Details
-
Bug
-
Status: Closed (View Workflow)
-
Major
-
Resolution: Fixed
-
10.2.13
-
None
-
CentOS 7 (3.10.0-693.17.1.el7.x86_64) on Virtual Box VM with 8 GB of RAM and 16 GB of storage. Note, I am also seeing the issue with a SuperMicro server running CentOS 7 (3.10.0-693.17.1.el7.x86_64) bare metal.
Description
I am attempting to move to the MariaDB 10.2 line for my production application. Previously my upgrade to 10.2 was blocked by MDEV-14857. Now in 10.2.13 this issue has been resolved, however I am running into a new problem. I have a nightly job that runs in a stored procedure that fails in MariaDB 10.2.13. The error I am getting is below:
ERROR 1114 (HY000): The table '#sql-daf_13' is fulldone
I realize that this error is normally due to a configuration issue, however look at the end of the error string, it says "done" after full.
I have narrowed down the issue using the following reproduction steps:
- I have uploaded a sql script to ftp.askmonty.org/private called brycejlowe-permtest.sql
- Install MariaDB 10.2.13 and source in the file, it creates a schema called scratch and a table called permtest1 which has about 422,000 rows in it.
- Execute the following statements against that table:
CREATE TEMPORARY TABLE scratch.temptest1 LIKE scratch.permtest1;
INSERT INTO scratch.temptest1 SELECT * FROM scratch.permtest1;
CREATE TEMPORARY TABLE scratch.temptest2 SELECT * FROM scratch.temptest1;
ALTER TABLE scratch.temptest2 ADD INDEX `part`(`Part_no`);
The result in MariaDB 10.2.8 through 10.2.12 is a temporary table scratch.temptest2 with an index called part. However the result in 10.2.13 is the error:
ERROR 1114 (HY000): The table '#sql-daf_13' is fulldone
Attachments
Activity
brycejlowe,
Thanks for the report and test case.
Can we use data from brycejlowe-permtest.sql on public resources, or do you prefer it to be kept confidential?
Reproducible on 10.2.13, not on 10.3.4, not on current 10.2
Message is "table is full", there are leftovers after " Stage: 1 of 2 'copy to tmp table' 19.2% of stage done", and the word "done" remains.
|
|
CREATE TABLE `t1` ( |
`id` int(10) unsigned NOT NULL AUTO_INCREMENT, |
`Part_No` varchar(15) NOT NULL, |
`IQty` float DEFAULT 0, |
`DQty` float DEFAULT 0, |
`TQty` float DEFAULT 0, |
`OHQty` float DEFAULT 0, |
`OOQty` float DEFAULT 0, |
`RQty` float DEFAULT 0, |
`Level` int(10) unsigned DEFAULT 1, |
`Part_Type` char(1) DEFAULT '1', |
`Period` int(10) unsigned DEFAULT NULL, |
`EcoList` varchar(50) NOT NULL, |
`MQty` float DEFAULT NULL, |
`BQty` float DEFAULT NULL, |
PRIMARY KEY (`id`), |
KEY `Part` (`Part_No`), |
KEY `pnp` (`Part_No`,`Period`) |
);
|
|
|
INSERT INTO t1 (id,Part_No,EcoList,MQty,BQty) SELECT seq,seq,seq,seq,seq FROM seq_1_to_1000000; |
|
|
CREATE TEMPORARY TABLE tt1 LIKE t1; |
INSERT INTO tt1 SELECT * FROM t1; |
|
|
CREATE TEMPORARY TABLE tt2 SELECT * FROM tt1; |
ALTER TABLE tt2 ADD INDEX `part`(`Part_no`); |
Apparently it was fixed in 10.2 tree just a few days ago, by this commit:
commit 852c35f571b1b7454aab5768899e2faee8f941d4
|
Author: Marko Mäkelä <marko.makela@mariadb.com>
|
Date: Mon Feb 19 14:58:05 2018 +0200
|
|
|
MDEV-11581 follow-up fix: Correct a condition
|
|
fsp_fill_free_list(): Correctly determine whether the temporary
|
tablespace file should be extended in order to respond to a
|
page allocation request. The inverted condition was noticed
|
by Thiru when he analyzed MDEV-13013.
|
NOTE: I am not seeing the issue in 10.1.31.