Skip to content



pt-query-digest - Analyze MySQL queries from logs, processlist, and tcpdump.



pt-query-digest [OPTIONS] [FILES] [DSN]

pt-query-digest analyzes MySQL queries from slow, general, and binary log files. It can also analyze queries from SHOW PROCESSLIST and MySQL protocol data from tcpdump. By default, queries are grouped by fingerprint and reported in descending order of query time (i.e. the slowest queries first). If no FILES are given, the tool reads STDIN. The optional DSN is used for certain options like --since and --until.

Report the slowest queries from slow.log:

pt-query-digest slow.log

Report the slowest queries from the processlist on host1:

pt-query-digest --processlist h=host1

Capture MySQL protocol data with tcpdump, then report the slowest queries:

tcpdump -s 65535 -x -nn -q -tttt -i any -c 1000 port 3306 > mysql.tcp.txt

pt-query-digest --type tcpdump mysql.tcp.txt

Save query data from slow.log to host2 for later review and trend analysis:

pt-query-digest --review h=host2 --no-report slow.log


Percona Toolkit is mature, proven in the real world, and well tested, but all database tools can pose a risk to the system and the database server. Before using this tool, please:

  • Read the tool’s documentation

  • Review the tool’s known “BUGS”

  • Test the tool on a non-production server

  • Backup your production server and verify the backups


pt-query-digest is a sophisticated but easy to use tool for analyzing MySQL queries. It can analyze queries from MySQL slow, general, and binary logs. (Binary logs must first be converted to text, see --type). It can also use SHOW PROCESSLIST and MySQL protocol data from tcpdump. By default, the tool reports which queries are the slowest, and therefore the most important to optimize. More complex and custom-tailored reports can be created by using options like --group-by, --filter, and --embedded-attributes.

Query analysis is a best-practice that should be done frequently. To make this easier, pt-query-digest has two features: query review (--review) and query history (--history). When the --review option is used, all unique queries are saved to a database. When the tool is ran again with --review, queries marked as reviewed in the database are not printed in the report. This highlights new queries that need to be reviewed. When the --history option is used, query metrics (query time, lock time, etc.) for each unique query are saved to database. Each time the tool is ran with --history, the more historical data is saved which can be used to trend and analyze query performance over time.


pt-query-digest works on events, which are a collection of key-value pairs called attributes. You’ll recognize most of the attributes right away: Query_time, Lock_time, and so on. You can just look at a slow log and see them. However, there are some that don’t exist in the slow log, and slow logs may actually include different kinds of attributes (for example, you may have a server with the Percona patches).

See “ATTRIBUTES REFERENCE” near the end of this documentation for a list of common and --type specific attributes. A familiarity with these attributes is necessary for working with --filter, --ignore-attributes, and other attribute-related options.

With creative use of --filter, you can create new attributes derived from existing attributes. For example, to create an attribute called Row_ratio for examining the ratio of Rows_sent to Rows_examined, specify a filter like:

--filter '($event->{Row_ratio} = $event->{Rows_sent} / ($event->{Rows_examined})) && 1'

The && 1 trick is needed to create a valid one-line syntax that is always true, even if the assignment happens to evaluate false. The new attribute will automatically appears in the output:

# Row ratio        1.00    0.00      1    0.50      1    0.71    0.50

Attributes created this way can be specified for --order-by or any option that requires an attribute.


The default --output is a query analysis report. The --[no]report option controls whether or not this report is printed. Sometimes you may want to parse all the queries but suppress the report, for example when using --review or --history.

There is one paragraph for each class of query analyzed. A “class” of queries all have the same value for the --group-by attribute which is fingerprint by default. (See “ATTRIBUTES”.) A fingerprint is an abstracted version of the query text with literals removed, whitespace collapsed, and so forth. The report is formatted so it’s easy to paste into emails without wrapping, and all non-query lines begin with a comment, so you can save it to a .sql file and open it in your favorite syntax-highlighting text editor. There is a response-time profile at the beginning.

The output described here is controlled by --report-format. That option allows you to specify what to print and in what order. The default output in the default order is described here.

The report, by default, begins with a paragraph about the entire analysis run The information is very similar to what you’ll see for each class of queries in the log, but it doesn’t have some information that would be too expensive to keep globally for the analysis. It also has some statistics about the code’s execution itself, such as the CPU and memory usage, the local date and time of the run, and a list of input files read/parsed.

Following this is the response-time profile over the events. This is a highly summarized view of the unique events in the detailed query report that follows. It contains the following columns:

Column        Meaning
============  ==========================================================
Rank          The query's rank within the entire set of queries analyzed
Query ID      The query's fingerprint
Response time The total response time, and percentage of overall total
Calls         The number of times this query was executed
R/Call        The mean response time per execution
V/M           The Variance-to-mean ratio of response time
Item          The distilled query

A final line whose rank is shown as MISC contains aggregate statistics on the queries that were not included in the report, due to options such as --limit and --outliers. For details on the variance-to-mean ratio, please see

Next, the detailed query report is printed. Each query appears in a paragraph. Here is a sample, slightly reformatted so ‘perldoc’ will not wrap lines in a terminal. The following will all be one paragraph, but we’ll break it up for commentary.

# Query 2: 0.01 QPS, 0.02x conc, ID 0xFDEA8D2993C9CAF3 at byte 160665

This line identifies the sequential number of the query in the sort order specified by --order-by. Then there’s the queries per second, and the approximate concurrency for this query (calculated as a function of the timespan and total Query_time). Next there’s a query ID. This ID is a hex version of the query’s checksum in the database, if you’re using --review. You can select the reviewed query’s details from the database with a query like SELECT .... WHERE checksum=0xFDEA8D2993C9CAF3.

If you are investigating the report and want to print out every sample of a particular query, then the following --filter may be helpful:

pt-query-digest slow.log           \
   --no-report                     \
   --output slowlog                \
   --filter '$event->{fingerprint} \
        && make_checksum($event->{fingerprint}) eq "FDEA8D2993C9CAF3"'

Notice that you must remove the 0x prefix from the checksum.

Finally, in case you want to find a sample of the query in the log file, there’s the byte offset where you can look. (This is not always accurate, due to some anomalies in the slow log format, but it’s usually right.) The position refers to the worst sample, which we’ll see more about below.

Next is the table of metrics about this class of queries.

#           pct   total    min    max     avg     95%  stddev  median
# Count       0       2
# Exec time  13   1105s   552s   554s    553s    554s      2s    553s
# Lock time   0   216us   99us  117us   108us   117us    12us   108us
# Rows sent  20   6.26M  3.13M  3.13M   3.13M   3.13M   12.73   3.13M
# Rows exam   0   6.26M  3.13M  3.13M   3.13M   3.13M   12.73   3.13M

The first line is column headers for the table. The percentage is the percent of the total for the whole analysis run, and the total is the actual value of the specified metric. For example, in this case we can see that the query executed 2 times, which is 13% of the total number of queries in the file. The min, max and avg columns are self-explanatory. The 95% column shows the 95th percentile; 95% of the values are less than or equal to this value. The standard deviation shows you how tightly grouped the values are. The standard deviation and median are both calculated from the 95th percentile, discarding the extremely large values.

The stddev, median and 95th percentile statistics are approximate. Exact statistics require keeping every value seen, sorting, and doing some calculations on them. This uses a lot of memory. To avoid this, we keep 1000 buckets, each of them 5% bigger than the one before, ranging from .000001 up to a very big number. When we see a value we increment the bucket into which it falls. Thus we have fixed memory per class of queries. The drawback is the imprecision, which typically falls in the 5 percent range.

Next we have statistics on the users, databases and time range for the query.

# Users       1   user1
# Databases   2     db1(1), db2(1)
# Time range 2008-11-26 04:55:18 to 2008-11-27 00:15:15

The users and databases are shown as a count of distinct values, followed by the values. If there’s only one, it’s shown alone; if there are many, we show each of the most frequent ones, followed by the number of times it appears.

# Query_time distribution
#   1us
#  10us
# 100us
#   1ms
#  10ms  #####
# 100ms  ####################
#    1s  ##########
#  10s+

The execution times show a logarithmic chart of time clustering. Each query goes into one of the “buckets” and is counted up. The buckets are powers of ten. The first bucket is all values in the “single microsecond range” – that is, less than 10us. The second is “tens of microseconds,” which is from 10us up to (but not including) 100us; and so on. The charted attribute can be changed by specifying --report-histogram but is limited to time-based attributes.

# Tables
#    SHOW CREATE TABLE `table1`\G
SELECT * FROM table1\G

This section is a convenience: if you’re trying to optimize the queries you see in the slow log, you probably want to examine the table structure and size. These are copy-and-paste-ready commands to do that.

Finally, we see a sample of the queries in this class of query. This is not a random sample. It is the query that performed the worst, according to the sort order given by --order-by. You will normally see a commented # EXPLAIN line just before it, so you can copy-paste the query to examine its EXPLAIN plan. But for non-SELECT queries that isn’t possible to do, so the tool tries to transform the query into a roughly equivalent SELECT query, and adds that below.

If you want to find this sample event in the log, use the offset mentioned above, and something like the following:

tail -c +<offset> /path/to/file | head

See also --report-format.


A query --review is the process of storing all the query fingerprints analyzed. This has several benefits:

  • You can add metadata to classes of queries, such as marking them for follow-up, adding notes to queries, or marking them with an issue ID for your issue tracking system.

  • You can refer to the stored values on subsequent runs so you’ll know whether you’ve seen a query before. This can help you cut down on duplicated work.

  • You can store historical data such as the row count, query times, and generally anything you can see in the report.

To use this feature, you run pt-query-digest with the --review option. It will store the fingerprints and other information into the table you specify. Next time you run it with the same option, it will do the following:

  • It won’t show you queries you’ve already reviewed. A query is considered to be already reviewed if you’ve set a value for the reviewed_by column. (If you want to see queries you’ve already reviewed, use the --report-all option.)

  • Queries that you’ve reviewed, and don’t appear in the output, will cause gaps in the query number sequence in the first line of each paragraph. And the value you’ve specified for --limit will still be honored. So if you’ve reviewed all queries in the top 10 and you ask for the top 10, you won’t see anything in the output.

  • If you want to see the queries you’ve already reviewed, you can specify --report-all. Then you’ll see the normal analysis output, but you’ll also see the information from the review table, just below the execution time graph. For example,

    # Review information
    #      comments: really bad IN() subquery, fix soon!
    #    first_seen: 2008-12-01 11:48:57
    #   jira_ticket: 1933
    #     last_seen: 2008-12-18 11:49:07
    #      priority: high
    #   reviewed_by: xaprb
    #   reviewed_on: 2008-12-18 15:03:11

    This metadata is useful because, as you analyze your queries, you get your comments integrated right into the report.


A query fingerprint is the abstracted form of a query, which makes it possible to group similar queries together. Abstracting a query removes literal values, normalizes whitespace, and so on. For example, consider these two queries:

SELECT name, password FROM user WHERE id='12823';
select name,   password from user
   where id=5;

Both of those queries will fingerprint to

select name, password from user where id=?

Once the query’s fingerprint is known, we can then talk about a query as though it represents all similar queries.

What pt-query-digest does is analogous to a GROUP BY statement in SQL. (But note that “multiple columns” doesn’t define a multi-column grouping; it defines multiple reports!) If your command-line looks like this,

pt-query-digest               \
    --group-by fingerprint    \
    --order-by Query_time:sum \
    --limit 10                \

The corresponding pseudo-SQL looks like this:

SELECT WORST(query BY Query_time), SUM(Query_time), ...
FROM /path/to/slow.log

You can also use the value distill, which is a kind of super-fingerprint. See --group-by for more.

Query fingerprinting accommodates many special cases, which have proven necessary in the real world. For example, an IN list with 5 literals is really equivalent to one with 4 literals, so lists of literals are collapsed to a single one. If you find something that is not fingerprinted properly, please submit a bug report with a reproducible test case.

Here is a list of transformations during fingerprinting, which might not be exhaustive:

  • Group all SELECT queries from mysqldump together, even if they are against different tables. The same applies to all queries from pt-table-checksum.

  • Shorten multi-value INSERT statements to a single VALUES() list.

  • Strip comments.

  • Abstract the databases in USE statements, so all USE statements are grouped together.

  • Replace all literals, such as quoted strings. For efficiency, the code that replaces literal numbers is somewhat non-selective, and might replace some things as numbers when they really are not. Hexadecimal literals are also replaced. NULL is treated as a literal. Numbers embedded in identifiers are also replaced, so tables named similarly will be fingerprinted to the same values (e.g. users_2009 and users_2010 will fingerprint identically).

  • Collapse all whitespace into a single space.

  • Lowercase the entire query.

  • Replace all literals inside of IN() and VALUES() lists with a single placeholder, regardless of cardinality.

  • Collapse multiple identical UNION queries into a single one.


This tool accepts additional command-line arguments. Refer to the “SYNOPSIS” and usage information for details.


Prompt for a password when connecting to MySQL.


type: array; default: db|Schema

List of attribute|alias, etc.

Certain attributes have multiple names, like db and Schema. If an event does not have the primary attribute, pt-query-digest looks for an alias attribute. If it finds an alias, it creates the primary attribute with the alias attribute’s value and removes the alias attribute.

If the event has the primary attribute, all alias attributes are deleted.

This helps simplify event attributes so that, for example, there will not be report lines for both db and Schema.


type: int; default: 0

A sanity limit for attribute values.

This option deals with bugs in slow logging functionality that causes large values for attributes. If the attribute’s value is bigger than this, the last-seen value for that class of query is used instead. Disabled by default.


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.


type: Array

Read this comma-separated list of config files; if specified, this must be the first option on the command line.


default: yes

Continue parsing even if there is an error. The tool will not continue forever: it stops once any process causes 100 errors, in which case there is probably a bug in the tool or the input is invalid.


default: yes

Create the --history table if it does not exist.

This option causes the table specified by --history to be created with the default structure shown in the documentation for --history.


default: yes

Create the --review table if it does not exist.

This option causes the table specified by --review to be created with the default structure shown in the documentation for --review.


Fork to the background and detach from the shell. POSIX operating systems only.


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: array

Two Perl regex patterns to capture pseudo-attributes embedded in queries.

Embedded attributes might be special attribute-value pairs that you’ve hidden in comments. The first regex should match the entire set of attributes (in case there are multiple). The second regex should match and capture attribute-value pairs from the first regex.

For example, suppose your query looks like the following:

SELECT * from users -- file: /login.php, line: 493;

You might run pt-query-digest with the following option:

:program:`pt-query-digest` --embedded-attributes ' -- .*','(\w+): ([^\,]+)'

The first regular expression captures the whole comment:

" -- file: /login.php, line: 493;"

The second one splits it into attribute-value pairs and adds them to the event:

=========  ==========
file       /login.php
line       493

NOTE: All commas in the regex patterns must be escaped with otherwise the pattern will break.


type: array; default: 5,10

Explain items when there are more or fewer than expected.

Defines the number of items expected to be seen in the report given by --[no]report, as controlled by --limit and --outliers. If there are more or fewer items in the report, each one will explain why it was included.


type: DSN

Run EXPLAIN for the sample query with this DSN and print results.

This works only when --group-by includes fingerprint. It causes pt-query-digest to run EXPLAIN and include the output into the report. For safety, queries that appear to have a subquery that EXPLAIN will execute won’t be EXPLAINed. Those are typically “derived table” queries of the form

select ... from ( select .... ) der;

The EXPLAIN results are printed as a full vertical format in the event report, which appears at the end of each event report in vertical style (\G) just like MySQL prints it.


type: string

Discard events for which this Perl code doesn’t return true.

This option is a string of Perl code or a file containing Perl code that gets compiled into a subroutine with one argument: $event. This is a hashref. If the given value is a readable file, then pt-query-digest reads the entire file and uses its contents as the code. The file should not contain a shebang (#!/usr/bin/perl) line.

If the code returns true, the chain of callbacks continues; otherwise it ends. The code is the last statement in the subroutine other than return $event. The subroutine template is:

sub { $event = shift; filter && return $event; }

Filters given on the command line are wrapped inside parentheses like ( filter ). For complex, multi-line filters, you must put the code inside a file so it will not be wrapped inside parentheses. Either way, the filter must produce syntactically valid code given the template. For example, an if-else branch given on the command line would not be valid:

--filter 'if () { } else { }'  # WRONG

Since it’s given on the command line, the if-else branch would be wrapped inside parentheses which is not syntactically valid. So to accomplish something more complex like this would require putting the code in a file, for example filter.txt:

my $event_ok; if (...) { $event_ok=1; } else { $event_ok=0; } $event_ok

Then specify --filter filter.txt to read the code from filter.txt.

If the filter code won’t compile, pt-query-digest will die with an error. If the filter code does compile, an error may still occur at runtime if the code tries to do something wrong (like pattern match an undefined value). pt-query-digest does not provide any safeguards so code carefully!

An example filter that discards everything but SELECT statements:

--filter '$event->{arg} =~ m/^select/i'

This is compiled into a subroutine like the following:

sub { $event = shift; ( $event->{arg} =~ m/^select/i ) && return $event; }

It is permissible for the code to have side effects (to alter $event).

See “ATTRIBUTES REFERENCE” for a list of common and --type specific attributes.

Here are more examples of filter code:

Host/IP matches

–filter ‘($event->{host} || $event->{ip} || “”) =~ m/’

Sometimes MySQL logs the host where the IP is expected. Therefore, we check both.

User matches john

–filter ‘($event->{user} || “”) =~ m/john/’

More than 1 warning

–filter ‘($event->{Warning_count} || 0) > 1’

Query does full table scan or full join

–filter ‘(($event->{Full_scan} || “”) eq “Yes”) || (($event->{Full_join} || “”) eq “Yes”)’

Query was not served from query cache

–filter ‘($event->{QC_Hit} || “”) eq “No”’

Query is 1 MB or larger

–filter ‘$event->{bytes} >= 1_048_576’

Since --filter allows you to alter $event, you can use it to do other things, like create new attributes. See “ATTRIBUTES” for an example.


type: Array; default: fingerprint

Which attribute of the events to group by.

In general, you can group queries into classes based on any attribute of the query, such as user or db, which will by default show you which users and which databases get the most Query_time. The default attribute, fingerprint, groups similar, abstracted queries into classes; see below and see also “FINGERPRINTS”.

A report is printed for each --group-by value (unless --no-report is given). Therefore, --group-by user,db means “report on queries with the same user and report on queries with the same db”; it does not mean “report on queries with the same user and db.” See also “OUTPUT”.

Every value must have a corresponding value in the same position in --order-by. However, adding values to --group-by will automatically add values to --order-by, for your convenience.

There are several magical values that cause some extra data mining to happen before the grouping takes place:


This causes events to be fingerprinted to abstract queries into a canonical form, which is then used to group events together into a class. See “FINGERPRINTS” for more about fingerprinting.


This causes events to be inspected for what appear to be tables, and then aggregated by that. Note that a query that contains two or more tables will be counted as many times as there are tables; so a join against two tables will count the Query_time against both tables.


This is a sort of super-fingerprint that collapses queries down into a suggestion of what they do, such as INSERT SELECT table1 table2.


Show help and exit.


type: DSN

Save metrics for each query class in the given table. pt-query-digest saves query metrics (query time, lock time, etc.) to this table so you can see how query classes change over time.

The default table is percona_schema.query_history. Specify database (D) and table (t) DSN options to override the default. The database and table are automatically created unless --no-create-history-table is specified (see --[no]create-history-table).

pt-query-digest inspects the columns in the table. The table must have at least the following columns:

CREATE TABLE query_review_history (
  checksum     CHAR(32) NOT NULL,
  sample       TEXT NOT NULL

Any columns not mentioned above are inspected to see if they follow a certain naming convention. The column is special if the name ends with an underscore followed by any of these values:


If the column ends with one of those values, then the prefix is interpreted as the event attribute to store in that column, and the suffix is interpreted as the metric to be stored. For example, a column named Query_time_min will be used to store the minimum Query_time for the class of events.

The table should also have a primary key, but that is up to you, depending on how you want to store the historical data. We suggest adding ts_min and ts_max columns and making them part of the primary key along with the checksum. But you could also just add a ts_min column and make it a DATE type, so you’d get one row per class of queries per day.

The following table definition is used for --[no]create-history-table:

  checksum             CHAR(32) NOT NULL,
  sample               TEXT NOT NULL,
  ts_min               DATETIME,
  ts_max               DATETIME,
  ts_cnt               FLOAT,
  Query_time_sum       FLOAT,
  Query_time_min       FLOAT,
  Query_time_max       FLOAT,
  Query_time_pct_95    FLOAT,
  Query_time_stddev    FLOAT,
  Query_time_median    FLOAT,
  Lock_time_sum        FLOAT,
  Lock_time_min        FLOAT,
  Lock_time_max        FLOAT,
  Lock_time_pct_95     FLOAT,
  Lock_time_stddev     FLOAT,
  Lock_time_median     FLOAT,
  Rows_sent_sum        FLOAT,
  Rows_sent_min        FLOAT,
  Rows_sent_max        FLOAT,
  Rows_sent_pct_95     FLOAT,
  Rows_sent_stddev     FLOAT,
  Rows_sent_median     FLOAT,
  Rows_examined_sum    FLOAT,
  Rows_examined_min    FLOAT,
  Rows_examined_max    FLOAT,
  Rows_examined_pct_95 FLOAT,
  Rows_examined_stddev FLOAT,
  Rows_examined_median FLOAT,
  -- Percona extended slowlog attributes
  Rows_affected_sum             FLOAT,
  Rows_affected_min             FLOAT,
  Rows_affected_max             FLOAT,
  Rows_affected_pct_95          FLOAT,
  Rows_affected_stddev          FLOAT,
  Rows_affected_median          FLOAT,
  Rows_read_sum                 FLOAT,
  Rows_read_min                 FLOAT,
  Rows_read_max                 FLOAT,
  Rows_read_pct_95              FLOAT,
  Rows_read_stddev              FLOAT,
  Rows_read_median              FLOAT,
  Merge_passes_sum              FLOAT,
  Merge_passes_min              FLOAT,
  Merge_passes_max              FLOAT,
  Merge_passes_pct_95           FLOAT,
  Merge_passes_stddev           FLOAT,
  Merge_passes_median           FLOAT,
  InnoDB_IO_r_ops_min           FLOAT,
  InnoDB_IO_r_ops_max           FLOAT,
  InnoDB_IO_r_ops_pct_95        FLOAT,
  InnoDB_IO_r_ops_stddev        FLOAT,
  InnoDB_IO_r_ops_median        FLOAT,
  InnoDB_IO_r_bytes_min         FLOAT,
  InnoDB_IO_r_bytes_max         FLOAT,
  InnoDB_IO_r_bytes_pct_95      FLOAT,
  InnoDB_IO_r_bytes_stddev      FLOAT,
  InnoDB_IO_r_bytes_median      FLOAT,
  InnoDB_IO_r_wait_min          FLOAT,
  InnoDB_IO_r_wait_max          FLOAT,
  InnoDB_IO_r_wait_pct_95       FLOAT,
  InnoDB_IO_r_wait_stddev       FLOAT,
  InnoDB_IO_r_wait_median       FLOAT,
  InnoDB_rec_lock_wait_min      FLOAT,
  InnoDB_rec_lock_wait_max      FLOAT,
  InnoDB_rec_lock_wait_pct_95   FLOAT,
  InnoDB_rec_lock_wait_stddev   FLOAT,
  InnoDB_rec_lock_wait_median   FLOAT,
  InnoDB_queue_wait_min         FLOAT,
  InnoDB_queue_wait_max         FLOAT,
  InnoDB_queue_wait_pct_95      FLOAT,
  InnoDB_queue_wait_stddev      FLOAT,
  InnoDB_queue_wait_median      FLOAT,
  InnoDB_pages_distinct_min     FLOAT,
  InnoDB_pages_distinct_max     FLOAT,
  InnoDB_pages_distinct_pct_95  FLOAT,
  InnoDB_pages_distinct_stddev  FLOAT,
  InnoDB_pages_distinct_median  FLOAT,
  -- Boolean (Yes/No) attributes.  Only the cnt and sum are needed
  -- for these.  cnt is how many times is attribute was recorded,
  -- and sum is how many of those times the value was Yes.  So
  -- sum/cnt * 100 equals the percentage of recorded times that
  -- the value was Yes.
  QC_Hit_cnt          FLOAT,
  QC_Hit_sum          FLOAT,
  Full_scan_cnt       FLOAT,
  Full_scan_sum       FLOAT,
  Full_join_cnt       FLOAT,
  Full_join_sum       FLOAT,
  Tmp_table_cnt       FLOAT,
  Tmp_table_sum       FLOAT,
  Tmp_table_on_disk_cnt FLOAT,
  Tmp_table_on_disk_sum FLOAT,
  Filesort_cnt          FLOAT,
  Filesort_sum          FLOAT,
  Filesort_on_disk_cnt  FLOAT,
  Filesort_on_disk_sum  FLOAT,
  PRIMARY KEY(checksum, ts_min, ts_max)

Note that we store the count (cnt) for the ts attribute only; it will be redundant to store this for other attributes.

Starting from Percona Toolkit 3.0.11, the checksum function has been updated to use 32 chars in the MD5 sum. This causes the checksum field in the history table to have a different value than in the previous versions of the tool.


short form: -h; type: string

Connect to host.


type: array; default: arg, cmd, insert_id, ip, port, Thread_id, timestamp, exptime, flags, key, res, val, server_id, offset, end_log_pos, Xid

Do not aggregate these attributes. Some attributes are not query metrics but metadata which doesn’t need to be (or can’t be) aggregated.


type: array; default: db,ts

If missing, inherit these attributes from the last event that had them.

This option sets which attributes are inherited or carried forward to events which do not have them. For example, if one event has the db attribute equal to “foo”, but the next event doesn’t have the db attribute, then it inherits “foo” for its db attribute.


type: float; default: .1

How frequently to poll the processlist, in seconds.


type: int; default: 1

How many times to iterate through the collect-and-report cycle. If 0, iterate to infinity. Each iteration runs for --run-time amount of time. An iteration is usually determined by an amount of time and a report is printed when that amount of time elapses. With --run-time-mode interval, an interval is instead determined by the interval time you specify with --run-time. See --run-time and --run-time-mode for more information.


type: Array; default: 95%:20

Limit output to the given percentage or count.

If the argument is an integer, report only the top N worst queries. If the argument is an integer followed by the % sign, report that percentage of the worst queries. If the percentage is followed by a colon and another integer, report the top percentage or the number specified by that integer, whichever comes first.

The value is actually a comma-separated array of values, one for each item in --group-by. If you don’t specify a value for any of those items, the default is the top 95%.

See also --outliers.


type: string

Print all output to this file when daemonized.


type: int; default: 10

Trim host names in reports to this length. 0=Do not trim host names.


type: int; default: 74

Trim lines to this length. 0=Do not trim lines.


type: Array; default: Query_time:sum

Sort events by this attribute and aggregate function.

This is a comma-separated list of order-by expressions, one for each --group-by attribute. The default Query_time:sum is used for --group-by attributes without explicitly given --order-by attributes (that is, if you specify more --group-by attributes than corresponding --order-by attributes). The syntax is attribute:aggregate. See “ATTRIBUTES” for valid attributes. Valid aggregates are:

Aggregate Meaning
========= ============================
sum       Sum/total attribute value
min       Minimum attribute value
max       Maximum attribute value
cnt       Frequency/count of the query

For example, the default Query_time:sum means that queries in the query analysis report will be ordered (sorted) by their total query execution time (“Exec time”). Query_time:max orders the queries by their maximum query execution time, so the query with the single largest Query_time will be listed first. cnt refers more to the frequency of the query as a whole, how often it appears; “Count” is its corresponding line in the query analysis report. So any attribute and cnt should yield the same report wherein queries are sorted by the number of times they appear.

When parsing general logs (--type genlog), the default --order-by becomes Query_time:cnt. General logs do not report query times so only the cnt aggregate makes sense because all query times are zero.

If you specify an attribute that doesn’t exist in the events, then pt-query-digest falls back to the default Query_time:sum and prints a notice at the beginning of the report for each query class. You can create attributes with --filter and order by them; see “ATTRIBUTES” for an example.


type: array; default: Query_time:1:10

Report outliers by attribute:percentile:count.

The syntax of this option is a comma-separated list of colon-delimited strings. The first field is the attribute by which an outlier is defined. The second is a number that is compared to the attribute’s 95th percentile. The third is optional, and is compared to the attribute’s cnt aggregate. Queries that pass this specification are added to the report, regardless of any limits you specified in --limit.

For example, to report queries whose 95th percentile Query_time is at least 60 seconds and which are seen at least 5 times, use the following argument:

--outliers Query_time:60:5

You can specify an –outliers option for each value in --group-by.


type: string; default: report

How to format and print the query analysis results. Accepted values are:

=======        ===============================
report         Standard query analysis report
slowlog        MySQL slow log
json           JSON, one array per query class
json-anon      JSON without example queries
secure-slowlog JSON without example queries

The entire report output can be disabled by specifying --no-report (see --[no]report), and its sections can be disabled or rearranged by specifying --report-format.

json output was introduced in 2.2.1 and is still in development, so the data structure may change in future versions.


short form: -p; type: string

Password to use when connecting. If password contains commas they must be escaped with a backslash: “exam,ple”


type: string

Create the given PID file. The tool won’t start if the PID file already exists and the PID it contains is different than the current PID. However, if the PID file exists and the PID it contains is no longer running, the tool will overwrite the PID file with the current PID. The PID file is removed automatically when the tool exits.


short form: -P; type: int

Port number to use for connection.


Preserve numbers in database/table names when fingerprinting queries. The standard fingerprint method replaces numbers in db/tables names, making a query like ‘SELECT * FROM db1.table2’ to be fingerprinted as ‘SELECT * FROM db?.table?’. This option changes that behaviour and the fingerprint will become ‘SELECT * FROM db1.table2’.


type: DSN

Poll this DSN’s processlist for queries, with --interval sleep between.

If the connection fails, pt-query-digest tries to reopen it once per second.


type: array; default: time,30

Print progress reports to STDERR. The value is a comma-separated list with two parts. The first part can be percentage, time, or iterations; the second part specifies how often an update should be printed, in percentage, seconds, or number of iterations.


type: time; default: 0

Wait this long for an event from the input; 0 to wait forever.

This option sets the maximum time to wait for an event from the input. It applies to all types of input except --processlist. If an event is not received after the specified time, the script stops reading the input and prints its reports. If --iterations is 0 or greater than 1, the next iteration will begin, else the script will exit.

This option requires the Perl POSIX module.


default: yes

Print query analysis reports for each --group-by attribute. This is the standard slow log analysis functionality. See “OUTPUT” for the description of what this does and what the results look like.

If you don’t need a report (for example, when using --review or --history), it is best to specify --no-report because this allows the tool to skip some expensive operations.


Report all queries, even ones that have been reviewed. This only affects the report --output when using --review. Otherwise, all queries are always printed.


type: Array; default: rusage,date,hostname,files,header,profile,query_report,prepared

Print these sections of the query analysis report.

============ ======================================================
rusage       CPU times and memory usage reported by ps
date         Current local date and time
hostname     Hostname of machine on which :program:`pt-query-digest` was run
files        Input files read/parsed
header       Summary of the entire analysis run
profile      Compact table of queries for an overview of the report
query_report Detailed information about each unique query
prepared     Prepared statements

The sections are printed in the order specified. The rusage, date, files and header sections are grouped together if specified together; other sections are separated by blank lines.

See “OUTPUT” for more information on the various parts of the query report.


type: string; default: Query_time

Chart the distribution of this attribute’s values.

The distribution chart is limited to time-based attributes, so charting Rows_examined, for example, will produce a useless chart. Charts look like:

# Query_time distribution
#   1us
#  10us
# 100us
#   1ms
#  10ms  ###########################
# 100ms  ########################################################
#    1s  ########
#  10s+

See “OUTPUT” for more information.


type: string

If specified, the tool writes the last file offset, if there is one, to the given filename. When ran again with the same value for this option, the tool reads the last file offset from the file, seeks to that position in the log, and resumes parsing events from that point onward.


type: DSN

Save query classes for later review, and don’t report already reviewed classes.

The default table is percona_schema.query_review. Specify database (D) and table (t) DSN options to override the default. The database and table are automatically created unless --no-create-review-table is specified (see --[no]create-review-table).

If the table was created manually, it must have at least the following columns. You can add more columns for your own special purposes, but they won’t be used by pt-query-digest.

   checksum     CHAR(32) NOT NULL PRIMARY KEY,
   fingerprint  TEXT NOT NULL,
   sample       TEXT NOT NULL,
   first_seen   DATETIME,
   last_seen    DATETIME,
   reviewed_by  VARCHAR(20),
   reviewed_on  DATETIME,
   comments     TEXT

The columns are:

===========  ====================================================
checksum     A 64-bit checksum of the query fingerprint
fingerprint  The abstracted version of the query; its primary key
sample       The query text of a sample of the class of queries
first_seen   The smallest timestamp of this class of queries
last_seen    The largest timestamp of this class of queries
reviewed_by  Initially NULL; if set, query is skipped thereafter
reviewed_on  Initially NULL; not assigned any special meaning
comments     Initially NULL; not assigned any special meaning

Note that the fingerprint column is the true primary key for a class of queries. The checksum is just a cryptographic hash of this value, which provides a shorter value that is very likely to also be unique.

After parsing and aggregating events, your table should contain a row for each fingerprint. This option depends on --group-by fingerprint (which is the default). It will not work otherwise.


type: time

How long to run for each --iterations. The default is to run forever (you can interrupt with CTRL-C). Because --iterations defaults to 1, if you only specify --run-time, pt-query-digest runs for that amount of time and then exits. The two options are specified together to do collect-and-report cycles. For example, specifying --iterations 4 --run-time 15m with a continuous input (like STDIN or --processlist) will cause pt-query-digest to run for 1 hour (15 minutes x 4), reporting four times, once at each 15 minute interval.


type: string; default: clock

Set what the value of --run-time operates on. Following are the possible values for this option:


--run-time specifies an amount of real clock time during which the tool should run for each --iterations.


--run-time specifies an amount of log time. Log time is determined by timestamps in the log. The first timestamp seen is remembered, and each timestamp after that is compared to the first to determine how much log time has passed. For example, if the first timestamp seen is 12:00:00 and the next is 12:01:30, that is 1 minute and 30 seconds of log time. The tool will read events until the log time is greater than or equal to the specified --run-time value.

Since timestamps in logs are not always printed, or not always printed frequently, this mode varies in accuracy.


--run-time specifies interval boundaries of log time into which events are divided and reports are generated. This mode is different from the others because it doesn’t specify how long to run. The value of --run-time must be an interval that divides evenly into minutes, hours or days. For example, 5m divides evenly into hours (60/5=12, so 12 5 minutes intervals per hour) but 7m does not (60/7=8.6).

Specifying --run-time-mode interval --run-time 30m --iterations 0 is similar to specifying --run-time-mode clock --run-time 30m --iterations 0. In the latter case, pt-query-digest will run forever, producing reports every 30 minutes, but this only works effectively with continuous inputs like STDIN and the processlist. For fixed inputs, like log files, the former example produces multiple reports by dividing the log into 30 minutes intervals based on timestamps.

Intervals are calculated from the zeroth second/minute/hour in which a timestamp occurs, not from whatever time it specifies. For example, with 30 minute intervals and a timestamp of 12:10:30, the interval is not 12:10:30 to 12:40:30, it is 12:00:00 to 12:29:59. Or, with 1 hour intervals, it is 12:00:00 to 12:59:59. When a new timestamp exceeds the interval, a report is printed, and the next interval is recalculated based on the new timestamp.

Since --iterations is 1 by default, you probably want to specify a new value else pt-query-digest will only get and report on the first interval from the log since 1 interval = 1 iteration. If you want to get and report every interval in a log, specify --iterations 0.


type: int

Filter out all but the first N occurrences of each query. The queries are filtered on the first value in --group-by, so by default, this will filter by query fingerprint. For example, --sample 2 will permit two sample queries for each fingerprint. Useful in conjunction with --output slowlog to print the queries. You probably want to set --no-report to avoid the overhead of aggregating and reporting if you’re just using this to print out samples of queries. A complete example:

:program:`pt-query-digest` --sample 2 --no-report --output slowlog slow.log

type: string

Sets the user to be used to connect to the slaves. This parameter allows you to have a different user with less privileges on the slaves but that user must exist on all slaves.


type: string

Sets the password to be used to connect to the slaves. It can be used with –slave-user and the password for the user must be the same on all slaves.


type: Array

Set the MySQL variables in this comma-separated list of variable=value pairs.

By default, the tool sets:


Variables specified on the command line override these defaults. For example, specifying --set-vars wait_timeout=500 overrides the default value of 10000.

The tool prints a warning and continues if a variable cannot be set.


type: Hash

Show all values for these attributes.

By default pt-query-digest only shows as many of an attribute’s value as fit on a single line. This option allows you to specify attributes for which all values will be shown (line width is ignored). This only works for attributes with string values like user, host, db, etc. Multiple attributes can be specified, comma-separated.


type: string

Parse only queries newer than this value (parse queries since this date).

This option allows you to ignore queries older than a certain value and parse only those queries which are more recent than the value. The value can be several types:

* Simple time value N with optional suffix: N[shmd], where
  s=seconds, h=hours, m=minutes, d=days (default s if no suffix
  given); this is like saying "since N[shmd] ago"
* Full date with optional hours:minutes:seconds:
* Short, MySQL-style date:
* Any time expression evaluated by MySQL:

If you give a MySQL time expression, and you have not also specified a DSN for --explain, --processlist, or --review, then you must specify a DSN on the command line so that pt-query-digest can connect to MySQL to evaluate the expression.

The MySQL time expression is wrapped inside a query like “SELECT UNIX_TIMESTAMP(<expression>)”, so be sure that the expression is valid inside this query. For example, do not use UNIX_TIMESTAMP() because UNIX_TIMESTAMP(UNIX_TIMESTAMP()) returns 0.

Events are assumed to be in chronological order: older events at the beginning of the log and newer events at the end of the log. --since is strict: it ignores all queries until one is found that is new enough. Therefore, if the query events are not consistently timestamped, some may be ignored which are actually new enough.

See also --until.


short form: -S; type: string

Socket file to use for connection.


Show a timeline of events.

This option makes pt-query-digest print another kind of report: a timeline of the events. Each query is still grouped and aggregate into classes according to --group-by, but then they are printed in chronological order. The timeline report prints out the timestamp, interval, count and value of each class.

If all you want is the timeline report, then specify --no-report to suppress the default query analysis report. Otherwise, the timeline report will be printed at the end before the response-time profile (see --report-format and “OUTPUT”).

For example, this:

:program:`pt-query-digest` /path/to/log --group-by distill --timeline

will print something like:

# ########################################################
# distill report
# ########################################################
# 2009-07-25 11:19:27 1+00:00:01   2 SELECT foo
# 2009-07-27 11:19:30      00:01   2 SELECT bar
# 2009-07-27 11:30:00 1+06:30:00   2 SELECT foo

type: Array; default: slowlog

The type of input to parse. The permitted types are


Parse a binary log file that has first been converted to text using mysqlbinlog.

For example:

mysqlbinlog mysql-bin.000441 > mysql-bin.000441.txt

:program:`pt-query-digest` --type binlog mysql-bin.000441.txt


Parse a MySQL general log file. General logs lack a lot of “ATTRIBUTES”, notably Query_time. The default --order-by for general logs changes to Query_time:cnt.


Parse a log file in any variation of MySQL slow log format.


Inspect network packets and decode the MySQL client protocol, extracting queries and responses from it.

pt-query-digest does not actually watch the network (i.e. it does NOT “sniff packets”). Instead, it’s just parsing the output of tcpdump. You are responsible for generating this output; pt-query-digest does not do it for you. Then you send this to pt-query-digest as you would any log file: as files on the command line or to STDIN.

The parser expects the input to be formatted with the following options: -x -n -q -tttt. For example, if you want to capture output from your local machine, you can do something like the following (the port must come last on FreeBSD):

tcpdump -s 65535 -x -nn -q -tttt -i any -c 1000 port 3306 \
  > mysql.tcp.txt
:program:`pt-query-digest` --type tcpdump mysql.tcp.txt

The other tcpdump parameters, such as -s, -c, and -i, are up to you. Just make sure the output looks like this (there is a line break in the first line to avoid man-page problems):

2009-04-12 09:50:16.804849 IP
       > tcp 37
    0x0000:  4508 0059 6eb2 4000 4006 cde2 7f00 0001
    0x0010:  ....

Remember tcpdump has a handy -c option to stop after it captures some number of packets! That’s very useful for testing your tcpdump command. Note that tcpdump can’t capture traffic on a Unix socket. Read if you’re confused about this.

Devananda Van Der Veen explained on the MySQL Performance Blog how to capture traffic without dropping packets on busy servers. Dropped packets cause pt-query-digest to miss the response to a request, then see the response to a later request and assign the wrong execution time to the query. You can change the filter to something like the following to help capture a subset of the queries. (See for details.)

tcpdump -i any -s 65535 -x -n -q -tttt \
   'port 3306 and tcp[1] & 7 == 2 and tcp[3] & 7 == 2'

All MySQL servers running on port 3306 are automatically detected in the tcpdump output. Therefore, if the tcpdump output contains packets from multiple servers on port 3306 (for example,,, etc.), all packets/queries from all these servers will be analyzed together as if they were one server.

If you’re analyzing traffic for a MySQL server that is not running on port 3306, see --watch-server.

Also note that pt-query-digest may fail to report the database for queries when parsing tcpdump output. The database is discovered only in the initial connect events for a new client or when <USE db> is executed. If the tcpdump output contains neither of these, then pt-query-digest cannot discover the database.

Server-side prepared statements are supported. SSL-encrypted traffic cannot be inspected and decoded.


Raw logs are not MySQL logs but simple text files with one SQL statement per line, like:

/* Hello, world! */ SELECT * FROM t2 LIMIT 1
INSERT INTO t (a, b) VALUES ('foo', 'bar')

Since raw logs do not have any metrics, many options and features of pt-query-digest do not work with them.

One use case for raw logs is ranking queries by count when the only information available is a list of queries, from polling SHOW PROCESSLIST for example.


type: string

Parse only queries older than this value (parse queries until this date).

This option allows you to ignore queries newer than a certain value and parse only those queries which are older than the value. The value can be one of the same types listed for --since.

Unlike --since, --until is not strict: all queries are parsed until one has a timestamp that is equal to or greater than --until. Then all subsequent queries are ignored.


short form: -u; type: string

User for login if not current user.


type: Array

Report the number of variations in these attributes’ values.

Variations show how many distinct values an attribute had within a class. The usual value for this option is arg which shows how many distinct queries were in the class. This can be useful to determine a query’s cacheability.

Distinct values are determined by CRC32 checksums of the attributes’ values. These checksums are reported in the query report for attributes specified by this option, like:

# arg crc      109 (1/25%), 144 (1/25%)... 2 more

In that class there were 4 distinct queries. The checksums of the first two variations are shown, and each one occurred once (or, 25% of the time).

The counts of distinct variations is approximate because only 1,000 variations are saved. The mod (%) 1000 of the full CRC32 checksum is saved, so some distinct checksums are treated as equal.


Show version and exit.


default: yes

Check for the latest version of Percona Toolkit, MySQL, and other programs.

This is a standard “check for updates automatically” feature, with two additional features. First, the tool checks its own version and also the versions of the following software: operating system, Percona Monitoring and Management (PMM), MySQL, Perl, MySQL driver for Perl (DBD::mysql), and Percona Toolkit. Second, it checks for and warns about versions with known problems. For example, MySQL 5.5.25 had a critical bug and was re-released as 5.5.25a.

A secure connection to Percona’s Version Check database server is done to perform these checks. Each request is logged by the server, including software version numbers and unique ID of the checked system. The ID is generated by the Percona Toolkit installation script or when the Version Check database call is done for the first time.

Any updates or known problems are printed to STDOUT before the tool’s normal output. This feature should never interfere with the normal operation of the tool.

For more information, visit


default: yes

Output a trailing “G” in the reported SQL queries.

This makes the mysql client display the result using vertical format. Non-native MySQL clients like phpMyAdmin do not support this.


type: string

This option tells pt-query-digest which server IP address and port (like “”) to watch when parsing tcpdump (for --type tcpdump); all other servers are ignored. If you don’t specify it, pt-query-digest watches all servers by looking for any IP address using port 3306 or “mysql”. If you’re watching a server with a non-standard port, this won’t work, so you must specify the IP address and port to watch.

If you want to watch a mix of servers, some running on standard port 3306 and some running on non-standard ports, you need to create separate tcpdump outputs for the non-standard port servers and then specify this option for each. At present pt-query-digest cannot auto-detect servers on port 3306 and also be told to watch a server on a non-standard port.


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.

  • A

dsn: charset; copy: yes

Default character set.

  • D

dsn: database; copy: yes

Default database to use when connecting to MySQL.

  • F

dsn: mysql_read_default_file; copy: yes

Only read default options from the given file.

  • h

dsn: host; copy: yes

Connect to host.

  • p

dsn: password; copy: yes

Password to use when connecting. If password contains commas they must be escaped with a backslash: “exam,ple”

  • P

dsn: port; copy: yes

Port number to use for connection.

  • S

dsn: mysql_socket; copy: yes

Socket file to use for connection.

  • t

The --review or --history table.

  • u

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-query-digest ... > FILE 2>&1

Be careful: debugging output is voluminous and can generate several megabytes of output.


Using <PTDEBUG> might expose passwords. When debug is enabled, all command line parameters are shown in the 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

Please report bugs at Include the following information in your bug report:

  • Complete command-line used to run the tool

  • Tool --version

  • MySQL version of all servers involved

  • Output from the tool including STDERR

  • Input files (log/dump/config files, etc.)

If possible, include debugging output by running the tool with PTDEBUG; see “ENVIRONMENT”.


Visit to download the latest release of Percona Toolkit. Or, get the latest release from the command line:




You can also get individual tools from the latest release:


Replace TOOL with the name of any tool.


Events may have the following attributes. If writing a --filter, be sure to check that an attribute is defined in each event before using it, else the filter code may crash the tool with a “use of uninitialized value” error.

You can dump event attributes for any input like:

$ pt-query-digest                  \
    slow.log                       \
    --filter 'print Dumper $event' \
    --no-report                    \
    --sample 1

That will produce a lot of output with “attribute => value” pairs like:

$VAR1 = {
  Query_time => '0.033384',
  Rows_examined => '0',
  Rows_sent => '0',
  Thread_id => '10',
  Tmp_table => 'No',
  Tmp_table_on_disk => 'No',
  arg => 'SELECT col FROM tbl WHERE id=5',
  bytes => 103,
  cmd => 'Query',
  db => 'db1',
  fingerprint => 'select col from tbl where id=?',
  host => '',
  pos_in_log => 1334,
  ts => '071218 11:48:27',
  user => '[SQL_SLAVE]'


These attribute are common to all input --type and --processlist, except where noted.


The query text, or the command for admin commands like Ping.


The byte length of the arg.


“Query” or “Admin”.


The current database. The value comes from USE database statements. By default, Schema is an alias which is automatically changed to db; see --attribute-aliases.


An abstracted form of the query. See “FINGERPRINTS”.


Client host which executed the query.


The byte offset of the event in the log or tcpdump, except for --processlist.


The total time the query took, including lock time.


The timestamp of when the query ended.


Events have all available attributes from the log file. Therefore, you only need to look at the log file to see which events are available, but remember: not all events have the same attributes.

Percona Server adds many attributes to the slow log; see for more information.


These attributes are available when parsing --type tcpdump.


The MySQL error number if the query caused an error.


The client’s IP address. Certain log files may also contain this attribute.


Yes or No if no good index existed for the query (flag set by server).


Yes or No if the query did not use any index (flag set by server).


The client’s port number.


The number of warnings, as otherwise shown by SHOW WARNINGS.


If using --processlist, an id attribute is available for the process ID, in addition to the common attributes.


Baron Schwartz, Daniel Nichter, and Brian Fraser


This tool is part of Percona Toolkit, a collection of advanced command-line tools for MySQL developed by Percona. Percona Toolkit was forked from two projects in June, 2011: Maatkit and Aspersa. Those projects were created by Baron Schwartz and primarily developed by him and Daniel Nichter. Visit to learn about other free, open-source software from Percona.


pt-query-digest 3.6.0