Innodb – how to remove Mariadb temp files

innodbmariadb

I have a /mariadb_tmp directory (as specified in my mariadb-server.cnf) which is filled with files dating back to last November.

-rw-rw---- 1 mysql mysql   1032192 Nov 19 07:47 #sql899_5d5_0.ibd
-rw-rw---- 1 mysql mysql    278528 Nov 19 09:30 #sql899_ab8f_0.ibd
-rw-rw---- 1 mysql mysql    278528 Nov 19 09:30 #sql899_ad17_0.ibd
-rw-rw---- 1 mysql mysql    638976 Nov 19 09:32 #sql899_baf3_0.ibd
-rw-rw---- 1 mysql mysql    409600 Nov 19 09:32 #sql899_bbec_0.ibd
-rw-rw---- 1 mysql mysql    262144 Nov 19 09:35 #sql899_c96a_0.ibd
...
-rw-rw---- 1 mysql mysql    704512 Jan 16 10:08 #sql899_603c0c4_1.ibd
-rw-rw---- 1 mysql mysql    540672 Jan 16 10:08 #sql899_603c123_0.ibd
-rw-rw---- 1 mysql mysql 494927872 Jan 16 10:20 #sql899_603eb7b_0.ibd
-rw-rw---- 1 mysql mysql    606208 Jan 16 10:27 #sql899_60470ff_1.ibd

Googling on 'delete old mariadb temp files' leads to advice such as:

drop table `#sql899_5d5_0`;

which results in the error 'no database selected.' Of course, I don't know what database each tmp files comes from, so I don't know if that would work.

There are no *.frm files to go with these, and they are in the specified tmpdir directory, not the datadir.

Any idea how to clean these up? They are taking up a bunch of space on a disk that we really can't expand (for non-technical reasons) nor should we have to keep these around that long.

Thanks,

Ed Greenberg

Best Answer

After checking they aren't attached to an active process, ls -la /proc/$(pidof mysqld)/fd, or fuser {filename}, they can be removed with rm.

From MariaDB-10.4 onwards you shouldn't see these created on disk again ( MDEV-15584).

Related Solutions

MySQL Replication – Temp Table Still Exists on Slave

I hope I can answer this adequately because I hate temp tables that use InnoDB. Here is why

InnoDB Architecture

In this diagram, the data dictionary has that temp table as an entry. Once the temp table is done being used for ALTER TABLE, the data dictionary entry should disappear.

Here are your questions and my answer for them ...

1.. When alter has been successfully done then why this temp table still exist and not deleted.

Under normal circumstances, when a DB Connection terminates, all open temp tables are supposed to be dropped. The only reason a temp table will linger after an ALTER TABLE is done is if the DB Connection running the ALTER TABLE terminates before the temp table can be dropped. It is possible that the rollback segments (See Diagram) has the original info of the table before the ALTER TABLE was issued.

Since you mentioned this as happening on the Slave, the only suspect (or person of interest) for this problem would be the SQL thread for MySQL Replication. Why? It is the owner of the temp tables need to complete the SQL being executed from the relay logs. If you kill a SQL thread in a manner other than STOP SLAVE; while it is performing an ALTER TABLE, it may not drop temp tables in an orderly manner. Thus, you will have a lingering temp table regardless of the storage engine.

2.. Now how can I delete this temp table and what will be its impact.

You could go ahead and delete the .frm and .ibd. I only say this because it is temp table. The table should have disappeared. The impact ? The data dictionary will still have the temp table registered. Just a few bytes get wasted.

Don't worry about the temp table name being used again. Look at the name of the temp table

#sql-7a87_230c32.ibd

There are 10 hexadecimal digits. The chances of reusing that temp table name is 1 in ₁₆10 or ₂40 which 1,099,511,627,776. I think you will be OK in this.

Don't delete them just yet. Look at the your third question ...

3.. May I read this temp file or know by any command that for which table this temp table was created.

I understand the need for this question. Evidently, you want to know what the table was that was having the ALTER TABLE done. Unfortunately, you cannot read any temp table through standard SQL.

However, you could probably rename the .frm and use a utility to read the table structure.

Here is an old post of mine to do this with a .frm in the Windows environment : How can extract the table schema from just the .frm file?

EPILOGUE

Even though the temp table no longer has a useless purpose at this point, deleting the InnoDB temp table there is that 1 in trillion chance that InnoDB be behave erratically.

Your last resort when this happens is to perform a fully InnoDB Cleanup. See my post Howto: Clean a mysql InnoDB storage engine? in StackOverflow

UPDATE 2014-06-10 08:36 EDT

It should be evident from the file timestamps and the commentary in your second update that a shutdown must have taken place in the middle of an ALTER TABLE. All transactional information for MVCC regarding the InnoDB table was held in a state of suspended animation in one of the following places:

Transaction Logs (a.k.a. Redo Logs, which are the files ib_logfile0 and ib_logfile1),
Rollback Segments & Undo Space (located in the system tablespace file ibdata1)
Double Write Buffer (located in the system tablespace file ibdata1)

When you started mysql back up, the MVCC for the table was replayed to bring the original table into a consistent state. That may result in either the table being rolled back to the original state before the ALTER TABLE or if everything passed through the Redo Logs, the table now existing in the new design. The fact that the .ibd file has a later timestamp show that MVCC must have been applied to it on the startup of mysql (which includes crash recovery).

Once the shutdown happened, the InnoDB table lost its identity as a temp table and became a regular table in the eyes of mysqld. Thus, you should be able to delete the temp files with impunity.

Mysql – MariaDB/InnoDB: Temp file write failure

Look carefully at the error message

ERROR 1296 (HY000): Got error 64 'Temp file write failure' from InnoDB

What is error 64 ? perror says the following:

$ perror 64
OS error code  64:  Machine is not on the network

The ALTER TABLE command is create a temp file with the new compression. However, it is trying to create a temp file but it cannot write to the directory tmpdir is mapped to.

If tmpdir is mapped to a SAN that disconnected from the network, that error would make sense.
If tmpdir ran out of space, the error message makes no sense but the end result would still stop ALTER TABLE from completing its operation.

If this problem is repeatable, you may want to try doing this change manually.

create table positions_comp like positions;
alter table positions_comp KEY_BLOCK_SIZE=8;
insert into positions_comp select * from positions;
alter table positions rename positions_old;
alter table positions_comp rename positions;