MySQL binlogs are the foundation for replication, but them can be useful for tracking intense write operations on database when using binlog_format=ROW. With that we can see all the changes that are actually applied to the database. For example: a statement like delete from table where timestamp > ? could affect just 3 or 1 million rows. Besides that, sometimes we have tons of very fast statements, but when composed together in a transaction it can takes lots of time.

This application aims to aid in tracking which transactions are being too write intensves.

pip install mysql-binlog-explorer
mysql-binlog-explorer ~/logs/mysql-bin-changelog.411078 --tenant-identifier company_id  --schema-ddl schema/my_db.ddl

We have two not required parameters (but they are quite useful).

• schema-ddl is a file with the DDL instructions to create the database. It'll be use used to show the column names in the statements.
• tenant-identifier the name of a column which is used to store the tenant id (usually a column name that repeats across every table). This will aid in the generated charts.

• Enable binlog (configuration varies depending on environment)
• Enable row format for binlog: binlog_format=ROW
• This kind o change needs to restart the server so be careful

Given that you already have a local mysql installation with brew, doing the above steps would mean:

mysql.server start --log-bin=binlog --binlog-format=ROW

Now you need the actual logs. In a MySQL session do the following:

SHOW BINARY LOGS; -- get the name of the binary log that you want to check

Now download it:

mysqlbinlog -h <HOST> -u <USER> -p<PASSWORD> --read-from-remote-server --base64-output=decode-rows -vv <NAME_FROM_STATEMENT_ABOVE> > my-bin-log.txt

Don't forget the --base64-output=decode-rows -vv, it's mandatory for the parser to work!

Now just use it.

mysql-binlog-explorer my-bin-log.txt

When the tenant information is used, the app can plot a chart with transactions/changes distribution per tenant identifier.

• Tested only with MySQL 5.6 binlogs, other versions may not work as expected.
• The result of the parsing is stored entirely in-memory, so it may crash for very large file sets.