Parallel Region Create API calls can produce top-level regions with the same tree_id

[DevOpsFu]

NetBox version

v3.4.8

Python version

3.8

Steps to Reproduce

Ref: e-breuninger/terraform-provider-netbox#385

Using a tool which will produce parallel API calls to NetBox (e.g. Terraform), create two or more top-level Region objects with children, via the /dcim/regions endpoint.

In Terraform, this might look like this:

resource "netbox_region" "uk" {
  name = "United Kingdom"
}

resource "netbox_region" "ireland" {
  name = "Ireland"
}

resource "netbox_region" "london" {
  name             = "London"
  parent_region_id = netbox_region.uk.id
}

resource "netbox_region" "gatwick" {
  name             = "Gatwick"
  parent_region_id = netbox_region.uk.id
}



resource "netbox_region" "northeurope" {
  name             = "Azure North Europe"
  parent_region_id = netbox_region.ireland.id
}

resource "netbox_region" "uksouth" {
  name             = "Azure UK South"
  parent_region_id = netbox_region.london.id
}

Expected Behavior

The Regions are created with the correct tree structure in the NetBox UI

Observed Behavior

The top-level regions can sometimes be created with the same tree_id in the dcim_region table. This causes some weird UI issues whereby all the child regions will appear under one of the top-level regions, despite it not being their parent:

If you follow the child region links, the Parent IDs are set correctly.

To work around this, we can ensure that the top-level regions are created in serial so that they never receive the same tree_id value.

[arthanson]

This looks like it is due to django-mptt/django-mptt#677 see: django-mptt/django-mptt#555 but the PR is stagnant so marking this as blocked for either #6587 or #11421

[arthanson]

@DevOpsFu could you please test against #13942 I made the region API calls atomic which I think should fix the issue. I don't have terraform setup so don't have a good test scenario.

[DevOpsFu]

@DevOpsFu could you please test against #13942 I made the region API calls atomic which I think should fix the issue. I don't have terraform setup so don't have a good test scenario.

I'd be happy to test! I'll see if I can get the code at that PR built in Docker and get back to you soon 👍

[DevOpsFu]

Hi @arthanson , I have tested this against your branch but unfortunately I still get the same behaviour. The netbox stack was totally clean to start with, and running against a docker image built from your branch:

❯ docker ps
CONTAINER ID   IMAGE                                               COMMAND                  CREATED         STATUS                   PORTS                                       NAMES
e7dc6c711d55   netboxcommunity/netbox:12336-serialize-region-api   "/usr/bin/tini -- /o…"   5 minutes ago   Up 2 minutes (healthy)                                               netbox-docker-netbox-housekeeping-1
e2d05181e486   netboxcommunity/netbox:12336-serialize-region-api   "/usr/bin/tini -- /o…"   5 minutes ago   Up 2 minutes (healthy)                                               netbox-docker-netbox-worker-1
cf382b1dc07d   netboxcommunity/netbox:12336-serialize-region-api   "/usr/bin/tini -- /o…"   5 minutes ago   Up 5 minutes (healthy)   0.0.0.0:8000->8080/tcp, :::8000->8080/tcp   netbox-docker-netbox-1
7d36a9501ebc   postgres:15-alpine                                  "docker-entrypoint.s…"   5 minutes ago   Up 5 minutes             5432/tcp                                    netbox-docker-postgres-1
5676b041e21b   redis:7-alpine                                      "docker-entrypoint.s…"   5 minutes ago   Up 5 minutes             6379/tcp                                    netbox-docker-redis-cache-1
96767ca07745   redis:7-alpine                                      "docker-entrypoint.s…"   5 minutes ago   Up 5 minutes             6379/tcp                                    netbox-docker-redis-1

When running the Terraform code I pasted originally, I see the following region structure gets produced:

[arthanson]

Thanks @DevOpsFu I guess I will have to change to use locks, will get a new version out shortly.

[arthanson]

@DevOpsFu could you please try one more time, I've updated the PR to use postgres locks which I hope will fix it.

[DevOpsFu]

@arthanson Looks good! I did a few create/destroy cycles with the above TF code and the region tree looked correct every time. Thanks for looking at this!