Database inconsistency when using savepoint

[monetdb-team]

Date: 2015-06-12 18:09:44 +0200
From: Frédéric Jolliton <<frederic.jolliton+monetdb>>
To: SQL devs <>
Version: 11.19.15 (Oct2014-SP4)
CC: @njnes

Last updated: 2015-08-28 13:42:52 +0200

Comment 20921

Date: 2015-06-12 18:09:44 +0200
From: Frédéric Jolliton <<frederic.jolliton+monetdb>>

We're studying MonetDB for some time now, but we found yet another bug that worry us. We found that an article[1] on the MonetDB mailing list from 10 (ten) years ago reported the exact same problem. So, maybe it's a regression, or that nobody use SAVEPOINT (or we are using them incorrectly.)

Step to reproduce (here on command line, but same from a Python using the MonetDB modules):

 > $ mclient test
 > START TRANSACTION;
 > CREATE TABLE foo (a int);
 > SAVEPOINT sp;
 > RELEASE SAVEPOINT sp;
 > COMMIT;
 > SELECT * FROM foo;
 [ok]

Start another session:

 > $ mclient test
 > SELECT * FROM foo;
 SELECT: no such table 'foo'

The table seems to vanish at some point.

We got more curious error while testing this (like the table being listed in the system table `tables', but DROP test; reported that it was unknown.) At some other point, instead of getting the error "No such table" we got something completly unrelated: "monetdb.exceptions.DatabaseError: CREATE FUNCTION: name 'right_shift' (inet, inet) already in use". That is really puzzling.

I can provide more details if necessary. We made some automated tests to reproduce that.

Are we doing something wrong? We tested that both on Oct2014-SP4 and development.

[1] https://www.monetdb.org/pipermail/developers-list/2005-September/000154.html

Comment 20951

Date: 2015-07-01 17:10:47 +0200
From: @njnes

savepoint had multiple problems (which should now be fixed). Added the test.

Comment 20959

Date: 2015-07-02 18:08:15 +0200
From: Frédéric Jolliton <<frederic.jolliton+monetdb>>

Wonderful! We're going to test that next week.

Comment 21219

Date: 2015-08-28 13:42:52 +0200
From: @sjoerdmullender

Jul2015 has been released.