MySQL Bugs: #16017: Memory leak in mysql embedded library

Bug #16017	Memory leak in mysql embedded library
Submitted:	27 Dec 2005 13:50	Modified:	2 Aug 2006 17:52
Reporter:	Noam Ambar	Email Updates:
Status:	Closed	Impact on me:	None
Category:	MySQL Server: Embedded Library ( libmysqld )	Severity:	S3 (Non-critical)
Version:	4.1.13, 5.1	OS:	Linux (Linux (SuSE 9 kernel 2.6))
Assigned to:	Alexey Botchkov	CPU Architecture:	Any

Description:
When I run the program below (mysql_embedded), it causes a memory leak of ~6MB. When I add more iterations to the loop, it consumes the entire machine's memory. valgrind shows that the leak is on mysql_server_init but I'm not sure that this is not true since the leak is increased when I add iterations to the loop. Table T1 is an empty table with one column (int i1).

I ran the same program linked with the client library and changed the connection to a local mysql server and it worked fine. There were not leaks on both server and client.

Also when I replace the "SELECT" statement with "UPDATE" statement there is no memory leak so I guess that it is related to the result set but I don't know how. 

I get the same results on MySql4 and MySql5. 

Any response will be appreciated. 

Thanks, 

Noam 

valgrind output: 

==9484== 6004992 bytes in 2911 blocks are definitely lost in loss record 14 of 16 
==9484== at 0x1B904595: malloc (vg_replace_malloc.c:130) 
==9484== by 0x806757A: my_malloc (my_malloc.c:35) 
==9484== by 0x8061669: init_dynamic_array (in /a/nfs/iil/iec/sws/work/nambar/development/head/common/cpp/mysql/testfeeder) 
==9484== by 0x8165A61: _hash_init (in /a/nfs/iil/iec/sws/work/nambar/development/head/common/cpp/mysql/testfeeder) 
==9484== by 0x80AFF94: init_max_user_conn() (sql_parse.cc:503) 
==9484== by 0x8073AF2: init_embedded_server (mysqld.cc:3115) 
==9484== by 0x8056813: mysql_server_init (libmysql.c:162) 
==9484== by 0x804D6E8: CMySqlDatabase::startup(std::string&) (CMySqlDatabase.cpp:76) 
==9484== by 0x805241E: main (main.cpp:13)

How to repeat:
int main() 
{ 

string path = "/tmp/testarea"; 
CMySqlDatabase::startup(path); 
mysql_thread_init(); 

MYSQL* handle = mysql_init(NULL); 
mysql_real_connect(handle, NULL, NULL, NULL, "netstar", 0, NULL, 0); 
for(int i=0; i<3000; i++) 
{ 
cout << "iteration " << i << endl; 
MYSQL_STMT* stmt = mysql_stmt_init(handle); 
char* command = "SELECT 1 FROM T1"; 
//char* command = "UPDATE T1 SET i1 = 1"; 
mysql_stmt_prepare(stmt, command, strlen(command)); 
mysql_stmt_execute(stmt); 
mysql_stmt_free_result(stmt); 
mysql_stmt_close(stmt); 
} 
mysql_close(handle); 

} 
}

I wasn't able to test with latest 5.0 source, but with 4.1 i see the process growing over time (a megabyte every few seconds) when using the following source:

#include <stdlib.h>
#include <stdio.h>

#include <mysql.h>

static char *server_groups[] = {
  "embedded",
  "server",
  (char *)NULL
};

int main(int argc, char **argv) 
{
  MYSQL *mysql = NULL;
  int i;
  char query[256];

  if (mysql_server_init(argc, argv, server_groups)) {
	fputs("server init failed", stderr);
	return EXIT_FAILURE;
  }

  mysql_thread_init(); 
  
  MYSQL* handle = mysql_init(NULL); 
   mysql_options(handle, MYSQL_READ_DEFAULT_GROUP, "libmysqld_client");
   mysql_options(handle, MYSQL_OPT_USE_EMBEDDED_CONNECTION, NULL);  mysql_real_connect(handle, NULL, NULL, NULL, "test", 0, NULL, 0); 
  for(i=0; 1; i++) 
  { 
	printf("iteration %d\n", i); 
	MYSQL_STMT* stmt = mysql_stmt_init(handle); 
	char* command = "SELECT 1 FROM i1"; 
//char* command = "UPDATE T1 SET i1 = 1"; 
	mysql_stmt_prepare(stmt, command, strlen(command)); 
	mysql_stmt_execute(stmt); 
 	mysql_stmt_store_result(stmt);
	while (!mysql_stmt_fetch(stmt)) {}
	mysql_stmt_free_result(stmt); 
	mysql_stmt_close(stmt); 
  } 
  mysql_close(handle); 
  
}

Thank you for verifying it.

Changing category to libmysqld

After making some more tests I found out that this problem is related to the structure of the database. This problem exists in mysql5 when using database that was originally created with mysql4. If the database was originally created with mysql5, the memory leak does not show up

A patch for this bug has been committed. After review, it may
be pushed to the relevant source trees for release in the next
version. You can access the patch from:

  http://lists.mysql.com/commits/7155

Noted in 4.1.21 changelog.

Some memory leaks in the libmysqld embedded server were corrected.

Returning report to NDI status pending push of fix into 5.1.

Noted in 5.1.12 changelog.