Comments on: Doing a rolling upgrade of Percona XtraDB Cluster from 5.5 to 5.6 https://www.percona.com/blog/rolling-upgrade-percona-xtradb-cluster-5-5-5-6/ Fri, 12 May 2023 19:39:59 +0000 hourly 1 https://wordpress.org/?v=6.5.2 By: steph https://www.percona.com/blog/rolling-upgrade-percona-xtradb-cluster-5-5-5-6/#comment-10553717 Fri, 20 Feb 2015 12:56:36 +0000 https://www.percona.com/blog/?p=22614#comment-10553717 Jay

I wasnt expecting such a quick reply. Thanks alot. I feel quite bad for the question that will be placed forth because it feels more like request for consulting rarher giving “tips”. Nevertheless i will ask:

Thing is a bit more complecated. Haproxy is not used at all. The way the traffic is directed straight to server A is by using some sort of routing mechanism of the hosting provider. Using providers API you tell where to route the traffic. I have one more question:

During the upgrade of B and C, the server A will contain some data that others servers dont have (since they are read only). Assuming the traffic is switched from server A to C . At that poing, server A has data that server B,C dont have, and servers B,C (since read only is switch off and traffic is received by the servers) have data that server A does not have.

When server A upgrade is completed and server is back in cluster, will the missing data of server A to server B,C get replicated and vise versa?

Thanks once again

]]> By: Jay Janssen https://www.percona.com/blog/rolling-upgrade-percona-xtradb-cluster-5-5-5-6/#comment-10553653 Fri, 20 Feb 2015 12:08:51 +0000 https://www.percona.com/blog/?p=22614#comment-10553653 server A must be shut down the moment of traffic switching to B and C. This, if done carefully, should be the same as switching nodes on, say, a node failure. The speed here depends on your mechanism for Application to Cluster HA. For example, HAproxy could monitor for read_only, which would blend your steps 2 and 3 together (though realize HAproxy won’t switch instantaneously by default).

Hence, would do your steps in roughly the order:
2 + 3
1
4

]]> By: steph https://www.percona.com/blog/rolling-upgrade-percona-xtradb-cluster-5-5-5-6/#comment-10553612 Fri, 20 Feb 2015 11:47:39 +0000 https://www.percona.com/blog/?p=22614#comment-10553612 Hi Jay

Your article is most helpful however i am abit confused hence your comment would be mouch appriciated.
I am trying to figure out how is possble to have NO DOWNTIME following this procedure.

Assuming you have a PCX cluster of three serevrs. They are all in 5.5 version and all traffic is configured as such tha it is pointed to server A where all other servers (B,C) actually just receive the replications.

So, you take down node B, set it to read only mode, upgrade to 5.6 and up again. You do the same for server C. So you now have servers B,C in 5.6 in Read only mode and server A which handles all traffic to 5.5.

How do you proceed from there? the only way i can think of is:
1) shout down server A
2) disable read only for server B,C
3) switch traffic to either B or C
4) proceed upgrading server A as usual.

If that is the case, then indeed you have a down time between steps 1 to 3.

Am i thinking correct?

Thnaks in advance

]]>