fix(proxy): skip pricing lookup for passthrough:* models (#2585)

## Description

Internal synthetic model names such as `passthrough:count_tokens` are
used for passthrough endpoint accounting. They are not billable models,
but cost lookup still sent them through LiteLLM and logged pricing
failures, and they could appear as pseudo rows in per-model token
savings stats.

Closes #2578

## Type of Change

- [x] Bug fix (non-breaking change that fixes an issue)
- [ ] New feature (non-breaking change that adds functionality)
- [ ] Breaking change (fix or feature that would cause existing
functionality to change)
- [ ] Documentation update
- [ ] Performance improvement
- [ ] Code refactoring (no functional changes)

## Changes Made

- Short-circuit `CostTracker.estimate_cost` for models prefixed with
`passthrough:` so LiteLLM is never called and no pricing WARNING is
emitted.
- Omit `passthrough:*` models from `CostTracker.stats()` per-model rows
and related aggregate fields that feed dashboard cost/savings views.
- Add regression tests for silent skip and stats omission.

## Testing

- [x] Unit tests pass (`pytest`)
- [x] Linting passes (`ruff check .`)
- [ ] Type checking passes (`mypy headroom`)
- [x] New tests added for new functionality
- [ ] Manual testing performed

### Test Output

```text
$ python -m pytest tests/test_cost_pricing_warning_dedup.py -q
============================= test session starts ==============================
platform linux -- Python 3.11.15, pytest-9.1.0, pluggy-1.6.0
rootdir: /tmp/headroom-pr/headroom
configfile: pyproject.toml
collected 5 items

tests/test_cost_pricing_warning_dedup.py .....                           [100%]

========================= 5 passed, 1 warning in 0.26s =========================

$ python -m ruff check headroom/proxy/cost.py tests/test_cost_pricing_warning_dedup.py
All checks passed!

$ python -m ruff format --check headroom/proxy/cost.py tests/test_cost_pricing_warning_dedup.py
2 files already formatted
```

## Real Behavior Proof

- Environment: Linux x86_64, Python 3.11.15, shallow checkout of
headroomlabs-ai/headroom main at 904bc675 plus this branch commit
13526dc
- Exact command / steps: `python -m pytest
tests/test_cost_pricing_warning_dedup.py -q`; `python -m ruff check
headroom/proxy/cost.py tests/test_cost_pricing_warning_dedup.py`;
`python -m ruff format --check headroom/proxy/cost.py
tests/test_cost_pricing_warning_dedup.py`
- Observed result: 5 targeted tests passed, including passthrough skip
without warning and stats omitting passthrough models; ruff check and
format check clean on touched files
- Not tested: live `headroom proxy` session with Claude Code
count_tokens traffic; full monorepo `make ci-precheck`; mypy

## Review Readiness

- [x] I have performed a self-review
- [x] This PR is ready for human review

## Checklist

- [x] My code follows the project's style guidelines
- [x] I have performed a self-review of my code
- [ ] I have commented my code, particularly in hard-to-understand areas
- [ ] I have made corresponding changes to the documentation
- [x] My changes generate no new warnings
- [x] I have added tests that prove my fix is effective or that my
feature works
- [x] New and existing unit tests pass locally with my changes
- [x] I did **not** edit `CHANGELOG.md` - it is generated by
release-please from my Conventional Commit PR title (a CI guard enforces
this)

## Screenshots (if applicable)

N/A

Co-authored-by: JD Davis <mxjerrett@gmail.com>
This commit is contained in:
AxelRay
2026-10-01 00:13:35 -05:00
committed by GitHub
co-authored by JD Davis
parent 6ff74c5db5
commit f87848cf78
2 changed files with 47 additions and 2 deletions
+19 -2
View File
@@ -965,6 +965,9 @@ class CostTracker:
cache_read_tokens: Tokens served from cache (~10% of input rate)
cache_write_tokens: Tokens written to cache (~125% of input rate)
"""
if model.startswith("passthrough:"):
return None
litellm = _get_litellm_module()
if litellm is None:
_warn_pricing_once(
@@ -1479,6 +1482,8 @@ class CostTracker:
# ``_tokens_saved_by_model`` alone would drop such a model from the table
# entirely rather than merely under-report it.
for model in sorted(set(self._tokens_saved_by_model) | set(self._tool_saved_by_model)):
if model.startswith("passthrough:"):
continue
compression_saved = self._tokens_saved_by_model.get(model, 0)
tool_saved = self._tool_saved_by_model.get(model, 0)
# What the "Tokens Saved" column means: everything Headroom kept off
@@ -1513,6 +1518,8 @@ class CostTracker:
total_billed_input_tokens = 0
total_input_tokens = 0
for model in self._tokens_saved_by_model:
if model.startswith("passthrough:"):
continue
saved = self._tokens_saved_by_model[model]
sent = self._tokens_sent_by_model.get(model, 0)
cr = self._api_cache_read_by_model.get(model, 0)
@@ -1546,6 +1553,8 @@ class CostTracker:
# cache-aware valuation below is reported alongside it.
savings_usd = 0.0
for model in self._tokens_saved_by_model:
if model.startswith("passthrough:"):
continue
saved = self._tokens_saved_by_model[model]
if saved <= 0:
continue
@@ -1635,8 +1644,16 @@ class CostTracker:
"total_tool_tokens_saved": total_tool_saved,
"total_input_tokens": total_input_tokens,
"total_input_cost_usd": round(cost_with_headroom, 4),
"cache_write_5m_tokens": sum(self._api_cache_write_5m_by_model.values()),
"cache_write_1h_tokens": sum(self._api_cache_write_1h_by_model.values()),
"cache_write_5m_tokens": sum(
tokens
for model, tokens in self._api_cache_write_5m_by_model.items()
if not model.startswith("passthrough:")
),
"cache_write_1h_tokens": sum(
tokens
for model, tokens in self._api_cache_write_1h_by_model.items()
if not model.startswith("passthrough:")
),
"per_model": per_model,
# Input-only, unchanged: budgets, the per-model table and the
# persistent savings tracker all read it as input spend.
+28
View File
@@ -63,3 +63,31 @@ def test_litellm_unavailable_warns_once_per_model(monkeypatch, caplog):
unavailable = [r for r in caplog.records if "LiteLLM not available" in r.getMessage()]
assert len(unavailable) == 1
def test_passthrough_model_skips_pricing_without_warning(monkeypatch, caplog):
import headroom.proxy.cost as cost_mod
class _FakeLiteLLM:
@staticmethod
def cost_per_token(**_kwargs):
raise AssertionError("cost_per_token must not be called")
monkeypatch.setattr(cost_mod, "_get_litellm_module", lambda: _FakeLiteLLM())
tracker = cost_mod.CostTracker()
with caplog.at_level(logging.WARNING, logger="headroom.proxy"):
assert tracker.estimate_cost("passthrough:count_tokens", 100, 50) is None
assert not caplog.records
def test_stats_omit_passthrough_models(cost_tracker):
cost_tracker.record_tokens("passthrough:count_tokens", 10, 90)
cost_tracker.record_tokens("glm-5.2", 20, 80)
stats = cost_tracker.stats()
assert set(stats["per_model"]) == {"glm-5.2"}
assert stats["total_tokens_saved"] == 20
assert stats["total_input_tokens"] == 80