pt-table-checksum是percona公司提供的一个用于在线比对主从数据一致性的工具。
实现原理
将一张大表分成多个chunk，每次针对一个chunk进行校验，同时将校验的结果通过REPLACE INTO语句写入到percona.checksums表中，然后该语句通过主从复制，在SLAVE中同样执行一次，校验的结果同样是写入到percona.checksums表中，最后，通过查询percona.checksums来获取主从不一致的信息。
常见用法
1. 基本用法
# pt-table-checksum -h192.168.244.10 -umonitor -pmonitor123
其中，monitor的最小权限如下（第二个权限是针对percona.checksums的）：
GRANT SELECT, PROCESS, SUPER, REPLICATION SLAVE, REPLICATION CLIENT ON *.* TO 'monitor'@'192.168.244.10';
GRANT ALL PRIVILEGES ON `percona`.* TO 'monitor'@'192.168.244.10';
2. pt-table-checksum默认是运行在statement下，如果是其它日志格式，需加--no-check-binlog-format参数
# pt-table-checksum -h192.168.244.10 -umonitor -pmonitor123 --no-check-binlog-format
3. 如果主从复制中加了过滤条件，譬如binlog_ignore_db或replicate_do_db之类的参数，需加--no-check-replication-filters参数
# pt-table-checksum -h192.168.244.10 -umonitor -pmonitor123 --no-check-binlog-format --no-check-replication-filters
如果在对被过滤表进行校验时，命令hang住了，可加--replicate-database参数。
4. 基于指定库的校验
# pt-table-checksum -h192.168.244.10 -umonitor -pmonitor123 --no-check-binlog-format --databases=test,test1
5. 基于指定表的校验
# pt-table-checksum -h192.168.244.10 -umonitor -pmonitor123 --no-check-binlog-format --tables=test2.test
其它具体用法，可参考另外一篇博客：pt-table-checksum参数详解
通过打开general_log来看看其具体的执行过程，注意，测试表是test.test，共1000000条记录。
17 Query SHOW GLOBAL STATUS LIKE 'Threads_running'
查看Threads_running变量是为了查看当前系统的负载情况
17 Query /*!40101 SET @OLD_SQL_MODE := @@SQL_MODE, @@SQL_MODE := '', @OLD_QUOTE := @@SQL_QUOTE_SHOW_CREATE, @@SQL_QUOTE_SHOW_CREATE := 1 */
设置会话变量
17 Query USE `test` 17 Query SHOW CREATE TABLE `test`.`test`
查看test表的表结构，选取分片键，一般为主键或唯一索引 17 Query /*!40101 SET @@SQL_MODE := @OLD_SQL_MODE, @@SQL_QUOTE_SHOW_CREATE := @OLD_QUOTE */ 17 Query EXPLAIN SELECT * FROM `test`.`test` WHERE 1=1
查看test表的大概数量 17 Query SELECT /*!40001 SQL_NO_CACHE */ `id` FROM `test`.`test` FORCE INDEX(`PRIMARY`) ORDER BY `id` LIMIT 1 /*first lower boundary*/
选择第一个chunk的下标，即id的最小值 17 Query SELECT /*!40001 SQL_NO_CACHE */ `id` FROM `test`.`test` FORCE INDEX (`PRIMARY`) WHERE `id` IS NOT NULL ORDER BY `id` LIMIT 1 /*key_len*/
查看索引的长度 17 Query EXPLAIN SELECT /*!40001 SQL_NO_CACHE */ * FROM `test`.`test` FORCE INDEX (`PRIMARY`) WHERE `id` >= '1' /*key_len*/
查看实际使用的索引的长度，这个针对联合索引的场景。 17 Query USE `percona` 17 Query DELETE FROM `percona`.`checksums` WHERE db = 'test' AND tbl = 'test'
从percona.checksums表中删除之前的校验记录 17 Query USE `test` 17 Query EXPLAIN SELECT /*!40001 SQL_NO_CACHE */ `id` FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '1')) ORDER BY `id` LIMIT 2171, 2 /*next chunk boundary*/
17 Query SELECT /*!40001 SQL_NO_CACHE */ `id` FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '1')) ORDER BY `id` LIMIT 2171, 2 /*next chunk boundary*/ 确认本chunk的上限，以及下一个chunk的下限。
17 Query EXPLAIN SELECT COUNT(*) AS cnt, COALESCE(LOWER(CONV(BIT_XOR(CAST(CRC32(CONCAT_WS('#', `id`, `ip`, `no`, CONCAT(ISNULL(`ip`), ISNULL(`no`)))) AS UNSIGNED)), 10, 16)), 0) AS crc FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '1')) AND ((`id` <= '2172')) /*explain checksum chunk*/ 查看对本次chunk执行checksum操作的执行计划，确认读取的行数是否合理，选择的索引是否合适
17 Query REPLACE INTO `percona`.`checksums` (db, tbl, chunk, chunk_index, lower_boundary, upper_boundary, this_cnt, this_crc) SELECT 'test', 'test', '1', 'PRIMARY', '1', '2172', COUNT(*) AS cnt, COALESCE(LOWER(CONV(BIT_XOR(CAST(CRC32(CONCAT_WS('#', `id`, `ip`, `no`, CONCAT(ISNULL(`ip`), ISNULL(`no`)))) AS UNSIGNED)), 10, 16)), 0) AS crc FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '1')) AND ((`id` <= '2172')) /*checksum chunk*/ 关键操作，对本次chunk执行checksum操作，并将结果更新到percona.checksums表中。
17 Query SHOW WARNINGS 17 Query SELECT this_crc, this_cnt FROM `percona`.`checksums` WHERE db = 'test' AND tbl = 'test' AND chunk = '1'
查看本次操作校验的行数和校验和 17 Query UPDATE `percona`.`checksums` SET chunk_time = '0.059603', master_crc = '568c1ba3', master_cnt = '2172' WHERE db = 'test' AND tbl = 'test' AND chunk = '1' 将上面那个查询得到的行数和校验和更新到master_cnt和master_crc中。这样的话，主库的校验和在从库执行replace操作时被覆盖。
下面是针对第二个chunk执行的操作。 17 Query SHOW GLOBAL STATUS LIKE 'Threads_running' 17 Query EXPLAIN SELECT /*!40001 SQL_NO_CACHE */ `id` FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '2173')) ORDER BY `id` LIMIT 18219, 2 /*next chunk boundary*/ 17 Query SELECT /*!40001 SQL_NO_CACHE */ `id` FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '2173')) ORDER BY `id` LIMIT 18219, 2 /*next chunk boundary*/ 17 Query EXPLAIN SELECT COUNT(*) AS cnt, COALESCE(LOWER(CONV(BIT_XOR(CAST(CRC32(CONCAT_WS('#', `id`, `ip`, `no`, CONCAT(ISNULL(`ip`), ISNULL(`no`)))) AS UNSIGNED)), 10, 16)), 0) AS crc FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '2173')) AND ((`id` <= '20392')) /*explain checksum chunk*/ 17 Query REPLACE INTO `percona`.`checksums` (db, tbl, chunk, chunk_index, lower_boundary, upper_boundary, this_cnt, this_crc) SELECT 'test', 'test', '2', 'PRIMARY', '2173', '20392', COUNT(*) AS cnt, COALESCE(LOWER(CONV(BIT_XOR(CAST(CRC32(CONCAT_WS('#', `id`, `ip`, `no`, CONCAT(ISNULL(`ip`), ISNULL(`no`)))) AS UNSIGNED)), 10, 16)), 0) AS crc FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` >= '2173')) AND ((`id` <= '20392')) /*checksum chunk*/ 17 Query SHOW WARNINGS 17 Query SELECT this_crc, this_cnt FROM `percona`.`checksums` WHERE db = 'test' AND tbl = 'test' AND chunk = '2' 17 Query UPDATE `percona`.`checksums` SET chunk_time = '0.022960', master_crc = '83371365', master_cnt = '18220' WHERE db = 'test' AND tbl = 'test' AND chunk = '2' ... 下面的校验和上面的并不相同，上述id值的范围是1~1000000，下面两个chunk的范围是<1和>1000000,为什么要这么做呢?
主要是考虑到从库有可能存在上述两个范围的数据。
17 Query SHOW GLOBAL STATUS LIKE 'Threads_running' 17 Query EXPLAIN SELECT COUNT(*), '0' FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` < '1')) ORDER BY `id` /*explain past lower chunk*/ 17 Query REPLACE INTO `percona`.`checksums` (db, tbl, chunk, chunk_index, lower_boundary, upper_boundary, this_cnt, this_crc) SELECT 'test', 'test', '9', 'PRIMARY', NULL, '1', COUNT(*), '0' FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` < '1')) ORDER BY `id` /*past lower chunk*/ 17 Query SHOW WARNINGS 17 Query SELECT this_crc, this_cnt FROM `percona`.`checksums` WHERE db = 'test' AND tbl = 'test' AND chunk = '9' 17 Query UPDATE `percona`.`checksums` SET chunk_time = '0.004492', master_crc = '0', master_cnt = '0' WHERE db = 'test' AND tbl = 'test' AND chunk = '9' 17 Query SHOW GLOBAL STATUS LIKE 'Threads_running' 17 Query EXPLAIN SELECT COUNT(*), '0' FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` > '1000000')) ORDER BY `id` /*explain past upper chunk*/ 17 Query REPLACE INTO `percona`.`checksums` (db, tbl, chunk, chunk_index, lower_boundary, upper_boundary, this_cnt, this_crc) SELECT 'test', 'test', '10', 'PRIMARY', '1000000', NULL, COUNT(*), '0' FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` > '1000000')) ORDER BY `id` /*past upper chunk*/ 17 Query SHOW WARNINGS 17 Query SELECT this_crc, this_cnt FROM `percona`.`checksums` WHERE db = 'test' AND tbl = 'test' AND chunk = '10' 17 Query UPDATE `percona`.`checksums` SET chunk_time = '0.058622', master_crc = '0', master_cnt = '0' WHERE db = 'test' AND tbl = 'test' AND chunk = '10'
输出结果说明
# pt-table-checksum -h 192.168.244.10 -umonitor -pmonitor123 --no-check-binlog-format
TS ERRORS DIFFS ROWS CHUNKS SKIPPED TIME TABLE 11-09T21:19:01 0 0 0 1 0 2.768 h2.h3 11-09T21:19:06 0 0 0 1 0 3.903 hello.h1 11-09T21:19:06 0 0 0 1 0 0.620 hello.h2 11-09T21:19:10 0 0 0 1 0 1.725 mysql.columns_priv 11-09T21:19:10 0 1 4 1 0 0.457 mysql.db 11-09T21:19:11 0 0 0 1 0 0.306 mysql.event 11-09T21:19:12 0 0 0 1 0 0.721 mysql.func 11-09T21:19:13 0 0 40 1 0 0.649 mysql.help_category 11-09T21:19:14 0 0 611 6 0 1.313 mysql.help_keyword 11-09T21:19:18 0 0 1218 4 0 3.432 mysql.help_relation 11-09T21:19:20 0 0 583 1 0 1.632 mysql.help_topic 11-09T21:19:21 0 0 0 1 0 0.501 mysql.ndb_binlog_index 11-09T21:19:21 0 0 0 1 0 0.319 mysql.plugin 11-09T21:19:22 0 0 0 1 0 0.624 mysql.proc 11-09T21:19:22 0 0 0 1 0 0.626 mysql.procs_priv 11-09T21:19:23 0 0 2 1 0 0.375 mysql.proxies_priv 11-09T21:19:24 0 0 0 1 0 0.806 mysql.servers 11-09T21:19:24 0 1 3 1 0 0.292 mysql.tables_priv 11-09T21:19:24 0 0 0 1 0 0.382 mysql.time_zone 11-09T21:19:25 0 0 0 1 0 0.398 mysql.time_zone_leap_second 11-09T21:19:25 0 0 0 1 0 0.386 mysql.time_zone_name 11-09T21:19:25 0 0 0 1 0 0.393 mysql.time_zone_transition 11-09T21:19:26 0 0 0 1 0 0.313 mysql.time_zone_transition_type 11-09T21:19:26 0 1 18 1 0 0.393 mysql.user 11-09T21:19:27 0 0 1 1 0 0.642 percona.dsns 11-09T21:19:30 0 1 41 1 0 2.613 test.checksum 11-09T21:19:31 0 0 0 1 0 0.669 test.ta 11-09T21:20:13 0 1 1000000 7 0 42.009 test.test 11-09T21:20:15 0 0 5 1 0 1.818 test1.test 11-09T21:20:16 0 0 5 1 0 0.546 test1.test1 11-09T21:20:17 0 1 12 1 0 0.508 test2.test 11-09T21:20:17 0 1 0 1 0 0.387 test2.test2 11-09T21:20:18 0 1 1000 1 0 0.707 test3.test3
TS：校验完表后的时间戳
ERRORS：校验过程中出现的errors和warnings的次数。
DIFFS：所有SLAVE中checksum值不相同的chunk的数量，如一主两从中，SLAVE1的chunk1与MASTER的checksum不同，SLAVE2的chunk1和chunk2不相同，则DIFFS的值为2。如果SLAVE2的chunk2和chunk3不相同，则DIFFS为3。
ROWS：表中校验的记录数。通常情况下为表的总行数。如果指定了--where选项，则为符合条件的记录数。
CHUNKS：表被分割为多个chunk后，chunk的个数。
SKIPPED：跳过的chunk的个数，通常因为如下原因：
* MySQL not using the --chunk-index * MySQL not using the full chunk index (--[no]check-plan) * Chunk size is greater than --chunk-size * --chunk-size-limit * Lock wait timeout exceeded (--retries) * Checksum query killed (--retries)
TIME：校验表所花费的时间。
TABLE：校验的表名
replicate-check-only参数下的输出结果说明
如果指定了--replicate-check-only参数，则意味着不会校验任何表，直接获取上次校验的结果。
# pt-table-checksum -h 192.168.244.10 -umonitor -pmonitor123 --no-check-binlog-format --replicate-check-only
Differences on hbase TABLE CHUNK CNT_DIFF CRC_DIFF CHUNK_INDEX LOWER_BOUNDARY UPPER_BOUNDARY mysql.db 1 1 1 mysql.tables_priv 1 -3 1 mysql.user 1 -4 1 test.checksum 1 0 1 Differences on test TABLE CHUNK CNT_DIFF CRC_DIFF CHUNK_INDEX LOWER_BOUNDARY UPPER_BOUNDARY mysql.db 1 -1 1 mysql.tables_priv 1 -1 1 mysql.user 1 -2 1 test.checksum 1 0 1 test.test 7 1000000 0 PRIMARY 1000000 test2.test 1 -10 1 test2.test2 1 1 1 test3.test3 1 2000 1
可以看出，它分别输出了不同SLAVE中的差异部分。
TABLE：校验的表名。
CHUNK：checksum值不相同的chunk的数量。
CNT_DIFF：The number of chunk rows on the replica minus the number of chunk rows on the master.即SLAVE中被校验的记录数减去MASTER中的记录数。
CRC_DIFF：1 if the CRC of the chunk on the replica is different than the CRC of the chunk on the master, else 0.如果校验值相同，则CRC_DIFF为0，否则为1。
上述test.test中CRC_DIFF中crc为0的原因是SLAVE比MASTER多1000000条记录，且这1000000条记录正好又是在最后一个chunk中（如下所示）。注：master中id最大值为1000000。
319 Query REPLACE INTO `percona`.`checksums` (db, tbl, chunk, chunk_index, lower_boundary, upper_boundary, this_cnt, this_crc) SELECT 'test', 'test', '7', 'PRIMARY', '1000000', NULL, COUNT(*), '0' FROM `test`.`test` FORCE INDEX(`PRIMARY`) WHERE ((`id` > '1000000')) ORDER BY `id` /*past upper chunk*/
CHUNK_INDEX：用于将table切割成chunk的索引。
LOWER_BOUNDARY：The index values that define the lower boundary of the chunk.
UPPER_BOUNDARY：The index values that define the upper boundary of the chunk.
上述两个参数可用来定位具有不同checksum值的chunk对应的索引的上限和下限。
总结
1. pt-table-checksum对表进行校验时，并不需要表上面有任何索引。这时候，整张表即是一个chunk。
REPLACE INTO `percona`.`checksums` (db, tbl, chunk, chunk_index, lower_boundary, upper_boundary, this_cnt, this_crc) SELECT 'test2', 't1', '1', NULL, NULL, NULL, COUNT(*) AS cnt, COALESCE(LOWER(CONV(BIT_XOR(CAST(CRC32(CONCAT_WS('#', `id`, `name`, CONCAT(ISNULL(`id`), ISNULL(`name`)))) AS UNSIGNED)), 10, 16)), 0) AS crc FROM `test2`.`t1` /*checksum table*/
如果表的行过多的话，它会报如下错误：
# pt-table-checksum -h192.168.244.10 -umonitor -pmonitor123 --chunk-size-limit=100 --no-check-binlog-format --tables=test2.t1
11-17T11:41:53 Cannot checksum table test2.t1: There is no good index and the table is oversized. at /usr/local/bin/pt-table-checksum line 6528.
解决方法：
调整chunk-size-limit的值
# pt-table-checksum -h192.168.244.10 -umonitor -pmonitor123 --chunk-size-limit=1000 --no-check-binlog-format --tables=test2.t1
TS ERRORS DIFFS ROWS CHUNKS SKIPPED TIME TABLE 11-17T11:41:59 0 0 100004 1 0 0.600 test2.t1