Some strawberry/GraphQL issues porting NetBox DNS to NetBox 4

[peteeckel]

I'm currently in the process of porting NetBox DNS to NetBox 4. Currently I'm working on the GraphQL migration to strawberry.

Generally the migration is working fine, however I have some issues with generating filters using the @autotype_decorator decorator, due to some implementation specifics in NetBox DNS.

Problem with filtering on annotations for a model

The Zone model uses an annotation active that can be used to determine if a zone is in an active state or not. This annotation is used for the REST API as well, which works fine, but @autotype_decorator fails on creating a filter for it with the following error message:

Traceback (most recent call last):
  File "/opt/netbox/lib64/python3.11/site-packages/django/db/models/options.py", line 681, in get_field
    return self.fields_map[field_name]
           ~~~~~~~~~~~~~~~^^^^^^^^^^^^
KeyError: 'active'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/opt/netbox/netbox/manage.py", line 10, in <module>
    execute_from_command_line(sys.argv)
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/management/__init__.py", line 442, in execute_from_command_line
    utility.execute()
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/management/__init__.py", line 416, in execute
    django.setup()
  File "/opt/netbox/lib64/python3.11/site-packages/django/__init__.py", line 24, in setup
    apps.populate(settings.INSTALLED_APPS)
  File "/opt/netbox/lib64/python3.11/site-packages/django/apps/registry.py", line 124, in populate
    app_config.ready()
  File "/opt/netbox/netbox/netbox/plugins/__init__.py", line 125, in ready
    if graphql_schema := self._load_resource('graphql_schema'):
                         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/netbox/netbox/plugins/__init__.py", line 91, in _load_resource
    module = import_module(default_module)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.11/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen importlib._bootstrap>", line 1204, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1176, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1147, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 690, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 940, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "/vagrant/netbox-dns/netbox_dns/graphql/__init__.py", line 5, in <module>
    from .zone import NetBoxDNSZoneQuery
  File "/vagrant/netbox-dns/netbox_dns/graphql/zone.py", line 15, in <module>
    @autotype_decorator(ZoneFilterSet)
     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/netbox/netbox/graphql/filter_mixins.py", line 167, in wrapper
    field = model._meta.get_field(fieldname)
            ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/db/models/options.py", line 683, in get_field
    raise FieldDoesNotExist(
django.core.exceptions.FieldDoesNotExist: Zone has no field named 'active'

Definition of the annotation:

    def get_queryset(self):
        return (
            super()
            .get_queryset()
            .annotate(
                active=ExpressionWrapper(
                    Q(status__in=Zone.ACTIVE_STATUS_LIST), output_field=BooleanField()
                )
            )
        )

Definition in the filterset:

    active = django_filters.BooleanFilter(
        label="Zone is active",
    )

When I disable the definition in the filter, the error does not occur. It would help to be able to define that GraphQL filter manually and have it be ignored by @autotype_decorator, but so far I could not figure out how that can be done.

Problem with custom fields

For the Record and Zone models I am using some custom field types, NetworkField, AddressField, and RFC2317NetworkField that have special validation logic and lookups attached to them. With graphene I had them working by defining GraphQL conversions for them:

import graphene
from graphene_django.converter import convert_django_field

from netbox_dns.fields import NetworkField, AddressField, RFC2317NetworkField

@convert_django_field.register(NetworkField)
def convert_field_to_string(field, registry=None):
    return graphene.String(description=field.help_text, required=not field.null)

@convert_django_field.register(AddressField)
def convert_field_to_string(field, registry=None):
    return graphene.String(description=field.help_text, required=not field.null)

@convert_django_field.register(RFC2317NetworkField)
def convert_field_to_string(field, registry=None):
    return graphene.String(description=field.help_text, required=not field.null)

I assume something similar has to be done for strawberry, but couldn't find out how to do it yet. I could also live with having the fields excluded from filtering in GraphQL, as they are not really providing end user functionality.

@autotype_decorator fails with the following error for these fields:

Traceback (most recent call last):
  File "/opt/netbox/lib64/python3.11/site-packages/graphql/type/definition.py", line 1456, in fields
    fields = resolve_thunk(self._fields)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/graphql/type/definition.py", line 300, in resolve_thunk
    return thunk() if callable(thunk) else thunk
           ^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/strawberry/schema/schema_converter.py", line 410, in <lambda>
    fields=lambda: self.get_graphql_input_fields(type_definition),
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/strawberry/schema/schema_converter.py", line 388, in get_graphql_input_fields
    return _get_thunk_mapping(
           ^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/strawberry/schema/schema_converter.py", line 130, in _get_thunk_mapping
    field_type = field.type
                 ^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/strawberry/field.py", line 302, in type
    return self.resolve_type()
           ^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/strawberry_django/fields/base.py", line 183, in resolve_type
    resolved_type = resolve_model_field_type(
                    ^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/strawberry_django/fields/types.py", line 508, in resolve_model_field_type
    raise NotImplementedError(
NotImplementedError: GraphQL type for model field 'netbox_dns.Zone.arpa_network' has not been implemented

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/opt/netbox/netbox/manage.py", line 10, in <module>
    execute_from_command_line(sys.argv)
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/management/__init__.py", line 442, in execute_from_command_line
    utility.execute()
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/management/__init__.py", line 436, in execute
    self.fetch_command(subcommand).run_from_argv(self.argv)
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/management/base.py", line 413, in run_from_argv
    self.execute(*args, **cmd_options)
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/management/base.py", line 454, in execute
    self.check()
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/management/base.py", line 486, in check
    all_issues = checks.run_checks(
                 ^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/checks/registry.py", line 88, in run_checks
    new_errors = check(app_configs=app_configs, databases=databases)
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/checks/urls.py", line 14, in check_url_config
    return check_resolver(resolver)
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/core/checks/urls.py", line 24, in check_resolver
    return check_method()
           ^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/urls/resolvers.py", line 519, in check
    for pattern in self.url_patterns:
                   ^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/utils/functional.py", line 47, in __get__
    res = instance.__dict__[self.name] = self.func(instance)
                                         ^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/urls/resolvers.py", line 738, in url_patterns
    patterns = getattr(self.urlconf_module, "urlpatterns", self.urlconf_module)
                       ^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/utils/functional.py", line 47, in __get__
    res = instance.__dict__[self.name] = self.func(instance)
                                         ^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/django/urls/resolvers.py", line 731, in urlconf_module
    return import_module(self.urlconf_name)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.11/importlib/__init__.py", line 126, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "<frozen importlib._bootstrap>", line 1204, in _gcd_import
  File "<frozen importlib._bootstrap>", line 1176, in _find_and_load
  File "<frozen importlib._bootstrap>", line 1147, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 690, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 940, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "/opt/netbox/netbox/netbox/urls.py", line 9, in <module>
    from netbox.graphql.schema import schema
  File "/opt/netbox/netbox/netbox/graphql/schema.py", line 35, in <module>
    schema = strawberry.Schema(
             ^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/strawberry/schema/schema.py", line 143, in __init__
    self._schema = GraphQLSchema(
                   ^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/graphql/type/schema.py", line 224, in __init__
    collect_referenced_types(query)
  File "/opt/netbox/lib64/python3.11/site-packages/graphql/type/schema.py", line 435, in collect_referenced_types
    collect_referenced_types(arg.type)
  File "/opt/netbox/lib64/python3.11/site-packages/graphql/type/schema.py", line 438, in collect_referenced_types
    for field in named_type.fields.values():
                 ^^^^^^^^^^^^^^^^^
  File "/usr/lib64/python3.11/functools.py", line 1001, in __get__
    val = self.func(instance)
          ^^^^^^^^^^^^^^^^^^^
  File "/opt/netbox/lib64/python3.11/site-packages/graphql/type/definition.py", line 1459, in fields
    raise cls(f"{self.name} fields cannot be resolved. {error}") from error
TypeError: NetBoxDNSZoneFilter fields cannot be resolved. GraphQL type for model field 'netbox_dns.Zone.arpa_network' has not been implemented

First call to the GraphQL API is extremely slow

After a NetBox restart, the first call (per worker) to the GraphQL API is very slow, all subsequent calls are almost instantaneous:

$ time curl -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.example.com/graphql/ --data '{"query": "query {netbox_dns_record_list {name, type, value, zone{view {name}, name}}}" }'
[...]
0.06s user 0.01s system 0% cpu 13.374 total
$ time curl -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.example.com/graphql/ --data '{"query": "query {netbox_dns_record_list {name, type, value, zone{view {name}, name}}}" }'
[...]
0.05s user 0.01s system 50% cpu 0.123 total

Is this expected, or is there anything wrong? The GUI and the REST API do not show this behaviour.

Maybe @arthanson can shed some light on this? I'll investigate further and see what I can figure out as well.

[arthanson]

@peteeckel I'll try and take a look later today

[peteeckel]

I found one workaround (albeit an ugly one) for the first two issues:

• Subclass the original filterset for the model (e.g. class ZoneGraphQLFilterSet(ZoneFilterSet))
• Override the Meta class, removing the fields that are causing problems
• Use the subclass in the @autotype_decorator

That way, at least the other fields can be used for filtering.

[peteeckel]

Is this expected, or is there anything wrong? The GUI and the REST API do not show this behaviour.

I must correct myself - at least the GUI shows the same behaviour.

[peteeckel]

Update: I can query the annotated field active, I just can't define it in the filterset used for @autotype_decorator:

@strawberry_django.type(Zone, fields="__all__", filters=NetBoxDNSZoneFilter)
class NetBoxDNSZoneType(NetBoxObjectType):
    name: str
    status: str
    active: bool
[...]

curl -s -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.example.com/graphql/ --data '{"query": "query {netbox_dns_zone_list(filters: {name: {i_exact: \"zone1.example.com\"}}) {view {name}, name, active}}" }' | json_pp
{
   "data" : {
      "netbox_dns_zone_list" : [
         {
            "active" : true,
            "name" : "zone1.example.com",
            "view" : null
         }
      ]
   }
}

So it's very likely not a strawberry issue but one with @autotype_decorator.

[arthanson]

@peteeckel Answers to some of your issues / questions:

1. Problem with filtering on annotations for a model
   I'll have to look into supporting this, will probably have to skip it in the autotype_decorator and have it defined explicitly in your GraphQL code.
2. Problem with custom fields
   If you have a custom type defined for them you can just add a @strawberry.type to it (see: https://github.com/netbox-community/netbox/blob/feature/netbox/ipam/graphql/mixins.py#L13) or in the typing you can just define the field return type (prefix: str)

class AggregateType(NetBoxObjectType, BaseIPAddressFamilyType):
    prefix: str

3. @autotype_decorator fails with the following error for these fields:
   I think the issue here is that you are including a filter-method field in the fields statement, just remove it from the fields= line in the FilterSet.
4. First call to the GraphQL API is extremely slow
   I'll try to repro and take a look.

[peteeckel]

Hi @arthanson, thanks for the detailed response!

1. -> OK, will wait for it.
2. -> I had already added a type for it, works for queries -> OK (see the last update before yours).
3. -> ARGHH ... that fixed it, thanks! Never was a problem before ...
4. -> OK, will wait for it.

When the long response time issue occurs, the worker in question is at 100% CPU for a while, then goes back to normal.

Thanks a lot, that really helped, apart from the active annotation (the above workaround is still usable for this) everything now works as expected ... but there's one thing left still:

$ curl -s -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.excample.com/graphql/ --data '{"query": "query {netbox_dns_zone_list(filters: {rfc2317_prefix: {i_exact: \"10.0.0.0/27\"}}) {view {name}, name}}" }' | json_pp
{
   "data" : null,
   "errors" : [
      {
         "locations" : [
            {
               "column" : 55,
               "line" : 1
            }
         ],
         "message" : "String cannot represent a non string value: {i_exact: \"10.0.0.0/27\"}"
      }
   ]
}

Apparently some conversion from JSON to a netaddr.IPAddress field is required. Do you have any idea how to achieve that?

[peteeckel]

OK, that was filter syntax. This one works differently wrong:

curl -s -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.example.com/graphql/ --data '{"query": "query {netbox_dns_zone_list(filters: {rfc2317_prefix: \"10.0.0.0/27\"}) {view {name}, name}}" }' | json_pp
{
   "data" : null,
   "errors" : [
      {
         "locations" : [
            {
               "column" : 8,
               "line" : 1
            }
         ],
         "message" : "invalid input syntax for type inet: \"{10.0.0.0/27}\"\nLINE 1: ....\"id\") WHERE \"netbox_dns_zone\".\"rfc2317_prefix\" = '{10.0.0.0...\n                                                             ^",
         "path" : [
            "netbox_dns_zone_list"
         ]
      }
   ]
}

For some reason the PostgreSQL query generated for the GraphQL query contains illegal curly braces:

2024-03-25 20:29:32.338 UTC [105462] ERROR:  invalid input syntax for type inet: "{10.0.0.0/27}" at character 348
2024-03-25 20:29:32.338 UTC [105462] STATEMENT:  SELECT "netbox_dns_zone"."id", "netbox_dns_zone"."view_id", "netbox_dns_zone"."name", "netbox_dns_zone"."status" IN ('active') AS "active", "netbox_dns_view"."id", "netbox_dns_view"."name" FROM "netbox_dns_zone" LEFT OUTER JOIN "netbox_dns_view" ON ("netbox_dns_zone"."view_id" = "netbox_dns_view"."id") WHERE "netbox_dns_zone"."rfc2317_prefix" = '{10.0.0.0/27}' ORDER BY "netbox_dns_view"."name" ASC, "netbox_dns_zone"."name" ASC

Fun fact: It works perfectly on model level:

>>> from netbox_dns.models import Zone
>>> Zone.objects.filter(rfc2317_prefix='10.0.0.0/27')
<RestrictedQuerySet [<Zone: 0-31.0.0.10.in-addr.arpa>]>

And via the REST API:

curl "https://netbox.example.com/api/plugins/netbox-dns/zones/?rfc2317_prefix=10.0.0.0%2F27&brief=1" \
     -H 'Authorization: Token xxxxxxxxxxxxxx'

{
  "count": 1,
  "next": null,
  "previous": null,
  "results": [
    {
      "id": 8,
      "url": "http://netbox.example.com/api/plugins/netbox-dns/zones/8/",
      "name": "0-31.0.0.10.in-addr.arpa",
      "view": null,
      "display": "0-31.0.0.10.in-addr.arpa",
      "status": "active",
      "description": "",
      "rfc2317_prefix": "10.0.0.0/27",
      "active": true
    }
  ]
}

[peteeckel]

I could fix the problem with @autotype_decorator failing for the active annotation the same way as for the other filter methods, i.e. by removing it from fields.

And the best thing: It works, out of the box - so item 1 is fixed as well.

[peteeckel]

What do you think the problem with the broken PostgreSQL query comes from:

1. A problem with my query
2. A problem with my custom field class
3. A problem with @autotype_decorator
4. A problem with strawberry/strawberry_django

ATM I'm a bit lost where to start searching ...

[peteeckel]

Found it. I annotated the filter function with the type for rfc2317_prefix:

@strawberry_django.filter(Zone, lookups=True)
@autotype_decorator(ZoneFilterSet)
class NetBoxDNSZoneFilter(BaseFilterMixin):
    rfc2317_prefix: str | None

Now the query is generated correctly and the result is as expected:

$ curl -s -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.example.com/graphql/ --data '{"query": "query {netbox_dns_zone_list(filters: {rfc2317_prefix: \"10.0.0.0/27\"}) {view {name}, name, rfc2317_prefix}}" }' | json_pp
{
   "data" : {
      "netbox_dns_zone_list" : [
         {
            "name" : "0-31.0.0.10.in-addr.arpa",
            "rfc2317_prefix" : "10.0.0.0/27",
            "view" : null
         }
      ]
   }
}

[arthanson]

@peteeckel I've put in a PR to fix issue 1, it will skip fields not defined on the model so you will just need to explicitly define it on the GraphQL filter class. Should hopefully be merged in a day or so.

For item 4, there is a delay in startup time as Strawberry processes all the typing - it seems very slow at it for some reason. It just effects startup, so it is a one-time hit, but we are looking into it. It currently looks like it is an issue with Strawberry, so I may have to open up an issue with them.

[peteeckel]

Re. 1. - with "it actually works" I meant that "you can actually filter for the annotated field", so please check before removing a functionality already present and working :-)

Re. 4. - I suspected something like this. Might also be a consequence of many lazy object class resolutions ...

[peteeckel]

Update regarding the three functional issues: All of them have been fixed thanks to @arthanson's input. I implemented a hacked version of the GraphQL tests (necessary because I needed to modify the GraphQL Query class names and prefixes to avoid name clashes, which is to some extent my own fault) and all of them pass as well.

The only thing that's really annoying is the long startup time. That doesn't only affect GraphQL API operations but everything, from NetBox worker startup time over nbshell to tests. I hope this can be fixed soon - seems there was a similar issue with strawberry a while ago although it's unclear how it was fixed (went away after an update): strawberry-graphql/strawberry#2403

[peteeckel]

Unfortunately this has been un-fixed in one of the last commits ... not sure when it happened as it's difficult to bisect because of intervening migrations.

March 25th:

$ curl -s -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.example.com/graphql/ --data '{"query": "query {netbox_dns_zone_list(filters: {rfc2317_prefix: \"10.0.0.0/27\"}) {view {name}, name, rfc2317_prefix}}" }' | json_pp
{
   "data" : {
      "netbox_dns_zone_list" : [
         {
            "name" : "0-31.0.0.10.in-addr.arpa",
            "rfc2317_prefix" : "10.0.0.0/27",
            "view" : null
         }
      ]
   }
}

April 2nd:

$ curl -s -H "Authorization: Token xxxxxxxxxxxxxx" -H "Content-Type: application/json" -H "Accept: application/json"  https://netbox.example.com/graphql/ --data '{"query": "query {netbox_dns_zone_list(filters: {rfc2317_prefix: \"10.0.0.0/27\"}) {view {name}, name, rfc2317_prefix}}" }' | json_pp
{
   "data" : {
      "netbox_dns_zone_list" : []
   }
}

[peteeckel]

Update: Found it.

It works properly with strawberry-graphql-django-0.37.0, it doesn't with strawberry-graphql-django-0.34.0. So, following your remarks in #15577, I'll postpone that until the underlying issues have been fixed.