pt-table-checksum - Perform an online replication consistency check, or checksum MySQL tables efficiently on one or many servers.
Usage: pt-table-checksum [OPTION...] DSN [DSN...]
pt-table-checksum checksums MySQL tables efficiently on one or more hosts. Each host is specified as a DSN and missing values are inherited from the first host. If you specify multiple hosts, the first is assumed to be the master.
STOP! Are you checksumming slaves against a master? Then be sure to learn what “–replicate” does. It is probably the option you want to use.
Checksum all slaves against the master:
pt-table-checksum \ h=master-host \ --replicate mydb.checksums # Wait for first command to complete and replication to catchup # on all slaves, then... pt-table-checksum \ h=master-host \ --replicate mydb.checksums \ --replicate-check 2
Checksum all databases and tables on two servers and print the differences:
pt-table-checksum h=host1,u=user h=host2 | pt-checksum-filter
See “SPECIFYING HOSTS” for more on the syntax of the host arguments.
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-table-checksum executes queries that cause the MySQL server to checksum its data. This can cause significant server load. It is read-only unless you use the “–replicate” option, in which case it inserts a small amount of data into the specified table.
At the time of this release, we know of no bugs that could cause serious harm to users. There are miscellaneous bugs that might be annoying.
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-table-checksum.
See also “BUGS” for more information on filing bugs and getting help.
pt-table-checksum generates table checksums for MySQL tables, typically useful for verifying your slaves are in sync with the master. The checksums are generated by a query on the server, and there is very little network traffic as a result.
Checksums typically take about twice as long as COUNT(*) on very large InnoDB tables in my tests. For smaller tables, COUNT(*) is a good bit faster than the checksums. See “–algorithm” for more details on performance.
If you specify more than one server, pt-table-checksum assumes the first server is the master and others are slaves. Checksums are parallelized for speed, forking off a child process for each table. Duplicate server names are ignored, but if you want to checksum a server against itself you can use two different forms of the hostname (for example, “localhost 127.0.0.1”, or “h=localhost,P=3306 h=localhost,P=3307”).
If you want to compare the tables in one database to those in another database on the same server, just checksum both databases:
pt-table-checksum --databases db1,db2
You can then use pt-checksum-filter to compare the results in both databases easily.
pt-table-checksum examines table structure only on the first host specified, so if anything differs on the others, it won’t notice. It ignores views.
The checksums work on MySQL version 3.23.58 through 6.0-alpha. They will not necessarily produce the same values on all versions. Differences in formatting and/or space-padding between 4.1 and 5.0, for example, will cause the checksums to be different.
Each host is specified on the command line as a DSN. A DSN is a comma-separted list of option=value pairs. The most basic DSN is h=host to specify the hostname of the server and use default for everything else (port, etc.). See “DSN OPTIONS” for more information.
DSN options that are listed as copy: yes are copied from the first DSN to subsequent DSNs that do not specify the DSN option. For example, h=host1,P=12345 h=host2 is equivalent to h=host1,P=12345 h=host2,P=12345. This allows you to avoid repeating DSN options that have the same value for all DSNs.
Connection-related command-line options like “–user” and “–password” provide default DSN values for the corresponding DSN options indicated by the short form of each option. For example, the short form of “–user” is -u which corresponds to the u DSN option, so --user bob h=hostis equivalent to h=host,u=bob. These defaults apply to all DSNs that do not specify the DSN option.
The DSN option value precedence from higest to lowest is:
* explicit values in each DSN on the command-line * copied values from the first DSN * default values from connection-related command-line options
If you are confused about how pt-table-checksum will connect to your servers, use the “–explain-hosts” option and it will tell you.
HOW FAST IS IT?¶
Speed and efficiency are important, because the typical use case is checksumming large amounts of data.
pt-table-checksum is designed to do very little work itself, and generates very little network traffic aside from inspecting table structures with SHOW CREATE TABLE. The results of checksum queries are typically 40-character or shorter strings.
The MySQL server does the bulk of the work, in the form of the checksum queries. The following benchmarks show the checksum query times for various checksum algorithms. The first two results are simply running COUNT(col8) and CHECKSUM TABLE on the table. CHECKSUM TABLE is just CRC32 under the hood, but it’s implemented inside the storage engine layer instead of at the MySQL layer.
ALGORITHM HASH FUNCTION EXTRA TIME ============== ============= ============== ===== COUNT(col8) 2.3 CHECKSUM TABLE 5.3 BIT_XOR FNV_64 12.7 ACCUM FNV_64 42.4 BIT_XOR MD5 --optimize-xor 80.0 ACCUM MD5 87.4 BIT_XOR SHA1 --optimize-xor 90.1 ACCUM SHA1 101.3 BIT_XOR MD5 172.0 BIT_XOR SHA1 197.3
The tests are entirely CPU-bound. The sample data is an InnoDB table with the following structure:
CREATE TABLE test ( col1 int NOT NULL, col2 date NOT NULL, col3 int NOT NULL, col4 int NOT NULL, col5 int, col6 decimal(3,1), col7 smallint unsigned NOT NULL, col8 timestamp NOT NULL, PRIMARY KEY (col2, col1), KEY (col7), KEY (col1) ) ENGINE=InnoDB
The table has 4303585 rows, 365969408 bytes of data and 173457408 bytes of indexes. The server is a Dell PowerEdge 1800 with dual 32-bit Xeon 2.8GHz processors and 2GB of RAM. The tests are fully CPU-bound, and the server is otherwise idle. The results are generally consistent to within a tenth of a second on repeated runs.
CRC32 is the default checksum function to use, and should be enough for most cases. If you need stronger guarantees that your data is identical, you should use one of the other functions.
The “–algorithm” option allows you to specify which algorithm you would like to use, but it does not guarantee that pt-table-checksum will use this algorithm. pt-table-checksum will ultimately select the best algorithm possible given various factors such as the MySQL version and other command line options.
The three basic algorithms in descending order of preference are CHECKSUM, BIT_XOR and ACCUM. CHECKSUM cannot be used if any one of these criteria is true:
* --where is used * --since is used * --chunk-size is used * --replicate is used * --count is used * MySQL version less than 4.1.1
The BIT_XOR algorithm also requires MySQL version 4.1.1 or later.
After checking these criteria, if the requested “–algorithm” remains then it is used, otherwise the first remaining algorithm with the highest preference is used.
If you are using this tool to verify your slaves still have the same data as the master, which is why I wrote it, you should read this section.
The best way to do this with replication is to use the “–replicate” option. When the queries are finished running on the master and its slaves, you can go to the slaves and issue SQL queries to see if any tables are different from the master. Try the following:
SELECT db, tbl, chunk, this_cnt-master_cnt AS cnt_diff, this_crc <> master_crc OR ISNULL(master_crc) <> ISNULL(this_crc) AS crc_diff FROM checksum WHERE master_cnt <> this_cnt OR master_crc <> this_crc OR ISNULL(master_crc) <> ISNULL(this_crc);
The “–replicate-check” option can do this query for you. If you can’t use this method, try the following:
If your servers are not being written to, you can just run the tool with no further ado:pt-table-checksum server1 server2 ... serverN
If the servers are being written to, you need some way to make sure they are consistent at the moment you run the checksums. For situations other than master-slave replication, you will have to figure this out yourself. You may be able to use the “–where” option with a date or time column to only checksum data that’s not recent.
If you are checksumming a master and slaves, you can do a fast parallel checksum and assume the slaves are caught up to the master. In practice, this tends to work well except for tables which are constantly updated. You can use the “–slave-lag” option to see how far behind each slave was when it checksummed a given table. This can help you decide whether to investigate further.
The next most disruptive technique is to lock the table on the master, then take checksums. This should prevent changes from propagating to the slaves. You can just lock on the master (with “–lock”), or you can both lock on the master and wait on the slaves till they reach that point in the master’s binlog (“–wait”). Which is better depends on your workload; only you know that.
If you decide to make the checksums on the slaves wait until they’re guaranteed to be caught up to the master, the algorithm looks like this:For each table, Master: lock table Master: get pos In parallel, Master: checksum Slave(s): wait for pos, then checksum End Master: unlock table End
What I typically do when I’m not using the “–replicate” option is simply run the tool on all servers with no further options. This runs fast, parallel, non-blocking checksums simultaneously. If there are tables that look different, I re-run with “–wait”=600 on the tables in question. This makes the tool lock on the master as explained above.
Output is to STDOUT, one line per server and table, with header lines for each database. I tried to make the output easy to process with awk. For this reason columns are always present. If there’s no value, pt-table-checksum prints ‘NULL’.
The default is column-aligned output for human readability, but you can change it to tab-separated if you want. Use the “–tab” option for this.
Output is unsorted, though all lines for one table should be output together. For speed, all checksums are done in parallel (as much as possible) and may complete out of the order in which they were started. You might want to run them through another script or command-line utility to make sure they are in the order you want. If you pipe the output through pt-checksum-filter, you can sort the output and/or avoid seeing output about tables that have no differences.
The columns in the output are as follows. The database, table, and chunk come first so you can sort by them easily (they are the “primary key”).
Output from “–replicate-check” and “–checksum” are different.
The database the table is in.
The table name.
The chunk (see “–chunk-size”). Zero if you are not doing chunked checksums.
The server’s hostname.
The table’s storage engine.
The table’s row count, unless you specified to skip it. If OVERSIZE is printed, the chunk was skipped because the actual number of rows was greater than “–chunk-size” times “–chunk-size-limit”.
The table’s checksum, unless you specified to skip it or the table has no rows. some types of checksums will be 0 if there are no rows; others will print NULL.
How long it took to checksum the CHUNK, not including WAIT time. Total checksum time is WAIT + TIME.
How long the slave waited to catch up to its master before beginning to checksum. WAIT is always 0 for the master. See “–wait”.
The return value of MASTER_POS_WAIT(). STAT is always NULL for the master.
How far the slave lags the master, as reported by SHOW SLAVE STATUS. LAG is always NULL for the master.
REPLICATE TABLE MAINTENANCE¶
If you use “–replicate” to store and replicate checksums, you may need to perform maintenance on the replicate table from time to time to remove old checksums. This section describes when checksums in the replicate table are deleted automatically by pt-table-checksum and when you must manually delete them.
Before starting, pt-table-checksum calculates chunks for each table, even if “–chunk-size” is not specified (in that case there is one chunk: “1=1”). Then, before checksumming each table, the tool deletes checksum chunks in the replicate table greater than the current number of chunks. For example, if a table is chunked into 100 chunks, 0-99, then pt-table-checksum does:
DELETE FROM replicate table WHERE db=? AND tbl=? AND chunk > 99
That removes any high-end chunks from previous runs which no longer exist. Currently, this operation cannot be disabled.
If you use “–resume”, “–resume-replicate”, or “–modulo”, then you need to be careful that the number of rows in a table does not decrease so much that the number of chunks decreases too, else some checksum chunks may be deleted. The one exception is if only rows at the high end of the range are deleted. In that case, the high-end chunks are deleted and lower chunks remain unchanged. An increasing number of rows or chunks should not cause any adverse affects.
Changing the “–chunk-size” between runs with “–resume”, “–resume-replicate”, or “–modulo” can cause odd or invalid checksums. You should not do this. It won’t work with the resume options. With “–modulo”, the safest thing to do is manually delete all the rows in the replicate table for the table in question and start over.
If the replicate table becomes cluttered with old or invalid checksums and the auto-delete operation is not deleting them, then you will need to manually clean up the replicate table. Alternatively, if you specify “–empty-replicate-table”, then the tool deletes every row in the replicate table.
An exit status of 0 (sometimes also called a return value or return code) indicates success. If there is an error checksumming any table, the exit status is 1.
When running “–replicate-check”, if any slave has chunks that differ from the master, the exit status is 1.
If you are using innotop (see http://code.google.com/p/innotop), mytop, or another tool to watch currently running MySQL queries, you may see the checksum queries. They look similar to this:
REPLACE /*test.test_tbl:'2'/'5'*/ INTO test.checksum(db, ...
Since pt-table-checksum’s queries run for a long time and tend to be textually very long, and thus won’t fit on one screen of these monitoring tools, I’ve been careful to place a comment at the beginning of the query so you can see what it is and what it’s doing. The comment contains the name of the table that’s being checksummed, the chunk it is currently checksumming, and how many chunks will be checksummed. In the case above, it is checksumming chunk 2 of 5 in table test.test_tbl.
“–schema” is restricted to option groups Connection, Filter, Output, Help, Config, Safety.
“–empty-replicate-table”, “–resume” and “–resume-replicate” are mutually exclusive.
This tool accepts additional command-line arguments. Refer to the “SYNOPSIS” and usage information for details.
Checksum algorithm (ACCUM|CHECKSUM|BIT_XOR).
Specifies which checksum algorithm to use. Valid arguments are CHECKSUM, BIT_XOR and ACCUM. The latter two do cryptographic hash checksums. See also “ALGORITHM SELECTION”.
CHECKSUM is built into MySQL, but has some disadvantages. BIT_XOR and ACCUM are implemented by SQL queries. They use a cryptographic hash of all columns concatenated together with a separator, followed by a bitmap of each nullable column that is NULL (necessary because CONCAT_WS() skips NULL columns).
CHECKSUM is the default. This method uses MySQL’s built-in CHECKSUM TABLE command, which is a CRC32 behind the scenes. It cannot be used before MySQL 4.1.1, and various options disable it as well. It does not simultaneously count rows; that requires an extra COUNT(*) query. This is a good option when you are using MyISAM tables with live checksums enabled; in this case both the COUNT(*) and CHECKSUM queries will run very quickly.
The BIT_XOR algorithm is available for MySQL 4.1.1 and newer. It uses BIT_XOR(), which is order-independent, to reduce all the rows to a single checksum.
ACCUM uses a user variable as an accumulator. It reduces each row to a single checksum, which is concatenated with the accumulator and re-checksummed. This technique is order-dependent. If the table has a primary key, it will be used to order the results for consistency; otherwise it’s up to chance.
The pathological worst case is where identical rows will cancel each other out in the BIT_XOR. In this case you will not be able to distinguish a table full of one value from a table full of another value. The ACCUM algorithm will distinguish them.
However, the ACCUM algorithm is order-dependent, so if you have two tables with identical data but the rows are out of order, you’ll get different checksums with ACCUM.
If a given algorithm won’t work for some reason, pt-table-checksum falls back to another. The least common denominator is ACCUM, which works on MySQL 3.23.2 and newer.
The database.table with arguments for each table to checksum.
This table may be named anything you wish. It must contain at least the following columns:
CREATE TABLE checksum_args ( db char(64) NOT NULL, tbl char(64) NOT NULL, -- other columns as desired PRIMARY KEY (db, tbl) );
In addition to the columns shown, it may contain any of the other columns listed here (Note: this list is used by the code, MAGIC_overridable_args):
algorithm chunk-column chunk-index chunk-size columns count crc function lock modulo use-index offset optimize-xor chunk-size-limit probability separator save-since single-chunk since since-column sleep sleep-coef trim wait where
Each of these columns corresponds to the long form of a command-line option. Each column should be NULL-able. Column names with hyphens should be enclosed in backticks (e.g. `chunk-size`) when the table is created. The data type does not matter, but it’s suggested you use a sensible data type to prevent garbage data.
When pt-table-checksum checksums a table, it will look for a matching entry in this table. Any column that has a defined value will override the corresponding command-line argument for the table being currently processed. In this way it is possible to specify custom command-line arguments for any table.
If you add columns to the table that aren’t in the above list of allowable columns, it’s an error. The exceptions are db, tbl, and ts. The tscolumn can be used as a timestamp for easy visibility into the last time the since column was updated with “–save-since”.
This table is assumed to be located on the first server given on the command-line.
Prompt for a password when connecting to MySQL.
type: time; group: Throttle; default: 1s
How often to check for slave lag if “–check-slave-lag” is given.
default: yes; group: Safety
Do not “–replicate” if any replication filters are set. When –replicate is specified, pt-table-checksum tries to detect slaves and look for options that filter replication, such as binlog_ignore_db and replicate_do_db. If it finds any such filters, it aborts with an error. Replication filtering makes it impossible to be sure that the checksum queries won’t break replication or simply fail to replicate. If you are sure that it’s OK to run the checksum queries, you can negate this option to disable the checks. See also “–replicate-database”.
type: DSN; group: Throttle
Pause checksumming until the specified slave’s lag is less than “–max-lag”.
If this option is specified and “–throttle-method” is set to slavelagthen “–throttle-method” only checks this slave.
Print checksums and table names in the style of md5sum (disables “–[no]count”).
Makes the output behave more like the output of md5sum. The checksum is first on the line, followed by the host, database, table, and chunk number, concatenated with dots.
Prefer this column for dividing tables into chunks. By default, pt-table-checksum chooses the first suitable column for each table, preferring to use the primary key. This option lets you specify a preferred column, which pt-table-checksum uses if it exists in the table and is chunkable. If not, then pt-table-checksum will revert to its default behavior. Be careful when using this option; a poor choice could cause bad performance. This is probably best to use when you are checksumming only a single table, not an entire server. See also “–chunk-index”.
Prefer this index for chunking tables. By default, pt-table-checksum chooses an appropriate index for the “–chunk-column” (even if it chooses the chunk column automatically). This option lets you specify the index you prefer. If the index doesn’t exist, then pt-table-checksum will fall back to its default behavior. pt-table-checksum adds the index to the checksum SQL statements in a FORCE INDEX clause. Be careful when using this option; a poor choice of index could cause bad performance. This is probably best to use when you are checksumming only a single table, not an entire server.
type: string; default: open
Set which ends of the chunk range are open or closed. Possible values are one of MAGIC_chunk_range:
VALUE OPENS/CLOSES ========== ====================== open Both ends are open openclosed Low end open, high end closed
By default pt-table-checksum uses an open range of chunks like:
`id` < '10' `id` >= '10' AND < '20' `id` >= '20'
That range is open because the last chunk selects any row with id greater than (or equal to) 20. An open range can be a problem in cases where a lot of new rows are inserted with IDs greater than 20 while pt-table-checksumming is running because the final open-ended chunk will select all the newly inserted rows. (The less common case of inserting rows with IDs less than 10 would require a closedopen range but that is not currently implemented.) Specifying openclosed will cause the final chunk to be closed like:
`id` >= '20' AND `id` <= N
N is the MAX(\`id\`) that pt-table-checksum used when it first chunked the rows. Therefore, it will only chunk the range of rows that existed when the tool started and not any newly inserted rows (unless those rows happen to be inserted with IDs less than N).
See also “–chunk-size-limit”.
Approximate number of rows or size of data to checksum at a time. Allowable suffixes are k, M, G. Disallows --algorithm CHECKSUM.
If you specify a chunk size, pt-table-checksum will try to find an index that will let it split the table into ranges of approximately “–chunk-size” rows, based on the table’s index statistics. Currently only numeric and date types can be chunked.
If the table is chunkable, pt-table-checksum will checksum each range separately with parameters in the checksum query’s WHERE clause. If pt-table-checksum cannot find a suitable index, it will do the entire table in one chunk as though you had not specified “–chunk-size” at all. Each table is handled individually, so some tables may be chunked and others not.
The chunks will be approximately sized, and depending on the distribution of values in the indexed column, some chunks may be larger than the value you specify.
If you specify a suffix (one of k, M or G), the parameter is treated as a data size rather than a number of rows. The output of SHOW TABLE STATUS is then used to estimate the amount of data the table contains, and convert that to a number of rows.
type: float; default: 2.0; group: Safety
Do not checksum chunks with this many times more rows than “–chunk-size”.
When “–chunk-size” is given it specifies an ideal size for each chunk of a chunkable table (in rows; size values are converted to rows). Before checksumming each chunk, pt-table-checksum checks how many rows are in the chunk with EXPLAIN. If the number of rows reported by EXPLAIN is this many times greater than “–chunk-size”, then the chunk is skipped and OVERSIZEis printed for the COUNT column of the “OUTPUT”.
For example, if you specify “–chunk-size” 100 and a chunk has 150 rows, then it is checksummed with the default “–chunk-size-limit” value 2.0 because 150 is less than 100 * 2.0. But if the chunk has 205 rows, then it is not checksummed because 205 is greater than 100 * 2.0.
The minimum value for this option is 1 which means that no chunk can be any larger than “–chunk-size”. You probably don’t want to specify 1 because rows reported by EXPLAIN are estimates which can be greater than or less than the real number of rows in the chunk. If too many chunks are skipped because they are oversize, you might want to specify a value larger than 2.
You can disable oversize chunk checking by specifying “–chunk-size-limit” 0.
See also “–unchunkable-tables”.
short form: -c; type: array; group: Filter
Checksum only this comma-separated list of columns.
type: Array; group: Config
Read this comma-separated list of config files; if specified, this must be the first option on the command line.
Count rows in tables. This is built into ACCUM and BIT_XOR, but requires an extra query for CHECKSUM.
This is disabled by default to avoid an extra COUNT(*) query when “–algorithm” is CHECKSUM. If you have only MyISAM tables and live checksums are enabled, both CHECKSUM and COUNT will be very fast, but otherwise you may want to use one of the other algorithms.
Do a CRC (checksum) of tables.
Take the checksum of the rows as well as their count. This is enabled by default. If you disable it, you’ll just get COUNT(*) queries.
Create the replicate table given by “–replicate” if it does not exist.
Normally, if the replicate table given by “–replicate” does not exist, pt-table-checksum will die. With this option, however, pt-table-checksumwill create the replicate table for you, using the database.table name given to “–replicate”.
The structure of the replicate table is the same as the suggested table mentioned in “–replicate”. Note that since ENGINE is not specified, the replicate table will use the server’s default storage engine. If you want to use a different engine, you need to create the table yourself.
short form: -d; type: hash; group: Filter
Only checksum this comma-separated list of databases.
Only checksum databases whose names match this Perl regex.
short form: -F; type: string; group: Connection
Only read mysql options from the given file. You must give an absolute pathname.
DELETE all rows in the “–replicate” table before starting.
Issues a DELETE against the table given by “–replicate” before beginning work. Ignored if “–replicate” is not specified. This can be useful to remove entries related to tables that no longer exist, or just to clean out the results of a previous run.
If you want to delete entries for specific databases or tables you must do this manually.
short form: -e; type: hash; group: Filter
Do only this comma-separated list of storage engines.
Show, but do not execute, checksum queries (disables “–empty-replicate-table”).
Print full DSNs for each host and exit. This option allows you to see how pt-table-checksum parses DSNs from the command-line and how it will connect to those hosts. See “SPECIFYING HOSTS”.
Precision for FLOAT and DOUBLE number-to-string conversion. Causes FLOAT and DOUBLE values to be rounded to the specified number of digits after the decimal point, with the ROUND() function in MySQL. This can help avoid checksum mismatches due to different floating-point representations of the same values on different MySQL versions and hardware. The default is no rounding; the values are converted to strings by the CONCAT() function, and MySQL chooses the string representation. If you specify a value of 2, for example, then the values 1.008 and 1.009 will be rounded to 1.01, and will checksum as equal.
Hash function for checksums (FNV1A_64, MURMUR_HASH, SHA1, MD5, CRC32, etc).
You can use this option to choose the cryptographic hash function used for “–algorithm”=ACCUM or “–algorithm”=BIT_XOR. The default is to use CRC32, but MD5 and SHA1 also work, and you can use your own function, such as a compiled UDF, if you wish. Whatever function you specify is run in SQL, not in Perl, so it must be available to MySQL.
The FNV1A_64 UDF mentioned in the benchmarks is much faster than MD5. The C++ source code is distributed with Maatkit. It is very simple to compile and install; look at the header in the source code for instructions. If it is installed, it is preferred over MD5. You can also use the MURMUR_HASH function if you compile and install that as a UDF; the source is also distributed with Maatkit, and it is faster and has better distribution than FNV1A_64.
Show help and exit.
type: Hash; group: Filter
Ignore this comma-separated list of columns when calculating the checksum.
This option only affects the checksum when using the ACCUM or BIT_XOR “–algorithm”.
type: Hash; group: Filter
Ignore this comma-separated list of databases.
Ignore databases whose names match this Perl regex.
type: Hash; default: FEDERATED,MRG_MyISAM; group: Filter
Ignore this comma-separated list of storage engines.
type: Hash; group: Filter
Ignore this comma-separated list of tables.
Table names may be qualified with the database name.
Ignore tables whose names match the Perl regex.
Lock on master until done on slaves (implies “–slave-lag”).
This option can help you to get a consistent read on a master and many slaves. If you specify this option, pt-table-checksum will lock the table on the first server on the command line, which it assumes to be the master. It will keep this lock until the checksums complete on the other servers.
This option isn’t very useful by itself, so you probably want to use “–wait” instead.
Note: if you’re checksumming a slave against its master, you should use “–replicate”. In that case, there’s no need for locking, waiting, or any of that.
type: time; group: Throttle; default: 1s
Suspend checksumming if the slave given by “–check-slave-lag” lags.
This option causes pt-table-checksum to look at the slave every time it’s about to checksum a chunk. If the slave’s lag is greater than the option’s value, or if the slave isn’t running (so its lag is NULL), pt-table-checksum sleeps for “–check-interval” seconds and then looks at the lag again. It repeats until the slave is caught up, then proceeds to checksum the chunk.
This option is useful to let you checksum data as fast as the slaves can handle it, assuming the slave you directed pt-table-checksum to monitor is representative of all the slaves that may be replicating from this server. It should eliminate the need for “–sleep” or “–sleep-coef”.
Do only every Nth chunk on chunked tables.
This option lets you checksum only some chunks of the table. This is a useful alternative to “–probability” when you want to be sure you get full coverage in some specified number of runs; for example, you can do only every 7th chunk, and then use “–offset” to rotate the modulo every day of the week.
Just like with “–probability”, a table that cannot be chunked is done every time.
type: string; default: 0
Modulo offset expression for use with “–modulo”.
The argument may be an SQL expression, such as WEEKDAY(NOW()) (which returns a number from 0 through 6). The argument is evaluated by MySQL. The result is used as follows: if chunk_num % “–modulo” == “–offset”, the chunk will be checksummed.
Optimize BIT_XOR with user variables.
This option specifies to use user variables to reduce the number of times each row must be passed through the cryptographic hash function when you are using the BIT_XOR algorithm.
With the optimization, the queries look like this in pseudo-code:SELECT CONCAT( BIT_XOR(SLICE_OF(@user_variable)), BIT_XOR(SLICE_OF(@user_variable)), ... BIT_XOR(SLICE_OF(@user_variable := HASH(col1, col2... colN))));
The exact positioning of user variables and calls to the hash function is determined dynamically, and will vary between MySQL versions. Without the optimization, it looks like this:SELECT CONCAT( BIT_XOR(SLICE_OF(MD5(col1, col2... colN))), BIT_XOR(SLICE_OF(MD5(col1, col2... colN))), ... BIT_XOR(SLICE_OF(MD5(col1, col2... colN))));
The difference is the number of times all the columns must be mashed together and fed through the hash function. If you are checksumming really large columns, such as BLOB or TEXT columns, this might make a big difference.
short form: -p; type: string; group: Connection
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; group: Connection
Port number to use for connection.
type: int; default: 100
Checksums will be run with this percent probability.
This is an integer between 1 and 100. If 100, every chunk of every table will certainly be checksummed. If less than that, there is a chance that some chunks of some tables will be skipped. This is useful for routine jobs designed to randomly sample bits of tables without checksumming the whole server. By default, if a table is not chunkable, it will be checksummed every time even when the probability is less than 100. You can override this with “–single-chunk”.
See also “–modulo”.
type: array; default: time,30
Print progress reports to STDERR. Currently, this feature is only for when “–throttle-method” waits for slaves to catch up.
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.
short form: -q; group: Output
Do not print checksum results.
|--recheck||Re-checksum chunks that “–replicate-check” found to be different.|
type: int; group: Throttle
Number of levels to recurse in the hierarchy when discovering slaves. Default is infinite.
Preferred recursion method for discovering slaves.
Possible methods are:
METHOD USES =========== ================ processlist SHOW PROCESSLIST hosts SHOW SLAVE HOSTS
The processlist method is preferred because SHOW SLAVE HOSTS is not reliable. However, the hosts method is required if the server uses a non-standard port (not 3306). Usually pt-table-checksum does the right thing and finds the slaves, but you may give a preferred method and it will be used first. If it doesn’t find any slaves, the other methods will be tried.
Replicate checksums to slaves (disallows –algorithm CHECKSUM).
This option enables a completely different checksum strategy for a consistent, lock-free checksum across a master and its slaves. Instead of running the checksum queries on each server, you run them only on the master. You specify a table, fully qualified in db.table format, to insert the results into. The checksum queries will insert directly into the table, so they will be replicated through the binlog to the slaves.
When the queries are finished replicating, you can run a simple query on each slave to see which tables have differences from the master. With the “–replicate-check” option, pt-table-checksum can run the query for you to make it even easier. See “CONSISTENT CHECKSUMS” for details.
If you find tables that have differences, you can use the chunk boundaries in a WHERE clause with pt-table-sync to help repair them more efficiently. See pt-table-sync for details.
The table must have at least these columns: db, tbl, chunk, boundaries, this_crc, master_crc, this_cnt, master_cnt. The table may be named anything you wish. Here is a suggested table structure, which is automatically used for “–create-replicate-table” (MAGIC_create_replicate):
CREATE TABLE checksum ( db char(64) NOT NULL, tbl char(64) NOT NULL, chunk int NOT NULL, boundaries char(100) NOT NULL, this_crc char(40) NOT NULL, this_cnt int NOT NULL, master_crc char(40) NULL, master_cnt int NULL, ts timestamp NOT NULL, PRIMARY KEY (db, tbl, chunk) );
Be sure to choose an appropriate storage engine for the checksum table. If you are checksumming InnoDB tables, for instance, a deadlock will break replication if the checksum table is non-transactional, because the transaction will still be written to the binlog. It will then replay without a deadlock on the slave and break replication with “different error on master and slave.” This is not a problem with pt-table-checksum, it’s a problem with MySQL replication, and you can read more about it in the MySQL manual.
This works only with statement-based replication (pt-table-checksum will switch the binlog format to STATEMENT for the duration of the session if your server uses row-based replication).
In contrast to running the tool against multiple servers at once, using this option eliminates the complexities of synchronizing checksum queries across multiple servers, which normally requires locking and unlocking, waiting for master binlog positions, and so on. Thus, it disables “–lock”, “–wait”, and “–slave-lag” (but not “–check-slave-lag”, which is a way to throttle the execution speed).
The checksum queries actually do a REPLACE into this table, so existing rows need not be removed before running. However, you may wish to do this anyway to remove rows related to tables that don’t exist anymore. The “–empty-replicate-table” option does this for you.
Since the table must be qualified with a database (e.g. db.checksums), pt-table-checksum will only USE this database. This may be important if any replication options are set because it could affect whether or not changes to the table are replicated.
If the slaves have any –replicate-do-X or –replicate-ignore-X options, you should be careful not to checksum any databases or tables that exist on the master and not the slaves. Changes to such tables may not normally be executed on the slaves because of the –replicate options, but the checksum queries modify the contents of the table that stores the checksums, not the tables whose data you are checksumming. Therefore, these queries will be executed on the slave, and if the table or database you’re checksumming does not exist, the queries will cause replication to fail. For more information on replication rules, see http://dev.mysql.com/doc/en/replication-rules.html.
The table specified by “–replicate” will never be checksummed itself.
Check results in “–replicate” table, to the specified depth. You must use this after you run the tool normally; it skips the checksum step and only checks results.
It recursively finds differences recorded in the table given by “–replicate”. It recurses to the depth you specify: 0 is no recursion (check only the server you specify), 1 is check the server and its slaves, 2 is check the slaves of its slaves, and so on.
It finds differences by running the query shown in “CONSISTENT CHECKSUMS”, and prints results, then exits after printing. This is just a convenient way of running the query so you don’t have to do it manually.
The output is one informational line per slave host, followed by the results of the query, if any. If “–quiet” is specified, there is no output. If there are no differences between the master and any slave, there is no output. If any slave has chunks that differ from the master, pt-table-checksum’s exit status is 1; otherwise it is 0.
This option makes pt-table-checksum look for slaves by running SHOW PROCESSLIST. If it finds connections that appear to be from slaves, it derives connection information for each slave with the same default-and-override method described in “SPECIFYING HOSTS”.
If SHOW PROCESSLIST doesn’t return any rows, pt-table-checksum looks at SHOW SLAVE HOSTS instead. The host and port, and user and password if available, from SHOW SLAVE HOSTS are combined into a DSN and used as the argument. This requires slaves to be configured with report-host, report-port and so on.
This requires the @@SERVER_ID system variable, so it works only on MySQL 3.23.26 or newer.
USE only this database with “–replicate”. By default, pt-table-checksum executes USE to set its default database to the database that contains the table it’s currently working on. It changes its default database as it works on different tables. This is is a best effort to avoid problems with replication filters such as binlog_ignore_db and replicate_ignore_db. However, replication filters can create a situation where there simply is no one right way to do things. Some statements might not be replicated, and others might cause replication to fail on the slaves. In such cases, it is up to the user to specify a safe default database. This option specifies a default database that pt-table-checksum selects with USE, and never changes afterwards. See also <L”–[no]check-replication-filters”>.
Resume checksum using given output file from a previously interrupted run.
The given output file should be the literal output from a previous run of pt-table-checksum. For example:
pt-table-checksum host1 host2 -C 100 > checksum_results.txt pt-table-checksum host1 host2 -C 100 --resume checksum_results.txt
The command line options given to the first run and the resumed run must be identical (except, of course, for –resume). If they are not, the result will be unpredictable and probably wrong.
“–resume” does not work with “–replicate”; for that, use “–resume-replicate”.
This option resumes a previous checksum operation using “–replicate”. It is like “–resume” but does not require an output file. Instead, it uses the checksum table given to “–replicate” to determine where to resume the checksum operation.
When “–arg-table” and “–since” are given, save the current “–since” value into that table’s since column after checksumming. In this way you can incrementally checksum tables by starting where the last one finished.
The value to be saved could be the current timestamp, or it could be the maximum existing value of the column given by “–since-column”. It depends on what options are in effect. See the description of “–since” to see how timestamps are different from ordinary values.
|--schema||Checksum SHOW CREATE TABLE instead of table data.|
type: string; default: #
The separator character used for CONCAT_WS().
This character is used to join the values of columns when checksumming with “–algorithm” of BIT_XOR or ACCUM.
type: string; default: wait_timeout=10000; group: Connection
Set these MySQL variables. Immediately after connecting to MySQL, this string will be appended to SET and executed.
Checksum only data newer than this value.
If the table is chunk-able or nibble-able, this value will apply to the first column of the chunked or nibbled index.
This is not too different to “–where”, but instead of universally applying a WHERE clause to every table, it selectively finds the right column to use and applies it only if such a column is found. See also “–since-column”.
The argument may be an expression, which is evaluated by MySQL. For example, you can specify CURRENT_DATE - INTERVAL 7 DAY to get the date of one week ago.
A special bit of extra magic: if the value is temporal (looks like a date or datetime), then the table is checksummed only if the create time (or last modified time, for tables that report the last modified time, such as MyISAM tables) is newer than the value. In this sense it’s not applied as a WHERE clause at all.
The column name to be used for “–since”.
The default is for the tool to choose the best one automatically. If you specify a value, that will be used if possible; otherwise the best auto-determined one; otherwise none. If the column doesn’t exist in the table, it is just ignored.
Permit skipping with “–probability” if there is only one chunk.
Normally, if a table isn’t split into many chunks, it will always be checksummed regardless of “–probability”. This setting lets the probabilistic behavior apply to tables that aren’t divided into chunks.
Report replication delay on the slaves.
If this option is enabled, the output will show how many seconds behind the master each slave is. This can be useful when you want a fast, parallel, non-blocking checksum, and you know your slaves might be delayed relative to the master. You can inspect the results and make an educated guess whether any discrepancies on the slave are due to replication delay instead of corrupt data.
If you’re using “–replicate”, a slave that is delayed relative to the master does not invalidate the correctness of the results, so this option is disabled.
type: int; group: Throttle
Sleep time between checksums.
If this option is specified, pt-table-checksum will sleep the specified number of seconds between checksums. That is, it will sleep between every table, and if you specify “–chunk-size”, it will also sleep between chunks.
This is a very crude way to throttle checksumming; see “–sleep-coef” and “–check-slave-lag” for techniques that permit greater control.
type: float; group: Throttle
Calculate “–sleep” as a multiple of the last checksum time.
If this option is specified, pt-table-checksum will sleep the amount of time elapsed during the previous checksum, multiplied by the specified coefficient. This option is ignored if “–sleep” is specified.
This is a slightly more sophisticated way to throttle checksum speed: sleep a varying amount of time between chunks, depending on how long the chunks are taking. Even better is to use “–check-slave-lag” if you’re checksumming master/slave replication.
short form: -S; type: string; group: Connection
Socket file to use for connection.
Print tab-separated output, not column-aligned output.
short form: -t; type: hash; group: Filter
Do only this comma-separated list of tables.
Table names may be qualified with the database name.
Only checksum tables whose names match this Perl regex.
type: string; default: none; group: Throttle
Throttle checksumming when doing “–replicate”.
At present there is only one method: slavelag. When “–replicate” is used, pt-table-checksum automatically sets “–throttle-method” to slavelag and discovers every slave and throttles checksumming if any slave lags more than “–max-lag”. Specify -throttle-method none to disable this behavior completely, or specify “–check-slave-lag” and pt-table-checksum will only check that slave.
See also “–recurse” and “–recursion-method”.
Trim VARCHAR columns (helps when comparing 4.1 to >= 5.0).
This option adds a TRIM() to VARCHAR columns in BIT_XOR and ACCUMmodes.
This is useful when you don’t care about the trailing space differences between MySQL versions which vary in their handling of trailing spaces. MySQL 5.0 and later all retain trailing spaces in VARCHAR, while previous versions would remove them.
Checksum tables that cannot be chunked when “–chunk-size” is specified.
By default pt-table-checksum will not checksum a table that cannot be chunked when “–chunk-size” is specified because this might result in a huge, non-chunkable table being checksummed in one huge, memory-intensive chunk.
Specifying this option allows checksumming tables that cannot be chunked. Be careful when using this option! Make sure any non-chunkable tables are not so large that they will cause the tool to consume too much memory or CPU.
See also “–chunk-size-limit”.
Add FORCE INDEX hints to SQL statements.
By default pt-table-checksum adds an index hint (FORCE INDEX for MySQL v4.0.9 and newer, USE INDEX for older MySQL versions) to each SQL statement to coerce MySQL into using the “–chunk-index” (whether the index is specified by the option or auto-detected). Specifying --no-use-index causes pt-table-checksum to omit index hints.
short form: -u; type: string; group: Connection
User for login if not current user.
Verify checksum compatibility across servers.
This option runs a trivial checksum on all servers to ensure they have compatible CONCAT_WS() and cryptographic hash functions.
Versions of MySQL before 4.0.14 will skip empty strings and NULLs in CONCAT_WS, and others will only skip NULLs. The two kinds of behavior will produce different results if you have any columns containing the empty string in your table. If you know you don’t (for instance, all columns are integers), you can safely disable this check and you will get a reliable checksum even on servers with different behavior.
Show version and exit.
short form: -w; type: time
Wait this long for slaves to catch up to their master (implies “–lock” “–slave-lag”).
Note: the best way to verify that a slave is in sync with its master is to use “–replicate” instead. The “–wait” option is really only useful if you’re trying to compare masters and slaves without using “–replicate”, which is possible but complex and less efficient in some ways.
This option helps you get a consistent checksum across a master server and its slaves. It combines locking and waiting to accomplish this. First it locks the table on the master (the first server on the command line). Then it finds the master’s binlog position. Checksums on slaves will be deferred until they reach the same binlog position.
The argument to the option is the number of seconds to wait for the slaves to catch up to the master. It is actually the argument to MASTER_POS_WAIT(). If the slaves don’t catch up to the master within this time, they will unblock and go ahead with the checksum. You can tell whether this happened by examining the STAT column in the output, which is the return value of MASTER_POS_WAIT().
Do only rows matching this WHERE clause (disallows “–algorithm” CHECKSUM).
You can use this option to limit the checksum to only part of the table. This is particularly useful if you have append-only tables and don’t want to constantly re-check all rows; you could run a daily job to just check yesterday’s rows, for instance.
This option is much like the -w option to mysqldump. Do not specify the WHERE keyword. You may need to quote the value. Here is an example:
pt-table-checksum --where "foo=bar"
Add a chunk for rows with zero or zero-equivalent values. The only has an effect when “–chunk-size” is specified. The purpose of the zero chunk is to capture a potentially large number of zero values that would imbalance the size of the first chunk. For example, if a lot of negative numbers were inserted into an unsigned integer column causing them to be stored as zeros, then these zero values are captured by the zero chunk instead of the first chunk and all its non-zero values.
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-table-checksum ... > 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-table-checksum.
Please report bugs at https://bugs.launchpad.net/percona-toolkit. 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 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.
Claus Jeppesen, Francois Saint-Jacques, Giuseppe Maxia, Heikki Tuuri, James Briggs, Martin Friebe, and Sergey Zhuravlev
ABOUT PERCONA TOOLKIT¶
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.
COPYRIGHT, LICENSE, AND WARRANTY¶
This program is copyright 2007-2011 Baron Schwartz, 2011 Percona Inc. Feedback and improvements are welcome.
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.