Comments on: How Not to do MySQL High Availability: Geographic Node Distribution with Galera-Based Replication Misuse https://www.percona.com/blog/how-not-to-do-mysql-high-availability-geographic-node-distribution-with-galera-based-replication-misuse/ Thu, 28 Dec 2023 23:03:11 +0000 hourly 1 https://wordpress.org/?v=6.5.2 By: M A Wal-Ikram https://www.percona.com/blog/how-not-to-do-mysql-high-availability-geographic-node-distribution-with-galera-based-replication-misuse/#comment-10971606 Thu, 02 Jan 2020 05:05:21 +0000 https://www.percona.com/blog/?p=53654#comment-10971606 Is there a guide around to do this for 2 Kuberenetes clusters?

With https://www.percona.com/doc/kubernetes-operator-for-pxc/kubernetes.html we can get 2 or 3 clusters running. A guide on how to do the asynchronous replication across k8s clusters would be very helpful!

Thank you.

]]> By: Warren Johnson https://www.percona.com/blog/how-not-to-do-mysql-high-availability-geographic-node-distribution-with-galera-based-replication-misuse/#comment-10971588 Sat, 28 Dec 2019 02:38:49 +0000 https://www.percona.com/blog/?p=53654#comment-10971588 We have successfully run several MySQL Galera database clusters across AWS regions with minimal issues. The key has been knowing and understanding inherent latency for writing to the database and engineering around that. For example, if you have a single MySQL database running on the same network as your application server, you might do a series of consecutive writes without using a transaction. Its no problem to do that because the latency to the database is virtually 0. If you’re doing bulk writes between nodes separated by 10,000 miles, you make sure you use transactions. I wouldn’t try that with any other network infrastructure besides AWS. Whether this works for you or not, I think depends on your application and how much writing you’re doing. Being able to update one database node in Virginia and see those changes instantly in Japan is enormous for us even though we have had the odd issue over the years.

]]> By: Alex https://www.percona.com/blog/how-not-to-do-mysql-high-availability-geographic-node-distribution-with-galera-based-replication-misuse/#comment-10969973 Sun, 18 Nov 2018 18:14:36 +0000 https://www.percona.com/blog/?p=53654#comment-10969973 Hi Marco!

Did I get it right that the numbers you posted were generated with ping and netperf, not with the actual Galera replication? If so, I think here you have proven that latency and packet rate are inversely proportional to packet size when you have maxed out the network throughput. And indeed, Galera replication would behave the same if you max out the network throughput. Which would be a poor choice of network capacity for the task at hand then. Note that if the master generates more transactions than the network can handle, asynchronous replication would help you little, as the slave will simply lag behind being unable to receive transactions at a master rate.

What would be interesting to test is the actual geo-distributed Galera replication against varying writeset sizes since it won’t be able to saturate the network so easily. Even though there no way to set exact writeset size, they scale sufficiently linearly with the number of updated/inserted rows and it could be relatively easily done with mysqlslap, for instance. It would be really curious to see how trx rate and latency would behave with increasing writeset size.

Kind regards,
Alex

]]> By: Eero Teerikorpi https://www.percona.com/blog/how-not-to-do-mysql-high-availability-geographic-node-distribution-with-galera-based-replication-misuse/#comment-10969967 Fri, 16 Nov 2018 22:38:01 +0000 https://www.percona.com/blog/?p=53654#comment-10969967 You may also want to consider Continuent Clustering — Multisite/multimaster Continuent Clustering deployments for geo-distributed apps. See webinar at https://www.youtube.com/watch?v=qvXBegyCe08&feature=youtu.be

]]>