mysql复制错误：候选从属缺少复制用户

Question

Context: I'm trying to set up mysql replication with one master and one slave. I'm using the new GTID method in mysql 5.6, along with the mysql utilities for automatic failover.

I'm able to set up replication with a master node M and a slave node S.

I'm trying to run through a simple script to show myself that everything is set up correctly. The script is as follows:

1. Set up replication with node M as master and node S as slave
2. Ensure that changes made to tables in M are reflected in S
3. Shutdown M and ensure that S takes over as new master
4. Make a few table changes to S
5. Startup M and add it back to cluster as a slave
6. Ensure that M catches up to S with its updates
7. Make M the master node again

My script seems successful until the last step. Largely, I'm following the steps here: http://www.clusterdb.com/mysql/replication-and-auto-failover-made-easy-with-mysql-utilities

But when I run "mysqlrpladmin --master=root@S_NODE --new-master=root@M_NODE --demote-master --discover-slaves-login=root switchover", I'm getting the following error:

# Discovering slaves for master at S_NODE
# Discovering slave at M_NODE
# Found slave: M_NODE
# Checking privileges.
# Performing switchover from master at S_NODE to slave at M_NODE
# Checking candidate slave prerequisites.
Candidate slave is missing replication user.
ERROR: Candidate slave is missing replication user.
# Errors found. Switchover aborted.
#
# Replication Topology Health:
+-----------------+-------+---------+--------+------------+---------+
| host            | port  | role    | state  | gtid_mode  | health  |
+-----------------+-------+---------+--------+------------+---------+ 
| S_NODE          | 3306  | MASTER  | UP     | ON         | OK      |
| M_NODE          | 3306  | SLAVE   | UP     | ON         | OK      |
+-----------------+-------+---------+--------+------------+---------+

I can't seem to find any google or stackoverflow pages related to that error message "Candidate slave is missing replication user."

I believe I have a replication user on both node S and M, since both acted as a slave at one point in my test script.

Any thoughts?

Answer 1

I ran into this same problem while testing failover on some new clusters and noticed that after adding the --force flag as you suggest, the switchover completed successfully and created the user 'replication_user'@'S_NODE' without any prompting on my part.

It then occurred to me that the replication user I had set up was 'repl'@'specific_subnet_%' - which did not match what the script was looking for ('repl'@'S_NODE'), so the pre-check failed without --force .

I had another cluster I was testing, so to validate I did the following:

• ran the mysqlrpladmin ... switchover command without --force (it failed as expected)
• created both 'replication_user'@'S_NODE' and 'replication_user'@'M_NODE' with the same grant ( REPLICATION SLAVE ) and password as my existing replication_user. You need both S_NODE and M_NODE hosts if you want to be able to fail back and forth between them using this script.
• tried mysqlrpladmin ... switchover again without --force

It worked like a charm this time!

Moral of the story: the script expects the 'replication_user'@'new_master' user to exist. The --force flag will create that user for you, and should not be necessary after the first time you use it.

Answer 2

For some reason, adding "--force" flag fixed the issue. More details in the comments above. If anyone knows why this might be though, please comment.