Hi,

Context:

-----------

Version: MariaDB 10.0

Repo: http://yum.mariadb.org/10.0/centos7-amd64

OS: CentOS Linux release 7.2.1511

Kernel: 3.10.0-229.7.2.el7.x86_64

Cloud Provider: Windows Azure

I run a three-node MariaDB-Galera cluster on production for my company. With the custom config file at:

/etc/my.cnf.d/server.cnf

My datadisk is mounted over a lvm2 partition over two physical volumes, mounted on the path.

/datadrive/galera/mysql

The logical volume only takes up 50% of the volume group as I leave the empty space for snapshot operations.

This is reflected in my configuration as:

[mysqld]
datadir = /datadrive/galera/mysql
socket = /datadrive/galera/mysql/mysql.sock

Issue:

--------

Coming to the issue at hand;

When starting the primary component through:

$ sudo service mysql bootstrap

or,

$ sudo service mysqld start --wsrep-new-cluster

( Essentially same, as bootstrap differs to start --wsrep_new_cluster )

The bootstrap sequence correctly calls mysqldsafe, as below

/usr/bin/mysqld_safe --datadir=/datadrive/galera/mysql --pid-file=/datadrive/galera/mysql/ciq-test-db01.pid --wsrep-new-cluster

However,

it then gets stuck in the wait_for_ready() function of /etc/init.d/mysql

The sequence tries to ping mysqladmin through:

if $bindir/mysqladmin ping >/dev/null 2>&1; then
log_success_msg
return 0
elif kill -0 $! 2>/dev/null ; then
: # mysqld_safe is still running
else

# mysqld_safe is no longer running, abort the wait loop
break

It gets stuck here as the test:

if $bindir/mysqladmin ping >/dev/null 2>&1; then

Always fails and the counter keeps increasing till 900 and Error's out with the

log_failure_msg

return 1

Test:

I tested that line from the console as:

$mysqladmin ping >dev/null 2>&1

and, was thrown the following error:

mysqladmin: connect to server at 'localhost' failed
error: 'Can't connect to local MySQL server through socket '/var/lib/mysql/mysql.sock' (2 "No such file or directory")'
Check that mysqld is running and that the socket: '/var/lib/mysql/mysql.sock' exists!

Obviously, '/var/lib/mysql' does not exist in my setup, and the socket file was at '/datadrive/galera/mysql/mysql.sock'

But, my socket can be passed explicitly to mysqladmin by doing,

$mysqladmin --socket=/datadrive/galera/mysql/mysql.sock ping

This actually worked as expected and I got the desired output,

mysqld is alive

Solution:

-----------

What I surmised through my novice abilities was that the init script was not passing the variables it read from /etc/my.cnf.d/server.cnf to mysqladmin

So, for now I have done a hackjob by altering the init script function wait_for_ready() with the test:

if $bindir/mysqladmin --socket=/datadrive/galera/mysql/mysql.sock ping >/dev/null 2>&1; then
log_success_msg
return 0

And voila,

$ sudo service mysql start
$ sudo service mysql start --wsrep-new-cluster
$ sudo service mysql bootstrap

All above works. It is evident that I should be passing all the variables from /etc/my.cnf.d/server.cnf that mysqladmin needs to set it's environement correctly.

However I think that this could be / should be done on the pre-distributed init script itself rather than a user-side hack.

While I have tried to be thorough in my inspection of the issue, I may have missed either something very basic or inherently complex that's currently is inherent to the process of initializing the mysql service. Please redirect me to a corrected course if that is the case.

P.S. First time post in the list, I may have stated things unacceptably

P.P.S. Merry Christmas to you all.

Thanks and regards,

Joy Bhattacherjee

Mob: +91-9011235028