mmm_control show indicates all-good even though mmmd_agent could not access db

[Mike Pelley]

We have a simple two database server master-master setup with the monitor running on a third application server. To test our restore procedure, we reinstalled both database servers from scratch, set MySQL to replicate in both directions, restored our mysqldump file, and configured mmmd_agent. Using mmm_control on the third server, we "set_active" and then "set_online" the two db servers.

At this point, mmm_control show indicated that everything was good - one reader role on each server and one writer role on one of the servers. However, we had forgotten to do a "flush privileges" on the database server after the restore, so the mmmd_agent.error log was filled with messages like this:

2009/12/04 11:38:30 FATAL Couldn't allow writes: ERROR: Can't connect to MySQL (host = 192.168.11.24:3306, user = mmmd_agent)!

Because of this problem, mmmd_agent (rightly) would not configure the writer role IP on the interface, although mmm_control show indicated that it was assigned.

I'd suggest that mmm_control show should indicate a problem when one of the agents cannot access the database.

[Pascal Hofmann (pascalhofmann)]

Some problem with failed checks:

[root@monitor mysql-mmm]# /usr/bin/mysql-mmm/monitor/checker rep_threads
check host1
UNKNOWN: Unknown state. Execute error: Access denied; you need the SUPER,REPLICATION CLIENT privilege for this operation

.. but this error just leads to the host flapping between REPLICATION_FAIL and ONLINE... no trace of it in the mmm_control show output.

[Pascal Hofmann (pascalhofmann)]

The problem described in my previous comment is fixed now - the checker will report an error if it doesn't have the correct privileges.