Segmentation Fault on some requests handling ROWID

[gduvalsc]

I have written a python program able to "replay" Oracle requests captured in the Shared Pool of an Oracle Database. This program is using the last version of cx_Oracle (7.1.2)

[oracle@oraee122 ~]$ python
Python 2.7.5 (default, Nov 1 2018, 03:12:47)
[GCC 4.8.5 20150623 (Red Hat 4.8.5-36.0.1)] on linux2
Type "help", "copyright", "credits" or "license" for more information.

import cx_Oracle
cx_Oracle.version
'7.1.2'

In this example the version of Python used is 2.7 but the same problem occurs when python is in version 3.

To realize a test case, you have just to install the demo located in any Oracle Database at $ORACLE_HOME/demo/schema/human_resources.

After installation (run hr_main.sql under sqlplus), a HR schema is created and in this schema an "EMPLOYEE" table is also created.

The following python program produces a SEGMENTATION FAULT:

[oracle@oraee122 ~]$ cat x.py
import cx_Oracle, logging, coloredlogs
coloredlogs.install(milliseconds=True)
conn = cx_Oracle.connect('hr/hr@localhost:1521/orclpdb1')
curs = conn.cursor()
curs.execute("select e.rowid from employees e, employees")
columns = [i[0] for i in curs.description]
logging.info('Cursor description: ' + str(columns))
for row in curs:
logging.info(str(row))
[oracle@oraee122 ~]$ python x.py
2019-06-04 10:30:32,324 oraee122 root[2256] INFO Cursor description: ['ROWID']
2019-06-04 10:30:32,325 oraee122 root[2256] INFO ('AAAR6pAAMAAAADNABD',)
2019-06-04 10:30:32,325 oraee122 root[2256] INFO ('AAAR6pAAMAAAADNABD',)
Segmentation fault

The database version (here ORACLE 12.2) and the Oracle Instant client used are not involved in the problem.

I will join later the stack produced in the core dump, but clearly the problem is tightly coupled with the rowid handling in a cursor.

[gduvalsc]

For information, if the request "select e.rowid from employees e, employees" is transformed to "select rowidtochar(e.rowid) from employees e, employees", the program runs normally!

This can be a workaround for some people working with rowids.

In my case (replay in a test environment captured requests coming from production environment), this workaround is not valid.

[anthony-tuininga]

This is a known issue, I believe. Take a look at this StackOverflow question. Try setting the environment variable ORA_OCI_NO_OPTIMIZED_FETCH to the value 1 before you run the query to avoid the issue. Your other workaround also avoids the issue but probably won't perform as well.

[anthony-tuininga]

FYI, the problem only occurs when you have duplicate rowids being fetched.

[gduvalsc]

Thanks a lot, Anthony! I used your advice and indeed, the problem is solved!
It's a valid workaround for me.
Congrats for your work around cx_Oracle!

[anthony-tuininga]

You're welcome!