Reliable stack deployment
In some cases the
registrar run into a race condition when one is creating the shared instance. Sometimes, service that does not set up the instance starts with a half configured
settings.py leading to strange bugs.
Using a service dependency does not work, as e.g the
renderer will not always be present.
Ideas for a solution:
- using a per-container instance:
- reliable instance creation/configuration
- database as synchronization out of the box
- duplication of files (especially static seems to be > 4GB; but actually unnecessary)
- spinup times increased (how much?)
- sticking to shared instance, with better synchronization