Hi folks,

An interesting debate over the best HA setup up with MySQL on System z is
now motivated me to reach out to the community to ask for your opinion.

We have 2 production LPARS on a z/10 running Red Hat v5.3 and are connected
in a common lan segment using vswitch.  On each LPAR the customer wishes to
have an MySQL database server: one master (as primary) and one slave (as a
failover for backup).  If the primary fails, the database gets umounted and
mounted RW on the slave host.. ip takover happens and mysql starts up on
the filesystem that was swapped over.

The question simply is - what is the best high availability setup to allow
for this scenario to happen?   A few options that have lead to good debate:

1)  Because the database will be 200-300GB, we can use LVM and have it
available to mount RW on either system.   We need to be very careful not to
mount RW at the same time; else we risk destroying the data.  In cases
where customers are using VMWARE or Xen on distributed, you ensure this by
using PaceMaker to tell the hypervisor to issue a STONITH to totally kill
the primary machine.  This ensures, that the filesystem will be RW only on
the backup now.  However!  There is currently no capability on z/VM to
allow PaceMaker (or any other software that I know of) to do this.
Instead, we may be able to configure PaceMaker to issue a #CP LOG off
command, but how do we get this message across to the 2nd LPAR?   Perhaps
CSE (Cross - System - Extensions) will allow for it?  Or perhaps it would
be easier to host these two servers on the same LPAR.   Thoughts?

2)  To avoid the playing around with the issues in (1), we then thought to
use the data replication in MySQL instead; replication happens with the
primary writing what insert / delete statements the slave needs to run into
a bin.log file - which is single thread and serial.   The slave runs this
in sequence in the efforts to try and match what the master is getting.
However, our MySQL experts tell us that because of the single thread nature
of this asyncronous replication, there is risk that the slave server won't
be able to catch up.  Since the database will be running with 800
transactions / second, in the event of a failover, many tranactions will be
lost.  Thus, MySQL support recommends to use DRDB filesystem block
replication instead.    I wonder - any takers for this option?   Concerns
for performance?

Any thoughts will be appreciated it.   I also saw somebody write about
putting a NFS up and then use it over hipersockets.  This seems simple, but
alas, 800 transactions per second on a NFS system - performance issues
looming?

Any thoughts are welcomed.  Thanks!


- Frank Pani

----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to