Zabbix历史数据库迁移 及分区

公司zabbix用了很久了，大概有80G的历史数据。机器内存也不是多大，所以决定迁移zabbix后台数据库，并在新机器上开启分区。
zabbix数据库里 主要的大表为以下几个, 系统同事又提出需要保存trends两个表的数据：

history，history_text,history_log,history_str,history_unit (迁移时不需要保留）
trends, trend_unit (迁移时需要保留数据）

操作步骤：

1 . 在新机器上搭好mysql。
2 . 在老机器上上 dump出一份只有表定义的文件，传到新库：

mysqldump -uroot -p zabbix -d  > zabbix_d.sql

3 . 在老机器上上 dump出表内容文件，这个文件要排除掉一些不需要迁移的历史数据表：

mysqldump -uroot -p zabbix -t --ignore-table=zabbix.history_log --ignore-table=zabbix.history --ignore-table=zabbix.history_text --ignore-table=zabbix.history_uint --ignore-table=zabbix.history_str >zabbix_t.sql

这步骤可能需要一定时间，因为trends的那两个表也不小，我们同时在新库上继续操作。

后续确认 trends的两个表也可以不要清楚数据。

4 .新机器上修改两个表的结构添加索引：

use zabbix
truncate table history;
optimize  table history;
truncate table history_uint;
truncate table history_str;

alter table history_text drop primary key, add index (id), drop index history_text_2, add index history_text_2 (itemid, id);
alter table history_log drop primary key, add index (id), drop index history_log_2, add index history_log_2 (itemid, id);

5 .添加自动分区的存储过程：

DELIMITER $$
CREATE PROCEDURE `partition_create`(SCHEMANAME varchar(64), TABLENAME varchar(64), PARTITIONNAME varchar(64), CLOCK int)
BEGIN
        /*
           SCHEMANAME = The DB schema in which to make changes
           TABLENAME = The table with partitions to potentially delete
           PARTITIONNAME = The name of the partition to create
        */
        /*
           Verify that the partition does not already exist
        */

        DECLARE RETROWS INT;
        SELECT COUNT(1) INTO RETROWS
        FROM information_schema.partitions
        WHERE table_schema = SCHEMANAME AND table_name = TABLENAME AND partition_description >= CLOCK;

        IF RETROWS = 0 THEN
                /*
                   1. Print a message indicating that a partition was created.
                   2. Create the SQL to create the partition.
                   3. Execute the SQL from #2.
                */
                SELECT CONCAT( "partition_create(", SCHEMANAME, ",", TABLENAME, ",", PARTITIONNAME, ",", CLOCK, ")" ) AS msg;
                SET @sql = CONCAT( 'ALTER TABLE ', SCHEMANAME, '.', TABLENAME, ' ADD PARTITION (PARTITION ', PARTITIONNAME, ' VALUES LESS THAN (', CLOCK, '));' );
                PREPARE STMT FROM @sql;
                EXECUTE STMT;
                DEALLOCATE PREPARE STMT;
        END IF;
END$$
DELIMITER ;

DELIMITER $$
CREATE PROCEDURE `partition_drop`(SCHEMANAME VARCHAR(64), TABLENAME VARCHAR(64), DELETE_BELOW_PARTITION_DATE BIGINT)
BEGIN
        /*
           SCHEMANAME = The DB schema in which to make changes
           TABLENAME = The table with partitions to potentially delete
           DELETE_BELOW_PARTITION_DATE = Delete any partitions with names that are dates older than this one (yyyy-mm-dd)
        */
        DECLARE done INT DEFAULT FALSE;
        DECLARE drop_part_name VARCHAR(16);

        /*
           Get a list of all the partitions that are older than the date
           in DELETE_BELOW_PARTITION_DATE.  All partitions are prefixed with
           a "p", so use SUBSTRING TO get rid of that character.
        */
        DECLARE myCursor CURSOR FOR
                SELECT partition_name
                FROM information_schema.partitions
                WHERE table_schema = SCHEMANAME AND table_name = TABLENAME AND CAST(SUBSTRING(partition_name FROM 2) AS UNSIGNED) < DELETE_BELOW_PARTITION_DATE;
        DECLARE CONTINUE HANDLER FOR NOT FOUND SET done = TRUE;

        /*
           Create the basics for when we need to drop the partition.  Also, create
           @drop_partitions to hold a comma-delimited list of all partitions that
           should be deleted.
        */
        SET @alter_header = CONCAT("ALTER TABLE ", SCHEMANAME, ".", TABLENAME, " DROP PARTITION ");
        SET @drop_partitions = "";

        /*
           Start looping through all the partitions that are too old.
        */
        OPEN myCursor;
        read_loop: LOOP
                FETCH myCursor INTO drop_part_name;
                IF done THEN
                        LEAVE read_loop;
                END IF;
                SET @drop_partitions = IF(@drop_partitions = "", drop_part_name, CONCAT(@drop_partitions, ",", drop_part_name));
        END LOOP;
        IF @drop_partitions != "" THEN
                /*
                   1. Build the SQL to drop all the necessary partitions.
                   2. Run the SQL to drop the partitions.
                   3. Print out the table partitions that were deleted.
                */
                SET @full_sql = CONCAT(@alter_header, @drop_partitions, ";");
                PREPARE STMT FROM @full_sql;
                EXECUTE STMT;
                DEALLOCATE PREPARE STMT;

                SELECT CONCAT(SCHEMANAME, ".", TABLENAME) AS `table`, @drop_partitions AS `partitions_deleted`;
        ELSE
                /*
                   No partitions are being deleted, so print out "N/A" (Not applicable) to indicate
                   that no changes were made.
                */
                SELECT CONCAT(SCHEMANAME, ".", TABLENAME) AS `table`, "N/A" AS `partitions_deleted`;
        END IF;
END$$
DELIMITER ;

DELIMITER $$
CREATE PROCEDURE `partition_maintenance`(SCHEMA_NAME VARCHAR(32), TABLE_NAME VARCHAR(32), KEEP_DATA_DAYS INT, HOURLY_INTERVAL INT, CREATE_NEXT_INTERVALS INT)
BEGIN
        DECLARE OLDER_THAN_PARTITION_DATE VARCHAR(16);
        DECLARE PARTITION_NAME VARCHAR(16);
        DECLARE OLD_PARTITION_NAME VARCHAR(16);
        DECLARE LESS_THAN_TIMESTAMP INT;
        DECLARE CUR_TIME INT;

        CALL partition_verify(SCHEMA_NAME, TABLE_NAME, HOURLY_INTERVAL);
        SET CUR_TIME = UNIX_TIMESTAMP(DATE_FORMAT(NOW(), '%Y-%m-%d 00:00:00'));

        SET @__interval = 1;
        create_loop: LOOP
                IF @__interval > CREATE_NEXT_INTERVALS THEN
                        LEAVE create_loop;
                END IF;

                SET LESS_THAN_TIMESTAMP = CUR_TIME + (HOURLY_INTERVAL * @__interval * 3600);
                SET PARTITION_NAME = FROM_UNIXTIME(CUR_TIME + HOURLY_INTERVAL * (@__interval - 1) * 3600, 'p%Y%m%d%H00');
                IF(PARTITION_NAME != OLD_PARTITION_NAME) THEN
      CALL partition_create(SCHEMA_NAME, TABLE_NAME, PARTITION_NAME, LESS_THAN_TIMESTAMP);
    END IF;
                SET @__interval=@__interval+1;
                SET OLD_PARTITION_NAME = PARTITION_NAME;
        END LOOP;

        SET OLDER_THAN_PARTITION_DATE=DATE_FORMAT(DATE_SUB(NOW(), INTERVAL KEEP_DATA_DAYS DAY), '%Y%m%d0000');
        CALL partition_drop(SCHEMA_NAME, TABLE_NAME, OLDER_THAN_PARTITION_DATE);

END$$
DELIMITER ;

DELIMITER $$
CREATE PROCEDURE `partition_verify`(SCHEMANAME VARCHAR(64), TABLENAME VARCHAR(64), HOURLYINTERVAL INT(11))
BEGIN
        DECLARE PARTITION_NAME VARCHAR(16);
        DECLARE RETROWS INT(11);
        DECLARE FUTURE_TIMESTAMP TIMESTAMP;

        /*
         * Check if any partitions exist for the given SCHEMANAME.TABLENAME.
         */
        SELECT COUNT(1) INTO RETROWS
        FROM information_schema.partitions
        WHERE table_schema = SCHEMANAME AND table_name = TABLENAME AND partition_name IS NULL;

        /*
         * If partitions do not exist, go ahead and partition the table
         */
        IF RETROWS = 1 THEN
                /*
                 * Take the current date at 00:00:00 and add HOURLYINTERVAL to it.  This is the timestamp below which we will store values.
                 * We begin partitioning based on the beginning of a day.  This is because we don't want to generate a random partition
                 * that won't necessarily fall in line with the desired partition naming (ie: if the hour interval is 24 hours, we could
                 * end up creating a partition now named "p201403270600" when all other partitions will be like "p201403280000").
                 */
                SET FUTURE_TIMESTAMP = TIMESTAMPADD(HOUR, HOURLYINTERVAL, CONCAT(CURDATE(), " ", '00:00:00'));
                SET PARTITION_NAME = DATE_FORMAT(CURDATE(), 'p%Y%m%d%H00');

                -- Create the partitioning query
                SET @__PARTITION_SQL = CONCAT("ALTER TABLE ", SCHEMANAME, ".", TABLENAME, " PARTITION BY RANGE(`clock`)");
                SET @__PARTITION_SQL = CONCAT(@__PARTITION_SQL, "(PARTITION ", PARTITION_NAME, " VALUES LESS THAN (", UNIX_TIMESTAMP(FUTURE_TIMESTAMP), "));");

                -- Run the partitioning query
                PREPARE STMT FROM @__PARTITION_SQL;
                EXECUTE STMT;
                DEALLOCATE PREPARE STMT;
        END IF;
END$$
DELIMITER ;

6 . 把存储过程导入数据库：

use zabbix
source zabbix_par.sql;

7 . 创建一个批量调用产生分区的存储过程

cat zabbix_cal.sql

DELIMITER $$
CREATE PROCEDURE `partition_maintenance_all`(SCHEMA_NAME VARCHAR(32))
BEGIN
                CALL partition_maintenance(SCHEMA_NAME, 'history', 28, 24, 14);
                CALL partition_maintenance(SCHEMA_NAME, 'history_log', 28, 24, 14);
                CALL partition_maintenance(SCHEMA_NAME, 'history_str', 28, 24, 14);
                CALL partition_maintenance(SCHEMA_NAME, 'history_text', 28, 24, 14);
                CALL partition_maintenance(SCHEMA_NAME, 'history_uint', 28, 24, 14);
                CALL partition_maintenance(SCHEMA_NAME, 'trends', 730, 24, 14);
                CALL partition_maintenance(SCHEMA_NAME, 'trends_uint', 730, 24, 14);
END$$
DELIMITER ;

8 . 把这个存储过程载入zabbix库

mysql -uzabbix -pzabbix zabbix  <  zabbix_cal.sql

CALL partition_maintenance(SCHEMA_NAME, ‘history_text’, 28, 24, 14);
保留28天数据， 每24小时创建一次分区，每次创建14个分区。
修改的时候尽量改28. 和 14 ，中间的24不要修改。

9 . 以后每天定期调度这个存储过程：

crontab -l

00 01 * * * /usr/local/bin/mysql -uzabbix -pzabbix zabbix -e "CALL partition_maintenance_all('zabbix');"

其它重要参考：
http://blog.sina.com.cn/s/blog_4cbf97060101fcfw.html