On Mon, Jan 11, 2021 at 11:13 PM nikita.zinov...@gmail.com < nikita.zinov...@gmail.com> wrote:
> I'm really amazed that Payara finally made it to a fully clustered App > server setting, people say they have seamless green/blue deployments there > with a cluster of 2-3 payara servers. I wish we could also get a clustered > PostgreSQL or an equivalent... > We've been running a cluster of GF/Payara servers for years. It was an adhoc cluster, not using the in built GF facilities for deployment. Rather, we just did sticky load balancing with little session sharing, and JMS topics for crossleg communication. It was originally done for availability, keeping the system up if one leg were to go down, so it was just two legs. We eventually had to move to 3 for performance reasons as a single machine could not take the traffic. GF has some nice clustering facilities, but for us it was simply easier to manage the deploys to the legs manually than to practice and learn the GF kit. We were live and just not inconvenienced enough to try to pursue it. We also run an OpenMQ cluster across all of the machines. We had good success with OpenMQ, but we certainly beat the heck out of it and have had different issues surrounding it. The back end is Postgres that has a live streaming replica. The front end is HAProxy. We are running 3 Payara instances on each leg, each supporting a different part of the application. We also had an independent cluster of 3 Payara instances running a specific service that has since been migrated to a completely different service. All of these were integrated via SAML SSO. We broke the applications up in Payara mostly for monitoring and independence. When things went awry it was much easier to find which application was misbehaving when they were isolated like that. But we certainly had scenarios in the past in lower load environments where everything was on one instance of Payara. The dev server, for example, had everything in a single instance. > (PostgreSQL offering, say, in Google cloud seems to be nice, scalable, > auto backed-up, but not clustered, it seems Oracle Autonomous DB can now do > it, but I never had chance to try Oracle Cloud and have no idea how > expensive it is, probably, not so much... Oh, and it seems there's also > Google's Spanner DB). > AWS Aurora Postgres is an impressive piece of kit. It's quite remarkable how a database implementation can change when you assume you have dozens of machines, stable memory, high bandwidth networking, and geographical isolation. If you're friendly to AWS, it's worth considering, but you can not use a cloud database without the rest of your application co-deployed in the same infrastructure. The lag will crush you. Regards, Will Hartung
