I have a MySQL server (Serv-B) that acts as a Slave from a Master server (Serv-A).
This has been configured and is working perfectly fine (the replication has been checked to work).
Now, I'd like this "slave" server to also become the master to another slave server (Serv-C). For that, I do the same commands as for setting up the initial configuration :
I connect to the MySQL server Serv-B, and call the following commands :
FLUSH TABLES WITH READ LOCK;
SHOW MASTER STATUS;
On another terminal, I run the mysqldump command :
mysqldump -u root mydb > /home/cx42/mydb.sql
But it hangs. It hangs until I either close the MySQL terminal, or stop the "Flush" command with UNLOCK TABLES;
As soon as I unlock the tables, the mysqldump commands finish in a few seconds.
It seems that the FLUSH command is locking my tables in a way that mysqldump can't access them, but I don't know what is causing this.
The Serv-B server (the one I'm calling "FLUSH ..") has the following configuration :
server-id = 10
log_bin = /var/log/mysql/mysql-bin.log
expire_logs_days = 7
max_binlog_size = 100M
binlog_do_db = mydb
log_slave_updates = 1
relay-log = /var/log/mysql/mysql-relay-bin.log
What is wrong ?
Yes, it's correct that FLUSH TABLES WITH READ LOCK acquires locks. It should be apparent from the syntax, and also if you were to read the documentation:
FLUSH TABLES tbl_name [, tbl_name] ... WITH READ LOCK
Flushes and acquires read locks for the named tables.
You could release the lock with UNLOCK TABLES, but that means you might not get the right reading from SHOW MASTER STATUS. As soon as you unlock the tables, more updates could be executed, advancing the log position.
The better option is to let mysqldump do the work. If you invoke mysqldump with the --source-data option (or --master-data before version 8.0.26), the output of the dump will include the binary log position, so you don't have to read it in another window.
See https://dev.mysql.com/doc/refman/8.0/en/mysqldump.html#option_mysqldump_source-data
I have a live MySQL database which is configured as a master to a slave. This slave is already replicating from the master. Additionally, the slave is intentionally behind the master by 10 minutes. Now I have a need to take a mysql dump from the master to start another slave.
If I take a dump from the master using the mysqldump --flush-logs option, like so
$ mysqldump --skip-lock-tables --single-transaction --flush-logs --hex-blob --master-data=2 -A > ~/dump.sql
would this be ok? My concerns are:-
Will the bin-log files be flushed (as in purged), thus causing
problems for the existing slave? This slave is relying on the bin-log
files to remain up to date.
Or would this just cause a new bin-log
file to be created leaving the older files intact. Meaning no
problems for the existing slave.
Why even bother with adding --flush-logs ?
I think you have mistaken FLUSH with PURGE. Purpose of flush is to clear and re-load caches or put pending writes to disk. In mysql some writes are done on table close (for example), sometimes you need the data to be on-disk... FLUSH will ensure data is written.
Now "why bother"... in some cases you will want to start replication by dumping SQL and saving log position, so after you import the SQL to the slave you can start from exactly the place on which you took db snapshot to be sure data is not corrupted (eg. by running single query from master - multiple times on the slave).
BTW: --single-transaction without locks is unsafe for any DB which is having writes to myisam tables, you could get databases dumped in different state... and if you already have one slave (which i assume is working correctly), then why not dumping data from the slave using FLUSH TABLES WITH READ LOCK during the whole operation, which is most safe way and always works as intended. It also read-locks the whole server during the dump but if you have working slave anyway - why bother?
I'm using a standard MySQL replica setup with a single master and a single replica. Now I want to create a new replica so that I have one master and two replicas. To setup the new replica, I wanted to do a MySQL dump of the existing replica instead of doing this from the master.
When doing this, I think I first need to stop the replica to get a consistent dataset:
mysql> STOP SLAVE;
Then I would find the replication coordinates so I can start my new replica at the correct replication point:
mysql> SHOW SLAVE STATUS\G
Then I would take the dump by running this command on the replica:
bash> mysqldump --single-transaction --quick --lock-tables=false --insert-ignore --all-databases --master-data=2 | gzip > data-for-replica.sql.gz
and after the MySQL dump is complete, I would restart the replica to let it catch back up:
mysql> START SLAVE;
My question is: since I'm using --single-transaction when doing my dump, can I parallelize these tasks:
Stop replication, check coordinates, and then initiate the MySQL dump
Re-start the replica in a separate shell while the dump is taking place
Would anything go wrong? Would I get inconsistent data and coordinates that would prevent me from creating the new replica in a sane state?
My hope would be to avoid having to wait for the dump to finish for a large database before restarting the replica that the backup is taken from. Since the dump is happening in a single transaction, any updates from the master would not be included in the dump, and hence the dump would be consistent with the coordinates I checked while the replica was stopped. Am I missing anything?
mysqldump is a read only action, not a transaction. So if mysqldump dumps only non-committed data, and your transaction stays open for the duration of the dump, the import in the new slave and the activation of the new slave, then theoretically you could do this. In practice however I think this scenario is unlikely to work.
I've seen this post to re-synchronize a slave database when only one slave is connected to a master DB.
If I understood well, this solution is not the best when we need to resynchronize only 1 slave DB out of batch of slave DBs.
Is there a way to do this, without having to resynchronize all the slave DBs connected to the master ?
I guess the RESET MASTER would affect all the slave DBs state and would lead to resynchronize all of them.
Thank you for your help
Cheers
I found the solution using these mysqldump option to avoid using RESET MASTER way :
mysqldump -p --skip-lock-tables --single-transaction --flush-logs --hex-blob --master-data=2 <dbname> > /tmp/<dbname>.sql
This way, the master log bin and position are not changed, but just added on top of the file .sql.
If you set master-data=2 they will be commented in the file
If you set master-data=1 they will not be commented in the file
Once your dump is on your slave host you can reset the slave, import your dump and then run the mysql command change master to MASTER_LOG_FILE='mysqld-bin.xxxxxxx',MASTER_LOG_POS=xxxx; with the values indicated in the top of your dump file.
Cheers
What in that case with rest of your DBs? If you change binlog and position will they be up to date?
Regards
Robert
Mysql Server1 is running as MASTER.
Mysql Server2 is running as SLAVE.
Now DB replication is happening from MASTER to SLAVE.
Server2 is removed from network and re-connect it back after 1 day. After this there is mismatch in database in master and slave.
How to re-sync the DB again as after restoring DB taken from Master to Slave also doesn't solve the problem ?
This is the full step-by-step procedure to resync a master-slave replication from scratch:
At the master:
RESET MASTER;
FLUSH TABLES WITH READ LOCK;
SHOW MASTER STATUS;
And copy the values of the result of the last command somewhere.
Without closing the connection to the client (because it would release the read lock) issue the command to get a dump of the master:
mysqldump -u root -p --all-databases > /a/path/mysqldump.sql
Now you can release the lock, even if the dump hasn't ended yet. To do it, perform the following command in the MySQL client:
UNLOCK TABLES;
Now copy the dump file to the slave using scp or your preferred tool.
At the slave:
Open a connection to mysql and type:
STOP SLAVE;
Load master's data dump with this console command:
mysql -uroot -p < mysqldump.sql
Sync slave and master logs:
RESET SLAVE;
CHANGE MASTER TO MASTER_LOG_FILE='mysql-bin.000001', MASTER_LOG_POS=98;
Where the values of the above fields are the ones you copied before.
Finally, type:
START SLAVE;
To check that everything is working again, after typing:
SHOW SLAVE STATUS;
you should see:
Slave_IO_Running: Yes
Slave_SQL_Running: Yes
That's it!
The documentation for this at the MySQL site is woefully out of date and riddled with foot-guns (such as interactive_timeout). Issuing FLUSH TABLES WITH READ LOCK as part of your export of the master generally only makes sense when coordinated with a storage/filesystem snapshot such as LVM or zfs.
If you are going to use mysqldump, you should rely instead on the --master-data option to guard against human error and release the locks on the master as quickly as possible.
Assume the master is 192.168.100.50 and the slave is 192.168.100.51, each server has a distinct server-id configured, the master has binary logging on and the slave has read-only=1 in my.cnf
To stage the slave to be able to start replication just after importing the dump, issue a CHANGE MASTER command but omit the log file name and position:
slaveserver> CHANGE MASTER TO MASTER_HOST='192.168.100.50', MASTER_USER='replica', MASTER_PASSWORD='asdmk3qwdq1';
Issue the GRANT on the master for the slave to use:
masterserver> GRANT REPLICATION SLAVE ON *.* TO 'replica'#'192.168.100.51' IDENTIFIED BY 'asdmk3qwdq1';
Export the master (in screen) using compression and automatically capturing the correct binary log coordinates:
mysqldump --master-data --all-databases --flush-privileges | gzip -1 > replication.sql.gz
Copy the replication.sql.gz file to the slave and then import it with zcat to the instance of MySQL running on the slave:
zcat replication.sql.gz | mysql
Start replication by issuing the command to the slave:
slaveserver> START SLAVE;
Optionally update the /root/.my.cnf on the slave to store the same root password as the master.
If you are on 5.1+, it is best to first set the master's binlog_format to MIXED or ROW. Beware that row logged events are slow for tables which lack a primary key. This is usually better than the alternative (and default) configuration of binlog_format=statement (on master), since it is less likely to produce the wrong data on the slave.
If you must (but probably shouldn't) filter replication, do so with slave options replicate-wild-do-table=dbname.% or replicate-wild-ignore-table=badDB.% and use only binlog_format=row
This process will hold a global lock on the master for the duration of the mysqldump command but will not otherwise impact the master.
If you are tempted to use mysqldump --master-data --all-databases --single-transaction (because you only using InnoDB tables), you are perhaps better served using MySQL Enterprise Backup or the open source implementation called xtrabackup (courtesy of Percona)
Unless you are writing directly to the slave (Server2) the only problem should be that Server2 is missing any updates that have happened since it was disconnected. Simply restarting the slave with "START SLAVE;" should get everything back up to speed.
I am very late to this question, however I did encounter this problem and, after much searching, I found this information from Bryan Kennedy: http://plusbryan.com/mysql-replication-without-downtime
On Master take a backup like this:
mysqldump --skip-lock-tables --single-transaction --flush-logs --hex-blob --master-data=2 -A > ~/dump.sql
Now, examine the head of the file and jot down the values for MASTER_LOG_FILE and MASTER_LOG_POS. You will need them later:
head dump.sql -n80 | grep "MASTER_LOG"
Copy the "dump.sql" file over to Slave and restore it:
mysql -u mysql-user -p < ~/dump.sql
Connect to Slave mysql and run a command like this:
CHANGE MASTER TO MASTER_HOST='master-server-ip', MASTER_USER='replication-user', MASTER_PASSWORD='slave-server-password', MASTER_LOG_FILE='value from above', MASTER_LOG_POS=value from above; START SLAVE;
To check the progress of Slave:
SHOW SLAVE STATUS;
If all is well, Last_Error will be blank, and Slave_IO_State will report “Waiting for master to send event”.
Look for Seconds_Behind_Master which indicates how far behind it is.
YMMV. :)
I think, Maatkit utilits helps for you! You can use mk-table-sync. Please see this link: http://www.maatkit.org/doc/mk-table-sync.html
Here is what I typically do when a mysql slave gets out of sync. I have looked at mk-table-sync but thought the Risks section was scary looking.
On Master:
SHOW MASTER STATUS
The outputted columns (File, Position) will be of use to us in a bit.
On Slave:
STOP SLAVE
Then dump the master db and import it to the slave db.
Then run the following:
CHANGE MASTER TO
MASTER_LOG_FILE='[File]',
MASTER_LOG_POS=[Position];
START SLAVE;
Where [File] and [Position] are the values outputted from the "SHOW MASTER STATUS" ran above.
Hope this helps!
Following up on David's answer...
Using SHOW SLAVE STATUS\G will give human-readable output.
Master:
mysqldump -u root -p --all-databases --master-data | gzip > /tmp/dump.sql.gz
scp master:/tmp/dump.sql.gz slave:/tmp/ Move dump file to slave server
Slave:
STOP SLAVE;
zcat /tmp/dump.sql.gz | mysql -u root -p
START SLAVE;
SHOW SLAVE STATUS;
NOTE:
On master you can run SET GLOBAL expire_logs_days = 3 to keep binlogs for 3 days in case of slave issues.
Here is a complete answer that will hopefully help others...
I want to setup mysql replication using master and slave, and since the only thing I knew was that it uses log file(s) to synchronize, if the slave goes offline and gets out of sync, in theory it should only need to connect back to its master and keep reading the log file from where it left off, as user malonso mentioned.
So here are the test result after configuring the master and slave as mentioned by: http://dev.mysql.com/doc/refman/5.0/en/replication-howto.html ...
Provided you use the recommended master/slave configuration and don't write to the slave, he and I where right (as far as mysql-server 5.x is concerned). I didn't even need to use "START SLAVE;", it just caught up to its master. But there is a default 88000 something retries every 60 second so I guess if you exhaust that you might have to start or restart the slave. Anyways, for those like me who wanted to know if having a slave going offline and back up again requires manual intervention.. no, it doesn't.
Maybe the original poster had corruption in the log-file(s)? But most probably not just a server going off-line for a day.
pulled from /usr/share/doc/mysql-server-5.1/README.Debian.gz which probably makes sense to non debian servers as well:
* FURTHER NOTES ON REPLICATION
===============================
If the MySQL server is acting as a replication slave, you should not
set --tmpdir to point to a directory on a memory-based filesystem or to
a directory that is cleared when the server host restarts. A replication
slave needs some of its temporary files to survive a machine restart so
that it can replicate temporary tables or LOAD DATA INFILE operations. If
files in the temporary file directory are lost when the server restarts,
replication fails.
you can use something sql like: show variables like 'tmpdir'; to find out.
Adding to the popular answer to include this error:
"ERROR 1200 (HY000): The server is not configured as slave; fix in config file or with CHANGE MASTER TO",
Replication from slave in one shot:
In one terminal window:
mysql -h <Master_IP_Address> -uroot -p
After connecting,
RESET MASTER;
FLUSH TABLES WITH READ LOCK;
SHOW MASTER STATUS;
The status appears as below: Note that position number varies!
+------------------+----------+--------------+------------------+
| File | Position | Binlog_Do_DB | Binlog_Ignore_DB |
+------------------+----------+--------------+------------------+
| mysql-bin.000001 | 98 | your_DB | |
+------------------+----------+--------------+------------------+
Export the dump similar to how he described "using another terminal"!
Exit and connect to your own DB(which is the slave):
mysql -u root -p
The type the below commands:
STOP SLAVE;
Import the Dump as mentioned (in another terminal, of course!) and type the below commands:
RESET SLAVE;
CHANGE MASTER TO
MASTER_HOST = 'Master_IP_Address',
MASTER_USER = 'your_Master_user', // usually the "root" user
MASTER_PASSWORD = 'Your_MasterDB_Password',
MASTER_PORT = 3306,
MASTER_LOG_FILE = 'mysql-bin.000001',
MASTER_LOG_POS = 98; // In this case
Once logged, set the server_id parameter (usually, for new / non-replicated DBs, this is not set by default),
set global server_id=4000;
Now, start the slave.
START SLAVE;
SHOW SLAVE STATUS\G;
The output should be the same as he described.
Slave_IO_Running: Yes
Slave_SQL_Running: Yes
Note: Once replicated, the master and slave share the same password!
Rebuilding the slave using LVM
Here is the method we use to rebuild MySQL slaves using Linux LVM. This guarantees a consistent snapshot while requiring very minimal downtime on your master.
Set innodb max dirty pages percent to zero on the master MySQL server. This will force MySQL to write all the pages to the disk which will significantly speed up the restart.
set global innodb_max_dirty_pages_pct = 0;
To monitor the number of dirty pages run the command
mysqladmin ext -i10 | grep dirty
Once the number stop decreasing you have reach the point to continue. Next reset the master to clear the old bin logs / relay logs:
RESET MASTER;
Execute lvdisplay to get LV Path
lvdisplay
Output will look like this
--- Logical volume ---
LV Path /dev/vg_mysql/lv_data
LV Name lv_data
VG Name vg_mysql
Shutdown the master database with command
service mysql stop
Next take a snaphot, mysql_snapshot will be the new logical volume name. If binlogs are place on the OS drive those need to be snapshot as well.
lvcreate --size 10G --snapshot --name mysql_snapshot /dev/vg_mysql/lv_data
Start master again with command
service mysql start
Restore dirty pages setting to the default
set global innodb_max_dirty_pages_pct = 75;
Run lvdisplay again to make sure the snapshot is there and visible
lvdisplay
Output:
--- Logical volume ---
LV Path /dev/vg_mysql/mysql_snapshot
LV Name mysql_snapshot
VG Name vg_mysql
Mount the snapshot
mkdir /mnt/mysql_snapshot
mount /dev/vg_mysql/mysql_snapshot /mnt/mysql_snapshot
If you have an existing MySQL slave running you need to stop it
service mysql stop
Next you need to clear MySQL data folder
cd /var/lib/mysql
rm -fr *
Back to master. Now rsync the snapshot to the MySQL slave
rsync --progress -harz /mnt/mysql_snapshot/ targethostname:/var/lib/mysql/
Once rsync has completed you may unmount and remove the snapshot
umount /mnt/mysql_snapshot
lvremove -f /dev/vg_mysql/mysql_snapshot
Create replication user on the master if the old replication user doesn't exist or password is unknown
GRANT REPLICATION SLAVE on *.* to 'replication'#'[SLAVE IP]' identified by 'YourPass';
Verify that /var/lib/mysql data files are owned by the mysql user, if so you can omit the following command:
chown -R mysql:mysql /var/lib/mysql
Next record the binlog position
ls -laF | grep mysql-bin
You will see something like
..
-rw-rw---- 1 mysql mysql 1073750329 Aug 28 03:33 mysql-bin.000017
-rw-rw---- 1 mysql mysql 1073741932 Aug 28 08:32 mysql-bin.000018
-rw-rw---- 1 mysql mysql 963333441 Aug 28 15:37 mysql-bin.000019
-rw-rw---- 1 mysql mysql 65657162 Aug 28 16:44 mysql-bin.000020
Here the master log file is the highest file number in sequence and bin log position is the file size. Record these values:
master_log_file=mysql-bin.000020
master_log_post=65657162
Next start the slave MySQL
service mysql start
Execute change master command on the slave by executing the following:
CHANGE MASTER TO
master_host="10.0.0.12",
master_user="replication",
master_password="YourPass",
master_log_file="mysql-bin.000020",
master_log_pos=65657162;
Finally start the slave
SLAVE START;
Check slave status:
SHOW SLAVE STATUS;
Make sure Slave IO is running and there are no connection errors. Good luck!
I recently wrote this on my blog which is found here... There are few more details there but the story is the same.
http://www.juhavehnia.com/2015/05/rebuilding-mysql-slave-using-linux-lvm.html
I created a GitHub repo with an script to solve this problem quickly. Just change a couple variables and run it (First, the script creates a backup of your database).
I hope this help you (and others people too).
How to Reset (Re-Sync) MySQL Master-Slave Replication
sometimes you just need to give the slave a kick too
try
stop slave;
reset slave;
start slave;
show slave status;
quite often, slaves, they just get stuck guys :)
We are using master-master replication technique of MySQL and if one MySQL server say 1 is removed from the network it reconnects itself after the connection are restored and all the records that were committed in the in the server 2 which was in the network are transferred to the server 1 which has lost the connection after restoration.
Slave thread in the MySQL retries to connect to its master after every 60 sec by default. This property can be changed as MySQL ha a flag "master_connect_retry=5" where 5 is in sec. This means that we want a retry after every 5 sec.
But you need to make sure that the server which lost the connection show not make any commit in the database as you get duplicate Key error Error code: 1062