We would love to see no-downtime deployments in Mendix Cloud. Today, neither in the developer portal web UI or the Mendix platform API do we have the option to restart just one linux container at a time when deploying so there is a 20 to 30 second outage every time we deploy a change to Mendix Cloud. It is not uncommon for us to deploy multiple times a day.
In cases were there is only one linux container configured to run the app we understand downtime is much harder to avoid but in cases were there are 2 or more linux containers running an app restarting them one a time so that there is always at least one linux container serving requests is a common pattern. The database schema changes complicate things significantly I’m sure and I’m not glossing over that.
If we can get there technically, we would very much like to have zero downtime deployments.
Same functionality is required for also mendix for private cloud.
We host our mendix applications in our own datacenter with mendix for private cloud and run our applications on 3 openshift clusters as active-active. That means each application run actively on 3 openshift clusters. This is a mendix supported deployment architecture. We use "runtimeLeaderSelection" configuration to enable master pod on only one of the clusters. That the way mendix supports multi cluster active-active runtime on Mx4PC.
So, if a feature is being implemented, I expect it to be supported on every runtime architecture.
I am emphasizing the proposal from Paul. I would love for the Mendix Cloud to be able to do application deploys and restart for multi-instanced application environments, instance-by-instance so to avoid downtime. This is especially important for our B2C focussed application.
In DevOps times with automated deployment pipelines, would be great to have zero downtime.
Interestingly, this is already possible when running locally from the Modeller. So you would not necessarily need extra environments spinning up and others being shut down.
Concerning database schema changes; it would be a good idea to have specific support for conversion logic when changing the model anyway. That should help in doing zero downtime deployments when there are schema changes.
Absolutely. HA should allow for failover and zero-down time.
This is a cost-issue rather than a technical one, since zero-downtime deployment is certainly possible.