resourcegroupstaggingapi get_resources() returns duplicate entries in server mode (not reproducible with @mock_aws)

[Jbat1Jumper]

Environment

• moto: 5.2.2 (installed via moto[server])
• Python: 3.12.3
• boto3: 1.40.49
• Reproduced with moto_server (HTTP server mode). NOT reproducible with the @mock_aws decorator — see below.

How to reproduce

Against a running moto_server (default settings, MOTO_ACCOUNT_ID unset or set, doesn't matter):

import boto3

ENDPOINT = "http://localhost:4566"
REGION = "us-east-1"

sns = boto3.client("sns", endpoint_url=ENDPOINT, region_name=REGION)
rgt = boto3.client("resourcegroupstaggingapi", endpoint_url=ENDPOINT, region_name=REGION)

topic_arn = sns.create_topic(Name="repro-topic")["TopicArn"]
sns.tag_resource(ResourceArn=topic_arn, Tags=[{"Key": "instance", "Value": "test-instance"}])

result = rgt.get_resources(
    ResourceTypeFilters=["sns"],
    TagFilters=[{"Key": "instance", "Values": ["test-instance"]}],
)
for entry in result["ResourceTagMappingList"]:
    print(entry["ResourceARN"])
print(len(result["ResourceTagMappingList"]), "entries")

Actual output (server mode):

arn:aws:sns:us-east-1:000000000000:repro-topic
arn:aws:sns:us-east-1:000000000000:repro-topic
2 entries

Expected: 1 entry — the topic was only created/tagged once.

Notably, this does NOT reproduce with the decorator

The same create → tag → get_resources sequence, run in-process with @mock_aws instead of against moto_server, returns exactly one clean entry:

import boto3
from moto import mock_aws

with mock_aws():
    sns = boto3.client("sns", region_name="us-east-1")
    topic_arn = sns.create_topic(Name="repro-topic")["TopicArn"]
    sns.tag_resource(ResourceArn=topic_arn, Tags=[{"Key": "instance", "Value": "test-instance"}])

    rgt = boto3.client("resourcegroupstaggingapi", region_name="us-east-1")
    result = rgt.get_resources(
        ResourceTypeFilters=["sns"],
        TagFilters=[{"Key": "instance", "Values": ["test-instance"]}],
    )
    print(len(result["ResourceTagMappingList"]), "entries")  # -> 1

I also checked iter_taggable_backends()/iter_tagged_resources() directly in-process under @mock_aws and got exactly one backend and one topic back, so the core resource-tagging logic (moto/core/resource_tagging.py) looks correct — the duplication seems to be introduced somewhere in the server-mode request dispatch path specifically, not in the shared model/generator logic.

This isn't specific to having 2 TagFilters — a single filter is enough to trigger it, which rules out a per-filter-yield bug in make_tag_matcher/_get_resources_generator.

Impact

Any consumer using moto_server (Docker image, CI, etc.) with a workflow that creates a tagged resource and then queries it back via resourcegroupstaggingapi.get_resources() gets duplicate ResourceTagMappingList entries for that resource.

[bpandola]

@Jbat1Jumper I can't reproduce this on Moto 5.2.2 (or on master). I get the one expected entry when running Moto in server mode.

If you open a PR with a failing test case, we can see if it fails our server mode CI. If you provide more detail on how you're running Moto in server mode (via command, via Docker, etc.) and how you're executing your test against Moto server, maybe something will stand out that might be causing this behavior you're seeing.

[Jbat1Jumper]

Dug a little bit more and I think the issue was because I had a server running on a separate process for long enough to hit this issue. I was also setting a custom MOTO_ACCOUNT_ID because I was testing against specific ARNs and I think that was the trigger.

Added this test here #10179 which should catch the core issue.