TESTS

Dates and times are all measured relative to the same instant, when pt-find first asks the database server what time it is. All date and time manipulation is done in SQL, so if you say to find tables modified 5 days ago, that translates to SELECT DATE_SUB(CURRENT_TIMESTAMP, INTERVAL 5 DAY). If you specify "--day-start", if course it's relative to CURRENT_DATE instead.

However, table sizes and other metrics are not consistent at an instant in time. It can take some time for MySQL to process all the SHOW queries, and pt-find can't do anything about that. These measurements are as of the time they're taken.

If you need some test that's not in this list, file a bug report and I'll enhance pt-find for you. It's really easy.

--autoinc

type: string; group: Tests

Table's next AUTO_INCREMENT is n. This tests the Auto_increment column.

--avgrowlen

type: size; group: Tests

Table avg row len is n bytes. This tests the Avg_row_length column. The specified size can be "NULL" to test where Avg_row_length IS NULL.

--checksum

type: string; group: Tests

Table checksum is n. This tests the Checksum column.

--cmin

type: size; group: Tests

Table was created n minutes ago. This tests the Create_time column.

--collation

type: string; group: Tests

Table collation matches pattern. This tests the Collation column.

--column-name

type: string; group: Tests

A column name in the table matches pattern.

--column-type

type: string; group: Tests

A column in the table matches this type (case-insensitive).

Examples of types are: varchar, char, int, smallint, bigint, decimal, year, timestamp, text, enum.

--comment

type: string; group: Tests

Table comment matches pattern. This tests the Comment column.

--connection-id

type: string; group: Tests

Table name has nonexistent MySQL connection ID. This tests the table name for a pattern. The argument to this test must be a Perl regular expression that captures digits like this: (\d+). If the table name matches the pattern, these captured digits are taken to be the MySQL connection ID of some process. If the connection doesn't exist according to SHOW FULL PROCESSLIST, the test returns true. If the connection ID is greater than pt-find's own connection ID, the test returns false for safety.

Why would you want to do this? If you use MySQL statement-based replication, you probably know the trouble temporary tables can cause. You might choose to work around this by creating real tables with unique names, instead of temporary tables. One way to do this is to append your connection ID to the end of the table, thusly: scratch_table_12345. This assures the table name is unique and lets you have a way to find which connection it was associated with. And perhaps most importantly, if the connection no longer exists, you can assume the connection died without cleaning up its tables, and this table is a candidate for removal.

This is how I manage scratch tables, and that's why I included this test in pt-find.

The argument I use to "--connection-id" is "\D_(\d+)$". That finds tables with a series of numbers at the end, preceded by an underscore and some non-number character (the latter criterion prevents me from examining tables with a date at the end, which people tend to do: baron_scratch_2007_05_07 for example). It's better to keep the scratch tables separate of course.

If you do this, make sure the user pt-find runs as has the PROCESS privilege! Otherwise it will only see connections from the same user, and might think some tables are ready to remove when they're still in use. For safety, pt-find checks this for you.

See also "--server-id".

--createopts

type: string; group: Tests

Table create option matches pattern. This tests the Create_options column.

--ctime

type: size; group: Tests

Table was created n days ago. This tests the Create_time column.

--datafree

type: size; group: Tests

Table has n bytes of free space. This tests the Data_free column. The specified size can be "NULL" to test where Data_free IS NULL.

--datasize

type: size; group: Tests

Table data uses n bytes of space. This tests the Data_length column. The specified size can be "NULL" to test where Data_length IS NULL.

Note: Starting from MySQL 8.0, empty tables return 0 instead of NULL.

--dblike

type: string; group: Tests

Database name matches SQL LIKE pattern.

--dbregex

type: string; group: Tests

Database name matches this pattern.

--empty

group: Tests

Table has no rows. This tests the Rows column.

--engine

type: string; group: Tests

Table storage engine matches this pattern. This tests the Engine column, or in earlier versions of MySQL, the Type column.

--function

type: string; group: Tests

Function definition matches pattern.

--indexsize

type: size; group: Tests

Table indexes use n bytes of space. This tests the Index_length column. The specified size can be "NULL" to test where Index_length IS NULL.

--kmin

type: size; group: Tests

Table was checked n minutes ago. This tests the Check_time column.

--ktime

type: size; group: Tests

Table was checked n days ago. This tests the Check_time column.

--mmin

type: size; group: Tests

Table was last modified n minutes ago. This tests the Update_time column.

--mtime

type: size; group: Tests

Table was last modified n days ago. This tests the Update_time column.

--procedure

type: string; group: Tests

Procedure definition matches pattern.

--rowformat

type: string; group: Tests

Table row format matches pattern. This tests the Row_format column.

--rows

type: size; group: Tests

Table has n rows. This tests the Rows column. The specified size can be "NULL" to test where Rows IS NULL.

--server-id

type: string; group: Tests

Table name contains the server ID. If you create temporary tables with the naming convention explained in "--connection-id", but also add the server ID of the server on which the tables are created, then you can use this pattern match to ensure tables are dropped only on the server they're created on. This prevents a table from being accidentally dropped on a slave while it's in use (provided that your server IDs are all unique, which they should be for replication to work).

For example, on the master (server ID 22) you create a table called scratch_table_22_12345. If you see this table on the slave (server ID 23), you might think it can be dropped safely if there's no such connection 12345. But if you also force the name to match the server ID with "--server-id '\D_(\d+)_\d+$'", the table won't be dropped on the slave.

--tablesize

type: size; group: Tests

Table uses n bytes of space. This tests the sum of the Data_length and Index_length columns.

--tbllike

type: string; group: Tests

Table name matches SQL LIKE pattern.

--tblregex

type: string; group: Tests

Table name matches this pattern.

--tblversion

type: size; group: Tests

Table version is n. This tests the Version column.

--trigger

type: string; group: Tests

Trigger action statement matches pattern.

--trigger-table

type: string; group: Tests

"--trigger" is defined on table matching pattern.

--view

type: string; group: Tests

CREATE VIEW matches this pattern.

ACTIONS

--exec

type: string; group: Actions

Execute this SQL with each item found. The SQL can contain escapes and formatting directives (see "--printf").

--exec-dsn

type: string; group: Actions

Specify a DSN in key-value format to use when executing SQL with "--exec" and "--exec-plus". Any values not specified are inherited from command-line arguments.

--exec-plus

type: string; group: Actions

Execute this SQL with all items at once. This option is unlike "--exec". There are no escaping or formatting directives; there is only one special placeholder for the list of database and table names, %s. The list of tables found will be joined together with commas and substituted wherever you place %s.

You might use this, for example, to drop all the tables you found:

   DROP TABLE %s
    

This is sort of like GNU find's "-exec command {} +" syntax. Only it's not totally cryptic. And it doesn't require me to write a command-line parser.

--print

group: Actions

Print the database and table name, followed by a newline. This is the default action if no other action is specified.

--printf

type: string; group: Actions

Print format on the standard output, interpreting '\' escapes and '%' directives. Escapes are backslashed characters, like \n and \t. Perl interprets these, so you can use any escapes Perl knows about. Directives are replaced by %s, and as of this writing, you can't add any special formatting instructions, like field widths or alignment (though I'm musing over ways to do that).

Here is a list of the directives. Note that most of them simply come from columns of SHOW TABLE STATUS. If the column is NULL or doesn't exist, you get an empty string in the output. A % character followed by any character not in the following list is discarded (but the other character is printed).

   CHAR DATA SOURCE        NOTES
   ---- ------------------ ------------------------------------------
   a    Auto_increment
   A    Avg_row_length
   c    Checksum
   C    Create_time
   D    Database           The database name in which the table lives
   d    Data_length
   E    Engine             In older versions of MySQL, this is Type
   F    Data_free
   f    Innodb_free        Parsed from the Comment field
   I    Index_length
   K    Check_time
   L    Collation
   M    Max_data_length
   N    Name
   O    Comment
   P    Create_options
   R    Row_format
   S    Rows
   T    Table_length       Data_length+Index_length
   U    Update_time
   V    Version