Handling Large Transactions¶
Length: 443 words; Published: June 24, 2015; Updated: October 22, 2019; Category: Performance; Type: Best Practices
Large transactions, especially ones deleting removes millions of rows from a table at once, can lead to diminished performance. One reason is that the table may reindexed and rescanned after each row is deleted.
Scenario
Suppose you have a node called dbhost with a database called keystone. Suppose further that you execute a large transaction, which includes a DELETE statement that deletes expired tokens from their table in that database and on that host. If this transaction involves millions of rows, it could affect the overall performance of the cluster.
Recommendations
This problem might be easily resolved by changing the size of the InnoDB buffer pool. The pool is bytes of the memory area where InnoDB caches table and index data. The larger the pool (that is, the more RAM is used), the less the disk is accessed, which is especially important when dealing with the same data in tables multiple times as you might in a large transaction on the same table.
To change the buffer pool size, check the value of the innodb_buffer_pool_size variable. If your servers are dedicated only to database service, try setting it to 80% of the server’s physical memory size. You can use the free command to see how much memory you have. Once you determine how much memory you can spare for the InnoDB pool, add or change a line in the server’s configuration file like the following:
innodb_buffer_pool_size=128M
If you must frequently perform extremely large transactions including DELETE statements, you might consider using pt-archiver from the Percona Toolkit. It is very efficient at deleting millions of rows without reading them or reindexing after each row is deleted.
To use pt-archiver, you will have to install the Percona Toolkit. Once that’s done, you would enter something like the following from the command-line to delete rows from tables (that is, keystone.token) based on a WHERE clause (that is, datetime column expires with values before now):
$ pt-archiver --source h=dbhost,D=keystone,t=token \
--purge --where "expires < NOW()" --primary-key-only \
--sleep-coef 1.0 --txn-size 500
This allows you to delete rows efficiently from the cluster.
The --source parameter provides the host, database, and table. Since there is no --destination parameter given, it won’t move the data to another table for archiving, per the primary function of pt-archiver. The --purge parameter instructs pt-archiver to remove the rows from the database. The --where parameter provides the WHERE clause of the DELETE statement.
The --primary-key-only parameter is efficient when purging rows. It prevents fetching each row in its entirety, when only the primary key column is used in the WHERE clause for DELETE statements.
Related Documents
