You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I think GAE flex in some ways suits us better; it seems more mature, better documented and "cleaner" than Azure Web Apps for containers.
However, the pricing structure is a big, big disadvantage for typical shiny loads as explained in subugoe/shinycaas#11.
As of now, GAE flex will open a new VM for every version of every service, which gets very expensive fast and also just wastes a lot of empty cycles.
By contrast, in Azure Web Apps for Containers, we can run several apps and even their slots (~versions) in the same VM, which suits our loads much better.
Weirdly, this is actually kind of an inconsistent design in Azure, because a (micro)services architecture would imply that each of the services can be scaled independently, and sharing a VM kind of defeats that purpose.
But good for us, for now.
In the long run, a cleaner design such as GAE flex might be preferable, if:
GAE flex allows <1 vCPU/Instance (RAM would still make it expensive)
GAE flex allows us to share VMs (making it like Azure Web apps)
GAE flex scales to 0 and/or Google Cloud Run allows websockets and/or miraculously allows stateful apps (not happening) subugoe/shinycaas#13
Anyway, keeping this open, might check back in a year or so.
I think GAE flex in some ways suits us better; it seems more mature, better documented and "cleaner" than Azure Web Apps for containers.
However, the pricing structure is a big, big disadvantage for typical shiny loads as explained in subugoe/shinycaas#11.
As of now, GAE flex will open a new VM for every version of every service, which gets very expensive fast and also just wastes a lot of empty cycles.
By contrast, in Azure Web Apps for Containers, we can run several apps and even their slots (~versions) in the same VM, which suits our loads much better.
Weirdly, this is actually kind of an inconsistent design in Azure, because a (micro)services architecture would imply that each of the services can be scaled independently, and sharing a VM kind of defeats that purpose.
But good for us, for now.
In the long run, a cleaner design such as GAE flex might be preferable, if:
Anyway, keeping this open, might check back in a year or so.
Related: subugoe/hoad#25
The text was updated successfully, but these errors were encountered: