified in the handle. Fetches the Positions an index cursor to the index specified in the handle. Fetches the
......
/* Copyright (c) 2004, 2011, Oracle and/or its affiliates. All rights reserved. /* Copyright (c) 2004, 2013, Oracle and/or its affiliates. All rights reserved.
This program is free software; you can redistribute it and/or modify This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by it under the terms of the GNU General Public License as published by
...@@ -217,6 +217,9 @@ class ha_federated: public handler ...@@ -217,6 +217,9 @@ class ha_federated: public handler
int delete_row(const uchar *buf); int delete_row(const uchar *buf);
int index_init(uint keynr, bool sorted); int index_init(uint keynr, bool sorted);
ha_rows estimate_rows_upper_bound(); ha_rows estimate_rows_upper_bound();
int index_read_idx_map(uchar *buf, uint index, const uchar *key,
key_part_map keypart_map,
enum ha_rkey_function find_flag);
int index_read(uchar *buf, const uchar *key, int index_read(uchar *buf, const uchar *key,
uint key_len, enum ha_rkey_function find_flag); uint key_len, enum ha_rkey_function find_flag);
int index_read_idx(uchar *buf, uint idx, const uchar *key, int index_read_idx(uchar *buf, uint idx, const uchar *key,
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment