Minimal hub home-setup

[AliveDevil]

Hey,

I'm trying to setup Kerberos for the third or fourth time now, without really knowing how these components work together - never really got anywhere - something always broke.

In case I want to setup a self-contained, not publicly exposed Kerberos Hub setup (for monitoring some cameras, with timeline/scrubbing, central view, live ML object detection), what's the absolute minimum of services I need to install, to get this rolling?

From what I could scrape together in the helm-charts repository and here:

• Set up a dedicated database VM (db.domain.internal), on which FerretDb a MongoDB replacement with PostgreSQL runs.
• I'm colocating some services on a "hub"-VM (hub.domain.internal), that is:
  • Eclipse Mosquitto for MQTT (TCP, WebSockets)
  • RabbitMQ for AMQP
  • uugai/vault - configured with MQTT and MongoDB URI
  • uugai/hub-api
  • uugai/hub-frontend
• One physical machine forwarding an ONVIF stream to some Kerberos service.
• One VM with all the pipe-services
• One or multiple VMs with an ML object detection service

Now, in the helm-charts repository are a bunch of added services:

• vault-proxy/vault-forwarder: As far as I can tell, these are optional/not required, when my setup involves onf of each hub and vault
• hub-pipe-*, which of these is really necessary? By configuration pipe-sprite and pipe-thumbnail are optional, pipe-notify-test sounds optional as well
• For object-detection there is uugai/hub-yolov8, uugai/data-filter, and kerberos/hub-ml
  Which of these is necessary to get live-detection and record-detection?

There are some things I don't quite get with the dataflow from the agent to hub/vault - do agents need to forward to the vault first, or is this only really necessary, when forwarding from an edge vault to a cloud hub?
I.e. is Agent -> Hub <-> Vault as minimal configuration usable, or will this always be Agent -> Vault <-> Hub (is this single-vault setup even possible)?

I've found at least two different queues for the initial ingest of events: kcloud-event-queue, data-filtering.

What difference does it make when configuring Vault to do edge or cloud detection?

In both, deployment and helm-charts, is sample MongoDb data, but when creating an account with POST /account/user/create and the admin api key, the subscriptions-table isn't populated, but a nested subscription object used in the user table, what's up with that?

Is it possible to define different retention periods, for different artifacts? e.g. recordings, that are analyzed and triggered an event due to motion/object detection should be retained for a week, all other recordings should be dropped within a day.

Is

• uugai/hub-pipeline-notification
• uugai/hub-pipeline-sprite

superseding

• kerberos/pipe-notify
• kerberos/pipe-sprite?

[cedricve]

hey @AliveDevil thanks for the information. I'm actually working on this for the moment, and exactly good documentation is missing (and some changes were required). I'm still working this out but believe I would be able to have something well documented and working EOW - early next week. Will come back on your questions as well.

[AliveDevil]

Sounds good, looking forward to the improved documentation :)