Comments on: table_cache negative scalability

By: Chris Howells

Chris Howells — Tue, 23 Aug 2011 10:39:32 +0000

This has bitten me once or twice too. It still seems to be an issue in the most recent Percona release of MySQL 5.5.14 too.

A patch has been written, have you thought about including it in Percona?

http://bugs.mysql.com/bug.php?id=49177
http://lists.mysql.com/commits/121507

By: Shahryar Ghazi

Shahryar Ghazi — Tue, 07 Sep 2010 19:21:50 +0000

we recently faced a similar issue so I wrote a tutorial here –> http://www.geeksww.com/tutorials/database_management_systems/mysql/installation/mysql_tablecache_informationschema_and_open_files_limit.php

By: DaveJ

DaveJ — Wed, 14 Apr 2010 14:14:40 +0000

Great Post, Thank you all for your info.

Question:

After kicking off mk-table-checksum in the middle of the night I start getting errors such as:

100414 0:21:52 [ERROR] Error in accept: Too many open files
100414 0:21:52 [ERROR] /usr/local/mysql/bin/mysqld: Can’t open file: ‘./wc_miehr_posasmp/panel_decert.frm’ (errno: 24)
100414 0:21:52 [ERROR] /usr/local/mysql/bin/mysqld: Can’t open file: ‘./wc_miehr_posasmp/panel_decert.frm’ (errno: 24)

I have this many MyISAM tables:
[root@xxx /db/mydata]# find -L . -name “*.MYI”|wc
13713 13713 558523

Number of possible needed fd’s: 13713 * 2 = 27426

I have this many InnoDB tables:
[root@xxx /db/mydata]# find -L . -name “*.ibd”|wc
31014 31014 1144412

Innodb is handled differently right? So 31014 table handles are always opened?

If I add the two, 31014 + 27426 = 58440 os file descriptors.

So –

1) Do I get errno: 24 becuase I have the open_files_limit set to 40000?
2) Could you explain the difference between MyISAM and InnoDB table cache?
3) Would converting the MyISAM tables to InnoDB fix or help the os ulimit max open tables issue?

Thanks for your help!

Server info:

Server version: 5.0.79-enterprise-log MySQL Enterprise Server (Commercial)

my.cnf:

[mysqld]
server-id = 2
port = 3306
basedir = /usr/local/mysql
datadir = /db/mydata
tmpdir = /db-logs/ibdlogs/dbtmp
pid-file = /db/mysql.pid
log-error = /db/mysql.err
log-slow-queries = /db/slow-queries.log
log-bin = /db-logs/bin-logs/mysql-logs/mysql-bin
expire_logs_days = 30 # Expire logs after in X days.
long_query_time = 2
thread_cache_size = 100
max_connections = 400
table_cache = 20000
open_files_limit = 40000
max_binlog_size = 500M
max_allowed_packet = 20M
tmp_table_size = 100M
max_heap_table_size = 100M
query_cache_size = 100M
default_table_type = MYISAM
log_long_format

# Innodb
innodb_status_file = 1
innodb_file_per_table = 1
innodb_data_home_dir = /db2/innodb
innodb_log_group_home_dir = /db-logs/ibdlogs
innodb_data_file_path = ibdata1:10M:autoextend
innodb_buffer_pool_size = 30G
innodb_log_buffer_size = 8M
innodb_additional_mem_pool_size = 8M
innodb_log_file_size = 256M
innodb_log_files_in_group = 2
innodb_thread_concurrency = 16
innodb_open_files = 20000
innodb_flush_method = O_DIRECT
innodb_flush_log_at_trx_commit = 2
transaction_isolation = READ-COMMITTED

By: DaveJ

DaveJ — Wed, 14 Apr 2010 13:44:40 +0000

Great Post, Thank you all for your info.

Question:

After kicking off mk-table-checksum in the middle of the night I start getting errors such as:

[root@wcdb /db]# perror 24
OS error code 24: Too many open files

I have this many MyISAM tables:

[root@xxx /db/mydata]# find -L . -name “*.MYI”|wc
13713 13713 558523

Number of possible needed fd’s: 13713 * 2 = 27426

I have this many InnoDB tables:

[root@xxx /db/mydata]# find -L . -name “*.ibd”|wc
31014 31014 1144412

Innodb is handled differently right? So 31014 table handles are always opened and would not be the cause of the errno: 24?

If I add the two, 31014 + 27426 = 58440 os file descriptors.

So –

1) Do I get errno: 24 becuase I have the open_files_limit set to 40000?
2) Could you explain the difference between MyISAM and InnoDB and the difference in how the OS handles opened files?
3) Would converting the MyISAM tables to InnoDB fix or help the os ulimit max open tables issue?

Thanks for your help!

Server info:

Server version: 5.0.79-enterprise-log MySQL Enterprise Server (Commercial)

my.cnf:

By: Kostja

Kostja — Thu, 10 Dec 2009 22:24:33 +0000

And as you can see from the analysis in the bug report, this is not a table cache problem, but a MyISAM one.

By: Mark Callaghan

Mark Callaghan — Sat, 28 Nov 2009 18:02:37 +0000

@Sergei – I want it fixed and I shared that with MySQL.

@Peter – whether or not Innodb keeps a file descriptor open for the tablespace, index cardinality stats (~8 leaf blocks per index) are sampled when MySQL creates a table handler instance for an InnoDB table. That is the cost I mentioned above.

By: Sergei Golubchik

Sergei Golubchik — Sat, 28 Nov 2009 08:19:19 +0000

I see, you didn’t bother to submit a bug report about it.

Am I the only reader of your blog that want this fixed ?
http://bugs.mysql.com/49177

By: Igor

Igor — Mon, 23 Nov 2009 22:38:48 +0000

I think table_cache values between 1,000 and 5,000 are of much higher practical relevance for vast majority of MySQL users (those that are not hosting providers etc).

By: Wagner Bianchi

Wagner Bianchi — Sun, 22 Nov 2009 17:33:38 +0000

Maybe this number is tiny compared with the number up on the test did by Peter. That was 100.000.

Regards.

By: Chris

Chris — Sun, 22 Nov 2009 16:51:29 +0000

I don’t think 853 tables is tiny. Why so many? Are some of them duplicates providing duplicate funtionality to different groups of users? Why not combine like data in a single table for all users? Or am I missing something?