Home / server / Questions / 275249
Accepted
Fulvio Scapin
Asked: 2011-05-31 06:36:43 +0800 CST

mk-table-sync in a master-slave scenario: Changes not replicated to the slave

  • 772

I have being employing mk-table-sync to synchronise tables from a master to slave on mysql 5.1. Unfortunately, while differences are correctly detected, modifications done on the master (DELETE,REPLACE,ecc.) do not seem to be propagated to the slave. SHOW SLAVE STATUS doesn't reveal connection issues.

Basically, doing

mk-table-sync -v  --execute --databases=forum --sync-to-master 
h=localhost,D=forum,t=user
# Syncing D=forum,h=localhost,t=user
# DELETE REPLACE INSERT UPDATE ALGORITHM START    END      EXIT DATABASE.TABLE
#      0       7      0      0 Chunk     14:35:00 14:35:01 2    forum.user

repeatedly gives always the same results, without an actual change to the slave.

Log on the slave:

http://pastebin.com/kxuxks1P

Log on the master:

http://pastebin.com/kVjEWEdL

Same goes for DELETEs done on the master and for every other table in the replicated db.

Any ideas?

Thanks in advance

mysql mysql-replication master-slave maatkit

2 Answers

  1. Best Answer

    In all honesty, I have never trusted the --execute parameter for mk-table-sync. I always use --print instead.

    Replace this

    mk-table-sync -v  --execute --databases=forum --sync-to-master h=localhost,D=forum,t=user

    with this if you have binary logging enabled

    echo "SET SQL_LOG_BIN=0;" > DBChanges.sql
mk-table-sync -v  --print --databases=forum --sync-to-master h=localhost,D=forum,t=user >> DBChanges.sql

    or this if the slave does not have binary logging

    mk-table-sync -v  --print --databases=forum --sync-to-master h=localhost,D=forum,t=user > DBChanges.sql

    This way, you can see the actual SQL to run safely on the slave.

    UPDATE 2011-05-31 12:57

    "Interesting. Correct me if I am wrong, but shouldn't the queries run on the master be propagated to the slave through replication? I don't quite understand why that doesn't happen"

    That's a fair question. Yet, think of the way MySQL Replication works. When an SQL statement is completed on a master, it is recorded in the master's binary logs. The I/O thread of the slave reads any new entries in the master's binary logs and appends them to the last of the slave's relay logs. The slave's SQL thread reads the relay log entries as a FIFO queue and processes the SQL statements in the order of their recording. If the slave has log-slave-updates and log-bin in its configuration, the SQL statament upon completion will be recorded in the slave's binary logs.

    Enough small talk on MySQL Replication.

    Now, why would a master not replicate to a slave ???

    Here are some possibilities for you to explore:

    POSSIBILITY #1 : Network latency causing binlog entries from the master not to propogate over to the relay logs of the slave in a timely manner or not at all.

    POSSIBILITY #2 : MySQL packets are too small and errors being ignored. This could happen only in the following scenario: The max_allowed_packet in the master is bigger than the max_allowed_packet in the slave. This would normally stop replication cold in its tracks. If you are skipping all slave errors (if you have slave-skip-errors=all in /etc/my.cnf) then various kinds of normal data can be ignored in this unique scenario.

    POSSIBILITY #3 : Config to skip any duplicate key error in the slave's SQL thread

    [mysqld]
slave-skip-errors=1062 (skips duplicate key errors)

    POSSIBILITY #4 : Config to skip any SQL error in the slave's SQL thread

    [mysqld]
slave-skip-errors=all (skips every SQL error under the sun)

    POSSIBILITY #5 : Having the slave's I/O thread simply die without telling mysqld. This can happen. Simple correction? Do the following to reestablish the slave I/O thread:

    STOP SLAVE IO_THREAD;
START SLAVE IO_THREAD;

    POSSIBILITY #6 : Having a wrong combination of replicate-do-db and replicate-ignore-db in /etc/my.cnf (Disclaimer : This is strictly my opinion)

    Some mix both options in /etc/my.cnf and think nothing of it. IMHO, these options are supposed to be mutually exclusive. You follow the logic of filtering out data or filtering in data in a slave. They should not be used together for you can get spurious results from replication. Either data should be there or not, NOT data should be there and not.

    • 4

  2. It is possible that you are using a version of mk-table-sync that doesn't set the binlog format to STATEMENT, therefore not really changing any data on the master (as designed) and thus nothing goes into the binary log.

    • 3