CSV engine reports "out of memory" when you select from a missing file

[Monty Taylor]

I'm guessing this will get touched anyway once Stewart's new no-FRM is pushed. BUT...

create table bug15205_2 (val int not null) engine=csv;
--remove_file $MYSQLTEST_VARDIR/master-data/test/bug15205.CSV
select * from bug15205;

This _should_ get an error about can't open the data file. Instead, we get:

Out of memory; check if drizzled or some other process uses all available memory; if not, you may have to use 'ulimit' to allow drizzled to use more memory or you can add more swap space

[Jay Pipes (jaypipes)]

Confirmed:

$> cd tests/
$> ./dtr --start-and-exit
$> ../client/drizzle --user=root --port=9306

drizzle> create table bug15205_2 (val int not null) engine=csv;
Query OK, 0 rows affected (0.02 sec)

switch to another shell, then:

$> cd var/master_data/test
$> rm bug15205_2.CSV

switch back to the Drizzle client and:

drizzle> select * from bug15205_2;
ERROR 1041 (HY000): Out of memory; check if drizzled or some other process uses all available memory; if not, you may have to use 'ulimit' to allow drizzled to use more memory or you can add more swap space

NOTES:

If you delete the .frm file and not the .CSV file, a correct error is reported:

drizzle> select * from bug15205_2;
ERROR 1146 (42S02): Table 'test.bug15205_2' doesn't exist

If you delete the .CSM file and not the .CSV or .frm files, an expected but well-handled crash occurs:

drizzle> select * from bug15205_2;
Empty set, 1 warning (0.01 sec)

drizzle> show warnings;
+-------+------+----------------------------------------------------------------+
| Level | Code | Message |
+-------+------+----------------------------------------------------------------+
| Error | 1194 | Table 'bug15205_2' is marked as crashed and should be repaired |
+-------+------+----------------------------------------------------------------+
1 row in set (0.00 sec)

drizzle> repair table bug15205_2;
+-----------------+--------+----------+----------+
| Table | Op | Msg_type | Msg_text |
+-----------------+--------+----------+----------+
| test.bug15205_2 | repair | status | OK |
+-----------------+--------+----------+----------+
1 row in set (0.00 sec)

drizzle> select * from bug15205_2;
Empty set (0.00 sec)

[Jay Pipes (jaypipes)]

Assigning to low-hanging-fruit as this is likely an easy fix...probably just a wrong error message in the handler, but need to check of course.

[Padraig O'Sullivan (posulliv)]

I had a quick look at this. I think the problem is in storage/csv/ha_tina.cc lines 785-786 in ha_tina::open():

 if (!(share= get_share(name, table)))
    return(HA_ERR_OUT_OF_MEM);

I changed this to:

 if (!(share= get_share(name, table)))
    return(ENOENT);

and then when I try to reproduce the issue as Jay outlined, I get:

drizzle> use test
Database changed
drizzle> create table t1 (val int not null) engine=csv;
Query OK, 0 rows affected (0.01 sec)

Switch to another shell and delete the .CSV file. Now, switch back to drizzle client.

drizzle> select * from t1;
ERROR 1017 (HY000): Can't find file: 't1' (errno: 2)
drizzle>

Does this look right to you guys?

[Jay Pipes (jaypipes)]

Yep, that looks about right to me. Can you push to a branch on Launchpad? Just do:

bzr push lp:~posulliv/drizzle/bug-fixes

and we'll review and merge. :)

Cheers!

Jay

[Padraig O'Sullivan (posulliv)]

I tried to create a test case for this as Jay mentioned. I wasn't too sure what the procedure for this is so I added the following to csv.test:

#
# Bug#311104
# CSV engine reports "out of memory" when you select
# from a missing file
#
--disable_warnings
drop table if exists t1;
--enable_warnings

create table t1 (val int not null) engine=csv;
--remove_file $MYSQLTEST_VARDIR/master-data/test/t1.CSV

--error 1017
select * from t1;

I pushed my modified version of csv.test and csv.result to my branch on Launchpad. Is this the correct way to submit a test case with a bug fix? Or should a new file be created with the test case in it?

I realize now I should have created the test case before working on the bug but I'll know to do this in the future.

Thanks, Padraig

[Jay Pipes (jaypipes)]

Yes, that it perfect, Padraig. Only thing now is to click the "Propose to merge into another branch" link on the branch page which contains the patch.

[Padraig O'Sullivan (posulliv)]

Fixed in trunk.