pt-visual-explain - Format EXPLAIN output as a tree.
pt-visual-explain [OPTION...] [FILE...]
pt-visual-explain transforms EXPLAIN output into a tree representation of the query plan. If FILE is given, input is read from the file(s). With no FILE, or when FILE is -, read standard input.
pt-visual-explain <file_containing_explain_output> pt-visual-explain -c <file_containing_query> mysql -e "explain select * from mysql.user" | pt-visual-explain
The following section is included to inform users about the potential risks, whether known or unknown, of using this tool. The two main categories of risks are those created by the nature of the tool (e.g. read-only tools vs. read-write tools) and those created by bugs.
pt-visual-explain is read-only and very low-risk.
At the time of this release, we know of no bugs that could cause serious harm to users.
The authoritative source for updated information is always the online issue tracking system. Issues that affect this tool will be marked as such. You can see a list of such issues at the following URL: http://www.percona.com/bugs/pt-visual-explain.
See also “BUGS” for more information on filing bugs and getting help.
pt-visual-explain reverse-engineers MySQL’s EXPLAIN output into a query execution plan, which it then formats as a left-deep tree – the same way the plan is represented inside MySQL. It is possible to do this by hand, or to read EXPLAIN’s output directly, but it requires patience and expertise. Many people find a tree representation more understandable.
You can pipe input into pt-visual-explain or specify a filename at the command line, including the magical ‘-‘ filename, which will read from standard input. It can do two things with the input: parse it for something that looks like EXPLAIN output, or connect to a MySQL instance and run EXPLAIN on the input.
When parsing its input, pt-visual-explain understands three formats: tabular like that shown in the mysql command-line client, vertical like that created by using the G line terminator in the mysql command-line client, and tab separated. It ignores any lines it doesn’t know how to parse.
When executing the input, pt-visual-explain replaces everything in the input up to the first SELECT keyword with ‘EXPLAIN SELECT,’ and then executes the result. You must specify --connect to execute the input as a query.
Either way, it builds a tree from the result set and prints it to standard output. For the following query,
select * from sakila.film_actor join sakila.film using(film_id);
pt-visual-explain generates this query plan:
JOIN +- Bookmark lookup | +- Table | | table film_actor | | possible_keys idx_fk_film_id | +- Index lookup | key film_actor->idx_fk_film_id | possible_keys idx_fk_film_id | key_len 2 | ref sakila.film.film_id | rows 2 +- Table scan rows 952 +- Table table film possible_keys PRIMARY
The query plan is left-deep, depth-first search, and the tree’s root is the output node – the last step in the execution plan. In other words, read it like this:
Table scan the ‘film’ table, which accesses an estimated 952 rows.
For each row, find matching rows by doing an index lookup into the film_actor->idx_fk_film_id index with the value from sakila.film.film_id, then a bookmark lookup into the film_actor table.
For more information on how to read EXPLAIN output, please see http://dev.mysql.com/doc/en/explain.html, and this talk titled “MySQL query optimizer internals and upcoming features in v. 5.2”: from Timour Katchaounov, one of the MySQL developers: http://goo.gl/VIWvo
This program is actually a runnable module, not just an ordinary Perl script. In fact, there are two modules embedded in it. This makes unit testing easy, but it also makes it easy for you to use the parsing and tree-building functionality if you want.
The ExplainParser package accepts a string and parses whatever it thinks looks like EXPLAIN output from it. The synopsis is as follows:
require "pt-visual-explain"; my $p = ExplainParser->new(); my $rows = $p->parse("some text"); # $rows is an arrayref of hashrefs.
The ExplainTree package accepts a set of rows and turns it into a tree. For convenience, you can also have it delegate to ExplainParser and parse text for you. Here’s the synopsis:
require "pt-visual-explain"; my $e = ExplainTree->new(); my $tree = $e->parse("some text", \%options); my $output = $e->pretty_print($tree); print $tree;
This section explains the algorithm that converts EXPLAIN into a tree. You may be interested in reading this if you want to understand EXPLAIN more fully, or trying to figure out how this works, but otherwise this section will probably not make your life richer.
The tree can be built by examining the id, select_type, and table columns of each row. Here’s what I know about them:
The id column is the sequential number of the select. This does not indicate nesting; it just comes from counting SELECT from the left of the SQL statement. It’s like capturing parentheses in a regular expression. A UNION RESULT row doesn’t have an id, because it isn’t a SELECT. The source code actually refers to UNIONs as a fake_lex, as I recall.
If two adjacent rows have the same id value, they are joined with the standard single-sweep multi-join method.
The select_type column tells a) that a new sub-scope has opened b) what kind of relationship the row has to the previous row c) what kind of operation the row represents.
Tables that are JOINed all have the same select_type. For example, if you JOIN three tables inside a dependent subquery, they’ll all say the same thing: DEPENDENT SUBQUERY.
The table column usually specifies the table name or alias, but may also say <derivedN> or <unionN,N...N>. If it says <derivedN>, the row represents an access to the temporary table that holds the result of the subquery whose id is N. If it says <unionN,..N> it’s the same thing, but it refers to the results it UNIONs together.
Finally, order matters. If a row’s id is less than the one before it, I think that means it is dependent on something other than the one before it. For example,
explain select (select 1 from sakila.film), (select 2 from sakila.film_actor), (select 3 from sakila.actor); | id | select_type | table | +----+-------------+------------+ | 1 | PRIMARY | NULL | | 4 | SUBQUERY | actor | | 3 | SUBQUERY | film_actor | | 2 | SUBQUERY | film |
If the results were in order 2-3-4, I think that would mean 3 is a subquery of 2, 4 is a subquery of 3. As it is, this means 4 is a subquery of the nearest previous recent row with a smaller id, which is 1. Likewise for 3 and 2.
This structure is hard to programmatically build into a tree for the same reason it’s hard to understand by inspection: there are both forward and backward references. <derivedN> is a forward reference to selectN, while <unionM,N> is a backward reference to selectM and selectN. That makes recursion and other tree-building algorithms hard to get right (NOTE: after implementation, I now see how it would be possible to deal with both forward and backward references, but I have no motivation to change something that works). Consider the following:
select * from ( select 1 from sakila.actor as actor_1 union select 1 from sakila.actor as actor_2 ) as der_1 union select * from ( select 1 from sakila.actor as actor_3 union all select 1 from sakila.actor as actor_4 ) as der_2; | id | select_type | table | +------+--------------+------------+ | 1 | PRIMARY | <derived2> | | 2 | DERIVED | actor_1 | | 3 | UNION | actor_2 | | NULL | UNION RESULT | <union2,3> | | 4 | UNION | <derived5> | | 5 | DERIVED | actor_3 | | 6 | UNION | actor_4 | | NULL | UNION RESULT | <union5,6> | | NULL | UNION RESULT | <union1,4> |
This would be a lot easier to work with if it looked like this (I’ve bracketed the id on rows I moved):
| id | select_type | table | +------+--------------+------------+ |  | UNION RESULT | <union1,4> | | 1 | PRIMARY | <derived2> | |  | UNION RESULT | <union2,3> | | 2 | DERIVED | actor_1 | | 3 | UNION | actor_2 | | 4 | UNION | <derived5> | |  | UNION RESULT | <union5,6> | | 5 | DERIVED | actor_3 | | 6 | UNION | actor_4 |
In fact, why not re-number all the ids, so the PRIMARY row becomes 2, and so on? That would make it even easier to read. Unfortunately that would also have the effect of destroying the meaning of the id column, which I think is important to preserve in the final tree. Also, though it makes it easier to read, it doesn’t make it easier to manipulate programmatically; so it’s fine to leave them numbered as they are.
The goal of re-ordering is to make it easier to figure out which rows are children of which rows in the execution plan. Given the reordered list and some row whose table is <union...> or <derived>, it is easy to find the beginning of the slice of rows that should be child nodes in the tree: you just look for the first row whose ID is the same as the first number in the table.
The next question is how to find the last row that should be a child node of a UNION or DERIVED. I’ll start with DERIVED, because the solution makes UNION easy.
Consider how MySQL numbers the SELECTs sequentially according to their position in the SQL, left-to-right. Since a DERIVED table encloses everything within it in a scope, which becomes a temporary table, there are only two things to think about: its child subqueries and unions (if any), and its next siblings in the scope that encloses it. Its children will all have an id greater than it does, by definition, so any later rows with a smaller id terminate the scope.
Here’s an example. The middle derived table here has a subquery and a UNION to make it a little more complex for the example.
explain select 1 from ( select film_id from sakila.film limit 1 ) as der_1 join ( select film_id, actor_id, (select count(*) from sakila.rental) as r from sakila.film_actor limit 1 union all select 1, 1, 1 from sakila.film_actor as dummy ) as der_2 using (film_id) join ( select actor_id from sakila.actor limit 1 ) as der_3 using (actor_id);
Here’s the output of EXPLAIN:
| id | select_type | table | | 1 | PRIMARY | <derived2> | | 1 | PRIMARY | <derived6> | | 1 | PRIMARY | <derived3> | | 6 | DERIVED | actor | | 3 | DERIVED | film_actor | | 4 | SUBQUERY | rental | | 5 | UNION | dummy | | NULL | UNION RESULT | <union3,5> | | 2 | DERIVED | film |
The siblings all have id 1, and the middle one I care about is derived3. (Notice MySQL doesn’t execute them in the order I defined them, which is fine). Now notice that MySQL prints out the rows in the opposite order I defined the subqueries: 6, 3, 2. It always seems to do this, and there might be other methods of finding the scope boundaries including looking for the lower boundary of the next largest sibling, but this is a good enough heuristic. I am forced to rely on it for non-DERIVED subqueries, so I rely on it here too. Therefore, I decide that everything greater than or equal to 3 belongs to the DERIVED scope.
The rule for UNION is simple: they consume the entire enclosing scope, and to find the component parts of each one, you find each part’s beginning as referred to in the <unionN,...> definition, and its end is either just before the next one, or if it’s the last part, the end is the end of the scope.
This is only simple because UNION consumes the entire scope, which is either the entire statement, or the scope of a DERIVED table. This is because a UNION cannot be a sibling of another UNION or a table, DERIVED or not. (Try writing such a statement if you don’t see it intuitively). Therefore, you can just find the enclosing scope’s boundaries, and the rest is easy. Notice in the example above, the UNION is over <union3,5>, which includes the row with id 4 – it includes every row between 3 and 5.
Finally, there are non-derived subqueries to deal with as well. In this case I can’t look at siblings to find the end of the scope as I did for DERIVED. I have to trust that MySQL executes depth-first. Here’s an example:
explain select actor_id, ( select count(film_id) + (select count(*) from sakila.film) from sakila.film join sakila.film_actor using(film_id) where exists( select * from sakila.actor where sakila.actor.actor_id = sakila.film_actor.actor_id ) ) from sakila.actor; | id | select_type | table | | 1 | PRIMARY | actor | | 2 | SUBQUERY | film | | 2 | SUBQUERY | film_actor | | 4 | DEPENDENT SUBQUERY | actor | | 3 | SUBQUERY | film |
In order, the tree should be built like this:
But the only reason the nested subquery didn’t include select 3 is because select 4 came first. In other words, if EXPLAIN looked like this,
| id | select_type | table | | 1 | PRIMARY | actor | | 2 | SUBQUERY | film | | 2 | SUBQUERY | film_actor | | 3 | SUBQUERY | film | | 4 | DEPENDENT SUBQUERY | actor |
I would be forced to assume upon seeing select 3 that select 4 is a subquery of it, rather than just being the next sibling in the enclosing scope. If this is ever wrong, then the algorithm is wrong, and I don’t see what could be done about it.
UNION is a little more complicated than just “the entire scope is a UNION,” because the UNION might itself be inside an enclosing scope that’s only indicated by the first item inside the UNION. There are only three kinds of enclosing scopes: UNION, DERIVED, and SUBQUERY. A UNION can’t enclose a UNION, and a DERIVED has its own “scope markers,” but a SUBQUERY can wholly enclose a UNION, like this strange example on the empty table t1:
explain select * from t1 where not exists( (select t11.i from t1 t11) union (select t12.i from t1 t12)); | id | select_type | table | Extra | +------+--------------+------------+--------------------------------+ | 1 | PRIMARY | t1 | const row not found | | 2 | SUBQUERY | NULL | No tables used | | 3 | SUBQUERY | NULL | no matching row in const table | | 4 | UNION | t12 | const row not found | | NULL | UNION RESULT | <union2,4> | |
The UNION’s backward references might make it look like the UNION encloses the subquery, but studying the query makes it clear this isn’t the case. So when a UNION’s first row says SUBQUERY, it is this special case.
By the way, I don’t fully understand this query plan; there are 4 numbered SELECT in the plan, but only 3 in the query. The parens around the UNIONs are meaningful. Removing them will make the EXPLAIN different. Please tell me how and why this works if you know.
Armed with this knowledge, it’s possible to use recursion to turn the parent-child relationship between all the rows into a tree representing the execution plan.
MySQL prints the rows in execution order, even the forward and backward references. At any given scope, the rows are processed as a left-deep tree. MySQL does not do “bushy” execution plans. It begins with a table, finds a matching row in the next table, and continues till the last table, when it emits a row. When it runs out, it backtracks till it can find the next row and repeats. There are subtleties of course, but this is the basic plan. This is why MySQL transforms all RIGHT OUTER JOINs into LEFT OUTER JOINs and cannot do FULL OUTER JOIN.
This means in any given scope, say
| id | select_type | table | | 1 | SIMPLE | tbl1 | | 1 | SIMPLE | tbl2 | | 1 | SIMPLE | tbl3 |
The execution plan looks like a depth-first traversal of this tree:
JOIN / \ JOIN tbl3 / \ tbl1 tbl2
The JOIN might not be a JOIN. It might be a subquery, for example. This comes from the type column of EXPLAIN. The documentation says this is a “join type,” but I think “access type” is more accurate, because it’s “how MySQL accesses rows.”
pt-visual-explain decorates the tree significantly more than just turning rows into nodes. Each node may get a series of transformations that turn it into a subtree of more than one node. For example, an index scan not marked with ‘Using index’ must do a bookmark lookup into the table rows; that is a three-node subtree. However, after the above node-ordering and scoping stuff, the rest of the process is pretty simple.
This tool accepts additional command-line arguments. Refer to the “SYNOPSIS” and usage information for details.
Prompt for a password when connecting to MySQL.
short form: -A; type: string
Default character set. If the value is utf8, sets Perl’s binmode on STDOUT to utf8, passes the mysql_enable_utf8 option to DBD::mysql, and runs SET NAMES UTF8 after connecting to MySQL. Any other value sets binmode on STDOUT without the utf8 layer, and runs SET NAMES after connecting to MySQL.
Assume that PRIMARY KEY index accesses don’t need to do a bookmark lookup to retrieve rows. This is the case for InnoDB.
Read this comma-separated list of config files; if specified, this must be the first option on the command line.
Treat input as a query, and obtain EXPLAIN output by connecting to a MySQL instance and running EXPLAIN on the query. When this option is given, pt-visual-explain uses the other connection-specific options such as --user to connect to the MySQL instance. If you have a .my.cnf file, it will read it, so you may not need to specify any connection-specific options.
short form: -D; type: string
Connect to this database.
short form: -F; type: string
Only read mysql options from the given file. You must give an absolute pathname.
type: string; default: tree
Set output format.
The default is a terse pretty-printed tree. The valid values are:
Value Meaning ===== ================================================ tree Pretty-printed terse tree. dump Data::Dumper output (see Data::Dumper for more).
Show help and exit.
short form: -h; type: string
Connect to host.
short form: -p; type: string
Password to use when connecting.
Create the given PID file. The file contains the process ID of the script. The PID file is removed when the script exits. Before starting, the script checks if the PID file already exists. If it does not, then the script creates and writes its own PID to it. If it does, then the script checks the following: if the file contains a PID and a process is running with that PID, then the script dies; or, if there is no process running with that PID, then the script overwrites the file with its own PID and starts; else, if the file contains no PID, then the script dies.
short form: -P; type: int
Port number to use for connection.
type: string; default: wait_timeout=10000
Set these MySQL variables. Immediately after connecting to MySQL, this string will be appended to SET and executed.
short form: -S; type: string
Socket file to use for connection.
short form: -u; type: string
User for login if not current user.
Show version and exit.
These DSN options are used to create a DSN. Each option is given like option=value. The options are case-sensitive, so P and p are not the same option. There cannot be whitespace before or after the = and if the value contains whitespace it must be quoted. DSN options are comma-separated. See the percona-toolkit manpage for full details.
dsn: charset; copy: yes
Default character set.
dsn: database; copy: yes
dsn: mysql_read_default_file; copy: yes
Only read default options from the given file
dsn: host; copy: yes
Connect to host.
dsn: password; copy: yes
Password to use when connecting.
dsn: port; copy: yes
Port number to use for connection.
dsn: mysql_socket; copy: yes
Socket file to use for connection.
dsn: user; copy: yes
User for login if not current user.
The environment variable PTDEBUG enables verbose debugging output to STDERR. To enable debugging and capture all output to a file, run the tool like:
PTDEBUG=1 pt-visual-explain ... > FILE 2>&1
Be careful: debugging output is voluminous and can generate several megabytes of output.
You need Perl, DBI, DBD::mysql, and some core packages that ought to be installed in any reasonably new version of Perl.
For a list of known bugs, see http://www.percona.com/bugs/pt-visual-explain.
Please report bugs at https://bugs.launchpad.net/percona-toolkit. Include the following information in your bug report:
If possible, include debugging output by running the tool with PTDEBUG; see “ENVIRONMENT”.
Visit http://www.percona.com/software/percona-toolkit/ to download the latest release of Percona Toolkit. Or, get the latest release from the command line:
wget percona.com/get/percona-toolkit.tar.gz wget percona.com/get/percona-toolkit.rpm wget percona.com/get/percona-toolkit.deb
You can also get individual tools from the latest release:
Replace TOOL with the name of any tool.
This tool is part of Percona Toolkit, a collection of advanced command-line tools developed by Percona for MySQL support and consulting. Percona Toolkit was forked from two projects in June, 2011: Maatkit and Aspersa. Those projects were created by Baron Schwartz and developed primarily by him and Daniel Nichter, both of whom are employed by Percona. Visit http://www.percona.com/software/ for more software developed by Percona.
This program is copyright 2011-2013 Percona Ireland Ltd, 2007-2011 Baron Schwartz.
THIS PROGRAM IS PROVIDED “AS IS” AND WITHOUT ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, WITHOUT LIMITATION, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, version 2; OR the Perl Artistic License. On UNIX and similar systems, you can issue `man perlgpl’ or `man perlartistic’ to read these licenses.
You should have received a copy of the GNU General Public License along with this program; if not, write to the Free Software Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA.