Reflect on implementing Signalen at 's-Hertogenbosch
bartjkdp opened this issue · comments
Bart commented
These are some things we stumbled upon when bringing Signalen in production at 's-Hertogenbosch:
- Signalen loads Amsterdam-specific data by default. Currently we remove this data in a manual deployment step, would be nice if the backend is "empty by default" -> #81
- Managing Elasticsearch as an extra dependency takes some effort. Would be nice of Elasticsearch is not neccesary. #86
- Importing areas can only be done by a system administrator. Would be nice of an application administrator is able to do this in the admin panel. -> #134
- Importing the machine learning model can only be done by a system administrator. Would be nice of an application administrator is able to do this in the admin panel. -> Signalen/classification#3
- Dumping and restoring User and Profile data is difficult as Profile contains two unique keys (
id
anduser_id
). When creating aUser
, aProfile
is automatically created with a different primary key, that prevents a Profile from the data dump to be restored. - For monitoring we included the ddtrace package. Can we move towards a more vendor-neutral way of collecting application logs?
- Access to Django admin needs separate username / password. Would be nice to have single-sign-on on the Django admin as well. -> #133
- Improve audit logging (@CBuiVNG) -> #84
- Default e-mails are Amsterdam-specific