Assitance needed for the resolution of memory leak

On 10/11/23 12:15, Sasmit Utkarsh wrote:

Hi Team,

I'm trying to insert multiple duplicate rows into the table using libpq
in C. I'm getting a memory leak reported from addresssanitizer tool.
Please let me know if i can share the code snippet

That is the only way you are going to get an answer, so share away.

Also include the report information from addresssanitizer.

Regards,
Sasmit Utkarsh
+91-7674022625

--
Adrian Klaver
adrian.klaver@aklaver.com

HI Adrian,

I have a main program which actually parses the input data from the files
and inserts them to a table in postgres db. The input can have many
duplicates and we cannot avoid it for now but we have to achieve the
insertions of the records as quickly as possible. Below are the functions
which are called from main to insert the record in a loop untill it reaches
the end of the input file. If the input file has all unique records then
there are no leaks reported by addresssanitizer tool but when there are
duplicates it reports leaks.

Like for an ex: I have provided the test run for a file which has 4
duplicate entries in the input file and below is the leak reported. I have
followed
consistently using PQclear to free all PGresult objects. Not sure why it
references to *PQmakeEmptyPGresult and **pqResultAlloc *even though I have
not used it explicitly. Is anything missing here? How can I avoid memory
leaks? Please assist here

*addresssanitizer report :*

TOTAL rows inserted into fs_data table = 0
TOTAL rows FAILED to insert into table = 4
END TRANSACtION
CLOSE shc_data
FINISH

=================================================================
==336368==ERROR: LeakSanitizer: detected memory leaks

*Direct leak of 640 byte(s) in 4 object(s) allocated from: #0 0xf7a1d0bb
in __interceptor_malloc
(/nix/store/3qz5qcx23d522i1fivjgh2nw0bs0pywf-gcc-10.3.0-lib/lib/libasan.so.6+0xaf0bb)
#1 0xf792b7f3 in PQmakeEmptyPGresult
(/nix/store/0rh8jnm4xgav5916nasd1d2ciy4nccsp-postgresql-13.7-lib/lib/libpq.so.5+0x117f3)Indirect
leak of 8192 byte(s) in 4 object(s) allocated from: #0 0xf7a1d0bb in
__interceptor_malloc
(/nix/store/3qz5qcx23d522i1fivjgh2nw0bs0pywf-gcc-10.3.0-lib/lib/libasan.so.6+0xaf0bb)
#1 0xf792b3bc in pqResultAlloc
(/nix/store/0rh8jnm4xgav5916nasd1d2ciy4nccsp-postgresql-13.7-lib/lib/libpq.so.5+0x113bc)*

SUMMARY: AddressSanitizer: 8832 byte(s) leaked in 8 allocation(s).

*Insert function:*

int postgreSQL_insert(PGconn *conn, int fa, int ft, int ord, unsigned char
*data, int blk_size, char *recid, char *xaddr)
{
PGresult *res;
int rc = 0;
int rc2;
char *ErrorMsg;
char cur_recid[5] = "\0\0\0\0\0";
int ft3;
int ord3;
char Command[250];;

uint32_t fa2 = htonl(fa);
uint32_t ft2 = htonl(ft);
uint32_t ord2 = htonl(ord);
uint32_t blk_size2 = htonl(blk_size);
uint32_t xaddr_size = strlen(xaddr);
uint32_t recid_size = strlen(recid);

char *fa_val = (char *)&fa2;
char *ft_val = (char *)&ft2;
char *ord_val = (char *)&ord2;
char *blk_size_val = (char *)&blk_size2;

int nParams = 7;
//char *command = "INSERT INTO fs_data (file_address,
face_type, ordinal, xaddr, recid, blk_size, data) VALUES ($1, $2, $3, $4,
$5, $6, $7)";
//char *command = "INSERT INTO fs_data (file_address,
face_type, ordinal, xaddr, recid, blk_size, data) VALUES ($1, $2, $3, $4,
$5, $6, $7) ON CONFLICT (file_address) DO NOTHING";
char *command = "*INSERT INTO fs_data (file_address,
face_type, ordinal, xaddr, recid, blk_size, data) VALUES ($1, $2, $3, $4,
$5, $6, $7) ON CONFLICT (file_address) DO NOTHING RETURNING file_address";*
Oid paramTypes[7] = {23, 23, 23, 25, 25, 23, 17}; // {int, int,
int, text, text, int, bytea}
const char *paramValues[7] = {fa_val, ft_val, ord_val, xaddr, recid,
blk_size_val, data};
int paramLengths[7] = {sizeof(fa2), sizeof(ft2), sizeof(ord2),
xaddr_size, recid_size, sizeof(blk_size2), blk_size};
int paramFormats[7] = {1, 1, 1, 1, 1, 1, 1}; // identify all
parameters as binary
int resultFormat = 1; // Result will be in binary
format.

if(init_only == false)
{
rc = postgreSQL_delete(conn, fa);
}

if ( sql_debug ) { fprintf(stderr,"PostgreSQL_insert INPUT fa %i ft %i ord
%i xaddr %s id %s blk_size %i\n",fa,ft,ord,xaddr,recid,blk_size); }
/* INSERT row in table fs_data */

if ( sql_debug ) { fprintf(stderr,"Command : (%s)\n", command); }
if ( sql_debug ) { fprintf(stderr,"Param. 0: faddr (%i)\n", fa); }
if ( sql_debug ) { fprintf(stderr,"Param. 1: face_type (%i)\n", ft); }
if ( sql_debug ) { fprintf(stderr,"Param. 2: ordinal (%i)\n", ord); }
if ( sql_debug ) { fprintf(stderr,"Param. 3: xaddr (%s)\n", xaddr); }
if ( sql_debug ) { fprintf(stderr,"Param. 4: recid (%s)\n", recid); }
if ( sql_debug ) { fprintf(stderr,"Param. 5: blk_size (%i)\n",
blk_size); }
res = PQexecParams(conn, // shc_data database connection
command, // "INSERT INTO fs_data ...)";
nParams, // parameter count
paramTypes, // integer, integer, integer, bytea
paramValues, // file_address, face_type, ordinal, data
paramLengths, // Lenght of each parameter
paramFormats, // all parameters are binary format
resultFormat); // result is binary

//fprintf(stderr," res = %s\n",PQresStatus(PQresultStatus(res))); // show
the value of res

if ( ( PQresultStatus(res) == PGRES_TUPLES_OK)
&& (PQntuples(res) == 0) ) // if tuples == 0, we hit the ON CONFLICT, so
report the ERROR
{
rc = 1;
fprintf(stderr, "\nERROR: DUPLICATE file address %s\n", xaddr);
fprintf(stderr, "FAIL : INSERT into fs_data file_address = %-11i (%s) recid
= %s face_type = %i ordinal = %i\n", fa,xaddr,recid,ft,ord);
rc2 = SQL_SELECT_recid(fa, conn, cur_recid);
if(rc2 == 0)
{
sprintf(Command,"SELECT face_type, ordinal from fs_data where
file_address=%i",fa);
//fprintf(stderr,"Command = %s\n",Command);
res = PQexec(conn, Command);
if (PQresultStatus(res) == PGRES_TUPLES_OK)
{
int num_rows = PQntuples(res);
if (num_rows > 0)
{
ft3 = atoi(PQgetvalue(res, 0, 0));
ord3= atoi(PQgetvalue(res, 0, 1));
fprintf(stderr, "CURRENT data: in fs_data file_address = %-11i (%s) recid
= %s face_type = %i ordinal = %i\n", fa,xaddr,cur_recid,ft3,ord3);
}
else
{
fprintf(stderr,"FAIL postgreSQL_insert: No rows returned for
file_address=%i\n",fa);
}
}
else
fprintf(stderr,"FAIL postgreSQL_insert: Query execution FAILED: %s\n",
PQresultErrorMessage(res));

PQclear(res);
}
}
else if ( PQresultStatus(res) != PGRES_TUPLES_OK) // something else went
wrong
{
rc = 1;
fprintf(stderr,"ERROR: %s\n",PQresultErrorMessage(res));
PQclear(res);
}
else
{
rc = 0;
if ( sql_debug ) { fprintf(stderr,"INSERT SUCCESSFUL\n"); }
PQclear(res);
}

if ( sql_debug1 ) { printPostgreSQL_SELECT_results(fa,conn); }

return rc;
}

Regards,
Sasmit Utkarsh
+91-7674022625

On Thu, Oct 12, 2023 at 1:06 AM Adrian Klaver <adrian.klaver@aklaver.com>
wrote:

Sasmit Utkarsh <utkarshsasmit@gmail.com> writes:

Like for an ex: I have provided the test run for a file which has 4
duplicate entries in the input file and below is the leak reported. I have
followed
consistently using PQclear to free all PGresult objects.

No you haven't: you've got two PQexec-something calls and
only one PQclear. If control reaches here:

res = PQexec(conn, Command);

that will overwrite your only pointer to the PQexecParams
result, and you won't clear it.

BTW, if this is what your code actually looks like in your editor,
it's no wonder you can't follow its basic control flow. Indent
according to the brace structure, and your life will get easier.

regards, tom lane

Thanks Tom, It did work, and yeah I do have indentation in my editor, there
was some formatting issue due to the copy and paste of the code.
:)

Regards,
Sasmit Utkarsh
+91-7674022625

On Thu, Oct 12, 2023 at 2:07 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:

That's why you /attach/ files, not paste text.

On 10/11/23 16:19, Sasmit Utkarsh wrote:

Thanks Tom, It did work, and yeah I do have indentation in my editor,
there was some formatting issue due to the copy and paste of the code.
:)

Regards,
Sasmit Utkarsh
+91-7674022625

On Thu, Oct 12, 2023 at 2:07 AM Tom Lane <tgl@sss.pgh.pa.us> wrote:

Sasmit Utkarsh <utkarshsasmit@gmail.com> writes:

Like for an ex: I have provided the test run for a file which has 4
duplicate entries in the input file and below is the leak reported.

I have

followed
consistently using PQclear to free all PGresult objects.

No you haven't: you've got two PQexec-something calls and
only one PQclear.  If control reaches here:

res = PQexec(conn, Command);

that will overwrite your only pointer to the PQexecParams
result, and you won't clear it.

BTW, if this is what your code actually looks like in your editor,
it's no wonder you can't follow its basic control flow. Indent
according to the brace structure, and your life will get easier.

                        regards, tom lane

--
Born in Arizona, moved to Babylonia.