MySQL consuming all system memory on INSERT ... SELECT

Question

The mysql daemon is getting killed because Linux is reaching out of memory:

Oct 24 07:41:23 <hostname> kernel: [82297.673701] Out of memory: kill process 13816 (mysqld) score 1839626 or a child 

There is a link with some workaround on this.

That only happen when executing a query INSERT ... SELECT with a very huge resulset.

MySQLTuner script displays that maximum theorical memory is less than 8GB, but top and munim shows that is getting over all RAM and swap available:

[--] Total buffers: 560.0M global + 72.2M per thread (100 max threads) [OK] Maximum possible memory usage: 7.6G (43% of installed RAM) 

I'm tried to tune some options with not results, there are the relevant ones:

skip-locking max_connections = 100 key_buffer_size = 512M max_allowed_packet = 32M table_open_cache = 2000 open_files_limit = 3000 sort_buffer_size = 16M read_buffer_size = 16M read_rnd_buffer_size = 8M myisam_sort_buffer_size = 64M thread_cache_size = 4 query_cache_size = 16M query_cache_limit = 2M thread_concurrency = 4 join_buffer_size = 32M tmp_table_size = 32M max_heap_table_size = 32M query_cache_limit = 8M bulk_insert_buffer_size = 64M myisam_max_sort_file_size = 50GB myisam_mmap_size = 10GB 

And there is a system resume:

• OS: Linux Debian "Squeeze" 6.0.8 (upgraded yesterday)
• RAM: 18GB
• Swap: 18GB
• MySQL: 5.1.72-2 (official Debain release)

At this moment, update or change OS or MySQL version is not possible, there is any option that can help and i missed?

Sorry by my english, and thank you in advance!

Edit: I'm only using MyISAM tables, and cannot change to InnoDB.

Answer 1

If no upgrade is availible, add more swap :)

Answer 2

Finally I realized that the problem was on joining with a federated table (sorry, it's a really complex query and I found this while analyzing the SQL).

The federated table has only about 7000 rows but executing the same query with a native MyISAM table with the same data, runs much faster and consumes a normal amount of RAM. Seems that MySQL executes a SELECT * in the remote server for every local table row joined and the memory used for caching is not released until finish.

At least I learned that the federated tables should be used carefully and avoid using them on complex querys or big resultsets.

Thanks to all who have responded, I hope this can be usefull to anybody with the same problem!