Files
headroom/tests/test_persistent_metrics.py
eb9932a671 fix(dashboard): lifetime metrics and refresh (#2374)
## Description

Unifies Dashboard Lifetime and Historical refresh behavior behind one
active-view refresher and canonicalizes durable lifetime waste-signal
names so persisted `json_noise` data is normalized to the current
`json_bloat` key.

Closes #

## 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

- Set dashboard stats polling to 15 seconds and use one active-view
refresh path for Lifetime and Historical views.
- Keep Lifetime and Historical view refreshes at 30 seconds, with
immediate refresh on view switch or keyboard force refresh.
- Normalize persisted waste-signal aliases from `json_noise` to
`json_bloat` and bucket unknown waste signals as `unknown`.
- Bump savings history schema to 6 so existing persisted data is
rewritten with canonical waste-signal keys.
- Add and update tests for waste-signal normalization, schema migration,
dashboard polling markup, and schema-version responses.

## 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
uv run --extra dev pytest tests/test_persistent_metrics.py tests/test_persistent_metrics_persistence.py tests/test_proxy_dashboard_stats_cache.py -q
23 passed, 1 skipped, 1 warning

uv run --extra dev pytest tests/test_proxy_project_savings.py::test_funnel_attributes_savings_from_context_and_stats_exposes_them -q
1 passed, 1 warning

uv run --extra dev pytest tests/test_proxy_savings_history.py -q -k "not save_fsyncs_parent_directory"
41 passed, 1 deselected, 1 warning

uv run --extra dev ruff check headroom/config.py headroom/proxy/persistent_metrics.py headroom/proxy/savings_tracker.py tests/test_persistent_metrics.py tests/test_persistent_metrics_persistence.py tests/test_proxy_savings_history.py tests/test_proxy_dashboard_stats_cache.py tests/test_proxy_project_savings.py
All checks passed!

uv run --extra dev ruff format --check headroom/config.py headroom/proxy/persistent_metrics.py headroom/proxy/savings_tracker.py tests/test_persistent_metrics.py tests/test_persistent_metrics_persistence.py tests/test_proxy_savings_history.py tests/test_proxy_dashboard_stats_cache.py tests/test_proxy_project_savings.py
8 files already formatted

git diff --check headroomlabs/main...HEAD
passed
```

## Real Behavior Proof

- Environment: Windows, Python 3.13.3, local review worktree.
- Exact command / steps: Ran the focused pytest and Ruff commands listed
above against the dashboard, persistent metrics, savings history, and
project savings surfaces touched by this PR.
- Observed result: Focused tests passed, Ruff passed, and whitespace
checks passed. Full `tests/test_proxy_savings_history.py` on Windows
still has the existing directory-fsync assertion issue in
`test_savings_tracker_save_fsyncs_parent_directory`, so the broader
savings-history run excluded only that known platform-specific
assertion.
- Not tested: Browser-driven dashboard interaction; full repository test
suite; full `mypy headroom`.

## 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
- [x] 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

## Additional Notes

Documentation is not needed for this internal dashboard polling and
persistence-normalization fix.
<!-- devin-review-badge-begin -->

---

<a href="https://app.devin.ai/review/headroomlabs-ai/headroom/pull/2374"
target="_blank"><picture><source media="(prefers-color-scheme: dark)"
srcset="https://static.devin.ai/assets/gh-devin-review-dark.svg?v=4"><img
src="https://static.devin.ai/assets/gh-devin-review-light.svg?v=4"
alt="Devin Review"></picture></a>
<!-- devin-review-badge-end -->

### Maintainer current-main requalification

Exact head `2aaeeeb96e8aabd45a7cbd7128227c0714c47df1`, upstream main
`0ddaa56c21cdc01c4774391fdcc69e987ecc4e5c`.

Clean current-main0ddaa56c integration; final diff remains seven
lifetime-metrics/dashboard source/test files. Fresh325tests
pass,12skip,1warning, including ALL16browsercases (zero browser skips)
using existing Playwright1.59.0 overlay and installed Chrome in offline
contexts with service workers blocked. No package/browser installation.
Six changed Python files pass Rufflint/format; candidate offline lock
and whitespace pass. Full native
trackedinputset/everyhash/artifactchecksum verified before3982DLL
reuse,notfreshbuild.

Actual temporary-file load/save/reload proof: prior-schema alias-bearing
file is rewritten at construction to canonical json_bloat30/unknown10
and survives reopening; current-schema file normalizes in memory without
immediate rewrite, then persists canonical data on next record;
stateless prior-schema file remains byte-identical. Do not claim every
current-schema alias file is rewritten immediately. Actual
browser-rendered lifetime counters also show canonical
json_bloat/unknown. Fresh Chrome screenshots visually inspected. Strict
synthetic request fulfillment blocks all other requests, context
offline, no page errors/unexpected requests. Browser interactions
demonstrate no lazy lifetime/history fetch before selecting a view,
active-view refresh after virtual deadline, keyboard forced refresh and
refresh on tab reentry; synthetic counters visibly update. Observed
configuration: statsPollMs5000 and viewPollMs30000 (5-second session
polling,30-second active lifetime/history refresh); earlier15-second
description is not the implemented interval. Virtual clock and synthetic
responses are not production timing/server/billing proof. Initial
missing default Playwright package/browser revision and wrong 'History'
selector were harness issues; corrected to existing overlay, installed
Chrome and actual 'Historical' button. No PR failure attributed to those
attempts.

Skip qualification:11optional-provider cases plus1stale throughput test.
Narrow rerun5pass/1skip reports missing server._throughput_cache export;
generic skip text incorrectly calls it unbuilt Rust core. Server bytes
and throughput test AST match current main, so this is an inherited
test-API limitation, not this PR's introduced defect or native-build
failure. No unrelated throughput behavior rewritten.

Read author/reviewer and outstanding Devin discussion; fresh
disk/browser evidence addresses prior proof gaps, and polling interval
is qualified accurately above. No new unreviewed external findings at
prepublication audit. Source/test delta has no
workflow/setup/dependency/native/CLI paths; publication security gate
passes. Exact published-head CI/Actions/rollup, current-main
ancestry/mergeability and fresh discussion remain before approval.
Maintainers handle CI; no PR merged.


Exact-head hosted checks are being handled by maintainers. No new
finding in the reviewed scope; formal approval awaits exact-head hosted
qualification; nothing merged.


<!-- maintainer-refresh-12864fc8f01a1ca6c58440eccef0dabdd3e595b8 -->

Maintainer current-main refresh: exact head
`abbf9f0744ce7faeb5f46513b6a535b9f5f4c5ee`, authoritative main
`12864fc8f01a1ca6c58440eccef0dabdd3e595b8`. Clean integration patches
exactly match incoming main; contributor behavior preserved. Fresh
focused source tests plus tokenizer/response-header/HF/JSON-parser
incoming-main controls: 109 passed, 1 skipped, 1 warning in 12.17s.
Ruff/format/whitespace pass. Dashboard/template/browser-test inputs are
byte-identical to the prior reviewed head; prior offline Chrome
interaction evidence remains historical, not a fresh browser run. The
one throughput test skip retains the inherited missing
server._throughput_cache API limitation, not an unbuilt-core claim.
Native artifact reuse verified against full recorded input set/every
hash and artifact checksum; no fresh native build claimed. Static
publication gate excludes inherited workflow-validation/native
Docker/Init/Wrap installer paths, with no
workflow/setup/dependency/native/CLI changes in PR diff. Actual
exact-head hosted workflows and final independent review remain pending.
No merge performed.

---------

Co-authored-by: JerrettDavis <mxjerrett@gmail.com>
Co-authored-by: JD Davis <jd@jds-macbook-air.tail2a279.ts.net>
2026-11-07 13:35:26 -06:00

269 lines
9.6 KiB
Python

"""Tests for durable, aggregate-only proxy Lifetime metrics."""
from __future__ import annotations
from datetime import datetime, timezone
import pytest
from headroom.proxy.persistent_metrics import PersistentMetricsState
FIXED_NOW = datetime(2026, 7, 14, 8, 30, tzinfo=timezone.utc)
def _new_state() -> PersistentMetricsState:
return PersistentMetricsState(now=lambda: FIXED_NOW)
def test_snapshot_accumulates_request_token_cache_cost_and_waste_metrics() -> None:
state = _new_state()
state.record_request(
provider="anthropic",
stack="codex",
model="claude-test",
input_tokens=100,
output_tokens=20,
attempted_input_tokens=150,
tokens_saved=50,
cached=True,
cache_read_tokens=80,
cache_write_tokens=40,
cache_write_5m_tokens=10,
cache_write_1h_tokens=30,
uncached_input_tokens=20,
input_usd=0.4,
compression_savings_usd=0.2,
cache_savings_usd=0.1,
waste_signals={"repetition": 7},
)
state.record_failed(provider="anthropic", model="claude-test")
state.record_rate_limited(provider="anthropic", model="claude-test")
state.record_cache_bust(tokens_lost=9)
state.record_cache_miss(provider="anthropic", reason="prefix_change")
snapshot = state.snapshot(persistence={"enabled": True, "healthy": True})
assert snapshot["scope"] == "lifetime"
assert snapshot["requests"] == {
"total": 1,
"cached": 1,
"failed": 1,
"rate_limited": 1,
# Lifetime carries the same splits the Prometheus labels do.
"failed_by_provider": {"anthropic": 1},
"rate_limited_by_source": {"headroom": 1},
"by_provider": {"anthropic": 1},
"by_stack": {"codex": 1},
}
assert snapshot["tokens"] == {
"input": 100,
"output": 20,
"attempted_input": 150,
"saved": 50,
"token_savings_percent": pytest.approx(50 / 150 * 100),
}
assert snapshot["prefix_cache"]["requests"] == 1
assert snapshot["prefix_cache"]["hit_requests"] == 1
assert snapshot["prefix_cache"]["cache_read_tokens"] == 80
assert snapshot["prefix_cache"]["cache_write_tokens"] == 40
assert snapshot["prefix_cache"]["cache_hit_rate"] == 100.0
assert snapshot["prefix_cache"]["ttl_1h_percent"] == 75.0
assert snapshot["prefix_cache"]["ttl_5m_percent"] == 25.0
assert snapshot["prefix_cache"]["bust_count"] == 1
assert snapshot["prefix_cache"]["bust_tokens"] == 9
assert snapshot["prefix_cache"]["misses_by_reason"] == {"prefix_change": 1}
assert snapshot["cost"] == {
"input_usd": 0.4,
"compression_savings_usd": 0.2,
# This call records no list ceiling, so the ceiling defaults to the
# cache-aware figure -- the truthful reading of "these are the same
# number" for a request with no cache mix to price against. The basis
# stays `unknown` because the caller named none: a fresh aggregate must
# not claim its untouched total was list-priced.
"compression_savings_list_usd": 0.2,
"savings_basis": "unknown",
"cache_savings_usd": 0.1,
}
assert snapshot["waste_signals"] == {"repetition": 7}
assert snapshot["by_model"]["claude-test"]["input_tokens"] == 100
def test_snapshot_uses_null_for_ratios_without_a_denominator() -> None:
snapshot = _new_state().snapshot(persistence={"enabled": True, "healthy": True})
assert snapshot["tokens"]["token_savings_percent"] is None
assert snapshot["prefix_cache"]["cache_hit_rate"] is None
assert snapshot["prefix_cache"]["ttl_1h_percent"] is None
assert snapshot["prefix_cache"]["ttl_5m_percent"] is None
def test_candidate_models_remain_available_until_the_two_hundred_and_first_model() -> None:
state = _new_state()
for index in range(200):
state.record_request(
provider="provider",
stack="stack",
model=f"model-{index:03}",
input_tokens=index + 1,
)
persisted = state.to_dict()
snapshot = state.snapshot(persistence={"enabled": True, "healthy": True})
assert len(persisted["models"]["tracked"]) == 200
assert "model-000" not in snapshot["by_model"]
assert snapshot["by_model"]["other"]["input_tokens"] == sum(range(1, 101))
def test_two_hundred_and_first_model_permanently_compacts_non_top_candidates() -> None:
state = _new_state()
for index in range(201):
state.record_request(
provider="provider",
stack="stack",
model=f"model-{index:03}",
input_tokens=index + 1,
)
persisted = state.to_dict()
snapshot = state.snapshot(persistence={"enabled": True, "healthy": True})
assert len(persisted["models"]["tracked"]) == 100
assert set(snapshot["by_model"]) == {
*(f"model-{index:03}" for index in range(101, 201)),
"other",
}
assert snapshot["by_model"]["other"]["input_tokens"] == sum(range(1, 102))
def test_state_normalizes_invalid_values_and_unknown_dimension_labels() -> None:
state = PersistentMetricsState(
{
"requests": {"total": "not-a-number"},
"tokens": {"input": float("nan"), "output": -3},
"models": {"tracked": {"unknown": {"input_tokens": "7"}}},
},
now=lambda: FIXED_NOW,
)
state.record_request(
provider=" ",
stack=None,
model=" ",
input_tokens=-1,
output_tokens=float("inf"),
waste_signals={"unrecognized": 9},
)
snapshot = state.snapshot(persistence={"enabled": True, "healthy": True})
assert snapshot["tokens"]["input"] == 0
assert snapshot["tokens"]["output"] == 0
assert snapshot["requests"]["by_provider"] == {"other": 1}
assert snapshot["requests"]["by_stack"] == {"other": 1}
assert snapshot["by_model"]["other"]["input_tokens"] == 7
assert snapshot["waste_signals"] == {"unknown": 9}
def test_json_bloat_waste_signal_is_a_named_bucket() -> None:
"""The parser emits ``json_bloat`` (WasteSignals.to_dict); it must not fall to ``other``."""
state = _new_state()
state.record_request(
provider="anthropic",
stack="claude",
model="claude-sonnet-5",
waste_signals={"json_bloat": 500, "html_noise": 20},
)
snapshot = state.snapshot(persistence={"enabled": True, "healthy": True})
assert snapshot["waste_signals"] == {"json_bloat": 500, "html_noise": 20}
def test_state_canonicalizes_lifetime_waste_signal_aliases_and_fallbacks() -> None:
state = PersistentMetricsState(
{
"waste_signals": {
"json_noise": 10,
"json_bloat": 20,
"other": 7,
"unrecognized_persisted": 3,
},
},
now=lambda: FIXED_NOW,
)
state.record_request(
provider="openai",
stack="codex",
model="gpt-test",
waste_signals={"json_bloat": 5, "unrecognized_runtime": 2},
)
snapshot = state.snapshot(persistence={"enabled": True, "healthy": True})
assert snapshot["waste_signals"] == {"json_bloat": 35, "unknown": 12}
def test_miss_reasons_still_fall_back_to_unknown_on_reload() -> None:
state = PersistentMetricsState(
{"prefix_cache": {"misses_by_reason": {"ttl_expiry": 2, "bogus": 1}}},
now=lambda: FIXED_NOW,
)
snapshot = state.snapshot(persistence={"enabled": True, "healthy": True})
assert snapshot["prefix_cache"]["misses_by_reason"] == {"ttl_expiry": 2, "unknown": 1}
def test_rate_limited_is_split_by_source_and_failures_by_provider() -> None:
"""Headroom's own limiter and an upstream 429 must stay distinguishable.
Both land in the same ``rate_limited`` total (unchanged), but an operator
acts on them differently: ours firing means raise the cap, the provider's
means back off. See issue #3696.
"""
state = _new_state()
state.record_rate_limited(provider="anthropic", source="headroom")
state.record_rate_limited(provider="anthropic", source="upstream")
state.record_rate_limited(provider="openai", source="upstream")
state.record_failed(provider="anthropic")
state.record_failed(provider="openai")
state.record_failed(provider="openai")
requests = state.snapshot(persistence={"enabled": True, "healthy": True})["requests"]
assert requests["rate_limited"] == 3
assert requests["rate_limited_by_source"] == {"headroom": 1, "upstream": 2}
assert requests["failed"] == 3
assert requests["failed_by_provider"] == {"anthropic": 1, "openai": 2}
def test_rate_limit_source_defaults_to_headroom_and_clamps_unknown_values() -> None:
state = _new_state()
state.record_rate_limited(provider="anthropic")
state.record_rate_limited(provider="anthropic", source="bogus")
requests = state.snapshot(persistence={"enabled": True, "healthy": True})["requests"]
assert requests["rate_limited_by_source"] == {"headroom": 2}
def test_pre_3696_state_loads_without_backfilling_invented_history() -> None:
"""A state file written before the split keeps its totals and starts the maps empty.
Back-filling the legacy ``rate_limited`` total into one bucket would claim we
observed a split we never recorded.
"""
state = PersistentMetricsState(
{"requests": {"total": 10, "failed": 4, "rate_limited": 7}},
now=lambda: FIXED_NOW,
)
requests = state.snapshot(persistence={"enabled": True, "healthy": True})["requests"]
assert requests["rate_limited"] == 7
assert requests["failed"] == 4
assert requests["rate_limited_by_source"] == {}
assert requests["failed_by_provider"] == {}