mirror of
https://github.com/headroomlabs-ai/headroom.git
synced 2026-08-27 14:17:10 -04:00
## Description Anthropic signs the thinking **block**, not the request — the signature covers that block's own content. #2254 responded to real 400s by freezing the **entire body** whenever any thinking block appeared anywhere in history. That protects bytes no signature covers, including top-level `tools` and `system`, which are not even inside `messages`. Measured on 227,777 lines of real proxy logs from a user reporting ~1% savings: - **618 of 1,802 requests (34.3%)** had every computed compression discarded. **100%** were `client=claude-code`; Codex/GPT traffic was untouched. - One session logged turn 1 saving 428 tokens, then **229 consecutive turns saving exactly 0**. - **491.9s — 34.2% of all optimization time** — was spent computing compressions that were then thrown away. One request paid 21.2s to compute a real 8.0% reduction that never shipped. - It orphaned the turn-1 cache prefix on **12 of 35** sessions, corroborated by Headroom's own `CACHE-MISS-ATTRIBUTION` events (21/21 are `reason=prefix_change`, **none** TTL expiry), with an exact token match: `expected_cached=27,541` equalling turn 1's write. ## Changes Made - Replace the presence test with a **positional, order-sensitive fingerprint** of every `thinking` / `redacted_thinking` block, compared against the client's original. Byte-equal blocks → forward the edits. Any difference (edited text, edited signature, dropped, reordered, moved) or any failure to prove equality → today's verbatim passthrough. Keys are sorted so a dict rebuilt in a different order is not mistaken for an edit. - `outbound_body_is_client_bytes` mirrors the relaxation exactly, or the CCR buffering probe and the forwarder would disagree and re-create #2952 in reverse. - The #2990/#3015 accounting reset now **recomputes** the lock immediately before use instead of reusing the probe taken before the CCR branch. The predicate tests block *content* now, and `enforce_cache_control_ttl_order` rewrites `body["messages"]` in between, so the early answer can go stale. (Latent before this PR; load-bearing after.) - **Perf:** parse the client body once per decision, plus a substring prescreen. A 9.3 MB body (the real production maximum) could otherwise be parsed four times per request on a stage that already carries a 30s timeout whose expiry quarantines compression process-wide. ## Rollout safety **On by default at the maintainer's explicit direction.** `HEADROOM_THINKING_PRESERVING_MUTATIONS=0` restores the previous blanket lock with no deploy. The risk is recorded in the module rather than smoothed over: #2254's stated cause — a plain canonical re-encode — cannot alter parsed values and therefore cannot by itself invalidate a signature, and that report's own log shows a transform (`tool_search_deferral`) firing on the failing turn. So the stated cause does not hold up, **but the failure was real and its true trigger was never isolated.** This relaxation is strictly narrower than what broke: it forwards edits only when every block is provably identical, which is the property the blanket rule was a crude proxy for. ## Testing ```text uv run pytest tests/test_proxy_byte_faithful_forwarding.py tests/test_ccr_buffered_stream_signed_thinking.py \ tests/test_proxy/test_anthropic_ccr_deferred_injection.py 92 passed uv run mypy headroom/proxy/body_forwarding.py headroom/proxy/handlers/anthropic.py # Success uv run ruff check . && ruff format --check . # clean ``` Existing tests that encoded the blanket lock were **re-pointed at the correct trigger, not deleted** — each now tampers with a thinking block so it still guards what it was written for. `test_signed_thinking_discarded_mutation_uses_wire_truth_for_all_accounting` (#3015) now runs under the kill switch, which proves both that the accounting neutralisation still works and that the env-var rollback is a complete restoration. ## Real behavior proof - **Setup:** macOS arm64, Python 3.12, this branch, byte-capturing transport. - **After-fix evidence** — end-to-end through `/v1/messages` with a signed thinking block in history and a compactable tool schema (`test_untouched_thinking_lets_tool_compaction_reach_the_wire`): the annotation keys the compaction strips (`$schema`, `title`) are **absent from the captured upstream bytes**, and `wire["messages"][1]["content"][0]` is **byte-identical to the client's signed block**. Under the kill switch the same request forwards the client's bytes unchanged with accounting zeroed. - **Parse-count measured, not assumed:** 7.2 MB thinking-bearing body → 2 parses became 1. 2 MB body with no thinking blocks (~2 of 3 requests) → 1 parse became **0**, i.e. faster than before this feature existed. - **Projected effect on the reporting user's traffic**, derived from their unlocked requests: Claude Code headline **2.27% → roughly 5–6%**. Their unlocked requests already achieve 5.62% overall and 7.2–7.4% in the 20K–150K band, which matches our fleet beacon (~8%); the 2.27% is a blend where 60% of tokens sat in requests that shipped nothing. - **NOT tested: live paid Anthropic traffic with a real signed thinking block.** This is the one thing that matters most and I could not do it here. The signature-verification behaviour is Anthropic's, and no local test can prove it accepts a re-serialized body carrying an untouched block. **Please validate on live traffic before relying on the default.** Watch for 400 `invalid_request_error` mentioning `thinking`, and `CACHE-MISS-ATTRIBUTION reason=prefix_change` rates. ## Known risk not eliminated Enabling this changes the wire bytes for in-flight sessions, so expect a **one-time prefix change** on the first affected turn of each live conversation. Supporting evidence that this is bounded: canonical serialization is already the norm for the ~66% of traffic without thinking blocks, and that traffic sustains a 94.3% cache hit rate. 🤖 Generated with [Claude Code](https://claude.com/claude-code) --------- Co-authored-by: Tejas Chopra <tejas@Tejass-MacBook-Pro.local> Co-authored-by: Claude Opus 5 <noreply@anthropic.com>
339 lines
12 KiB
Python
339 lines
12 KiB
Python
"""Buffered-CCR streaming vs. byte-faithful passthrough (issue #2952).
|
|
|
|
The buffered-CCR path is the one place the Anthropic handler changes the
|
|
request *for its own benefit*: it flips ``stream`` to False so the reply comes
|
|
back as one JSON document it can inspect for ``headroom_retrieve`` calls, then
|
|
resynthesizes SSE for the client.
|
|
|
|
That only works if the flip reaches the wire. When conversation history carries
|
|
a signed ``thinking`` block, ``select_outbound_body`` forwards the client's
|
|
original bytes instead — ``"stream": true`` and all — so upstream streams, the
|
|
JSON parse fails, resynthesis is skipped, and the client is left with a 200 and
|
|
nothing it can read. These tests pin the three defenses: don't take the path,
|
|
survive the reply if we somehow do, and never cache a body in the wrong format.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import asyncio
|
|
import json
|
|
from datetime import datetime
|
|
|
|
import pytest
|
|
|
|
fastapi = pytest.importorskip("fastapi")
|
|
httpx = pytest.importorskip("httpx")
|
|
|
|
from fastapi.testclient import TestClient # noqa: E402
|
|
|
|
from headroom.proxy.models import CacheEntry # noqa: E402
|
|
from headroom.proxy.server import ProxyConfig, create_app # noqa: E402
|
|
|
|
RETRIEVE_TOOL = {
|
|
"name": "headroom_retrieve",
|
|
"description": "Retrieve original content",
|
|
"input_schema": {"type": "object", "properties": {}},
|
|
}
|
|
|
|
SIGNED_THINKING_TURN = {
|
|
"role": "assistant",
|
|
"content": [
|
|
{
|
|
"type": "thinking",
|
|
"thinking": "private reasoning",
|
|
"signature": "sig-abc123",
|
|
},
|
|
{"type": "text", "text": "Answered."},
|
|
],
|
|
}
|
|
|
|
SSE_BODY = (
|
|
b'event: message_start\ndata: {"type":"message_start","message":{"id":"msg_1"}}\n\n'
|
|
b'event: message_stop\ndata: {"type":"message_stop"}\n\n'
|
|
)
|
|
|
|
|
|
def _config() -> ProxyConfig:
|
|
return ProxyConfig(
|
|
optimize=False,
|
|
cache_enabled=True,
|
|
rate_limit_enabled=False,
|
|
memory_enabled=False,
|
|
)
|
|
|
|
|
|
@pytest.fixture
|
|
def ccr_marker() -> str:
|
|
"""A marker this proxy actually owns, so retrieval could really fire.
|
|
|
|
The buffered path is only taken when the outgoing body carries a redeemable
|
|
marker (#3071) — ``headroom_retrieve`` has nothing to expand otherwise. These
|
|
tests are about what happens *on* that path, so they have to earn it.
|
|
"""
|
|
from headroom.cache.backends import InMemoryBackend
|
|
from headroom.cache.compression_store import get_compression_store, reset_compression_store
|
|
|
|
reset_compression_store()
|
|
store = get_compression_store(backend=InMemoryBackend())
|
|
hash_key = store.store(
|
|
"the original, uncompressed tool output",
|
|
"<<ccr:placeholder>>",
|
|
original_tokens=100,
|
|
compressed_tokens=5,
|
|
tool_name="Read",
|
|
)
|
|
try:
|
|
yield hash_key
|
|
finally:
|
|
reset_compression_store()
|
|
|
|
|
|
def _body(*, with_thinking: bool, marker: str | None = None) -> dict:
|
|
first = "hi" if marker is None else f"hi — earlier output is at <<ccr:{marker}>>"
|
|
messages: list[dict] = [{"role": "user", "content": first}]
|
|
if with_thinking:
|
|
messages.append(SIGNED_THINKING_TURN)
|
|
messages.append({"role": "user", "content": "continue"})
|
|
return {
|
|
"model": "claude-sonnet-4-20250514",
|
|
"max_tokens": 64,
|
|
"stream": True,
|
|
"tools": [RETRIEVE_TOOL],
|
|
"messages": messages,
|
|
}
|
|
|
|
|
|
def _headers() -> dict[str, str]:
|
|
return {"Authorization": "Bearer test-key", "x-api-key": "test-key"}
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("with_thinking", "relaxation_enabled", "expect_plain_streaming"),
|
|
[
|
|
# Locked (kill switch engaged): the flip cannot reach upstream, so
|
|
# buffering would ask for a stream and then parse it as JSON, stranding
|
|
# the client with an unreadable 200. This is #2952 exactly.
|
|
(True, False, True),
|
|
# Relaxed (default): no transform touched the thinking block, so the
|
|
# flip DOES land and buffered retrieval becomes the coherent choice --
|
|
# the outcome #2952 wanted before the blanket lock made it unreachable.
|
|
(True, True, False),
|
|
# No thinking block: unaffected in either regime.
|
|
(False, True, False),
|
|
(False, False, False),
|
|
],
|
|
)
|
|
def test_signed_thinking_history_skips_the_buffered_ccr_path(
|
|
with_thinking: bool,
|
|
relaxation_enabled: bool,
|
|
expect_plain_streaming: bool,
|
|
ccr_marker: str,
|
|
monkeypatch: pytest.MonkeyPatch,
|
|
) -> None:
|
|
"""The buffered path is only chosen when the stream:false flip can land."""
|
|
monkeypatch.setenv("HEADROOM_THINKING_PRESERVING_MUTATIONS", "1" if relaxation_enabled else "0")
|
|
calls: dict[str, object] = {}
|
|
|
|
async def fake_stream_response(url, headers, body, *args, **kwargs): # noqa: ANN001
|
|
calls["stream_body"] = body
|
|
return fastapi.responses.StreamingResponse(iter([SSE_BODY]), media_type="text/event-stream")
|
|
|
|
async def fake_retry(method, url, headers, req_body, *args, **kwargs): # noqa: ANN001
|
|
calls["buffered_body"] = json.loads(json.dumps(req_body))
|
|
return httpx.Response(
|
|
200,
|
|
json={
|
|
"id": "msg_1",
|
|
"type": "message",
|
|
"role": "assistant",
|
|
"model": "claude-sonnet-4-20250514",
|
|
"content": [{"type": "text", "text": "ok"}],
|
|
"stop_reason": "end_turn",
|
|
"usage": {"input_tokens": 10, "output_tokens": 5},
|
|
},
|
|
headers={"content-type": "application/json"},
|
|
)
|
|
|
|
app = create_app(_config())
|
|
with TestClient(app) as client:
|
|
client.app.state.proxy._stream_response = fake_stream_response
|
|
client.app.state.proxy._retry_request = fake_retry
|
|
resp = client.post(
|
|
"/v1/messages",
|
|
json=_body(with_thinking=with_thinking, marker=ccr_marker),
|
|
headers=_headers(),
|
|
)
|
|
|
|
assert resp.status_code == 200, resp.text
|
|
if expect_plain_streaming:
|
|
# Passthrough is locked in, so we must not pretend we can buffer.
|
|
assert "stream_body" in calls, "expected the plain streaming path"
|
|
assert "buffered_body" not in calls
|
|
# The turn still leaves as a streaming request, matching the bytes
|
|
# that passthrough will actually forward.
|
|
assert calls["stream_body"]["stream"] is True
|
|
else:
|
|
assert "buffered_body" in calls, "expected the buffered CCR path"
|
|
assert calls["buffered_body"]["stream"] is False
|
|
|
|
|
|
@pytest.mark.parametrize("upstream_delay", [0.0, 1.2], ids=["prompt", "past-keepalive"])
|
|
def test_buffered_ccr_relays_an_unexpected_sse_reply_and_does_not_cache_it(
|
|
upstream_delay: float, ccr_marker: str
|
|
) -> None:
|
|
"""A 200 SSE reply on the buffered path reaches the client as a stream.
|
|
|
|
The delay matters: ``_BufferedCCRResponse`` commits SSE response headers
|
|
after a 1 s keepalive, and past that point it can only forward a result
|
|
that exposes a ``body_iterator``. A plain ``Response`` there degrades to a
|
|
bare ``event: error`` — which is what a real (multi-second) Anthropic turn
|
|
hit in #2952.
|
|
"""
|
|
|
|
async def fake_retry(method, url, headers, req_body, *args, **kwargs): # noqa: ANN001
|
|
if upstream_delay:
|
|
await asyncio.sleep(upstream_delay)
|
|
return httpx.Response(200, content=SSE_BODY, headers={"content-type": "text/event-stream"})
|
|
|
|
app = create_app(_config())
|
|
with TestClient(app) as client:
|
|
proxy = client.app.state.proxy
|
|
proxy._retry_request = fake_retry
|
|
resp = client.post(
|
|
"/v1/messages",
|
|
json=_body(with_thinking=False, marker=ccr_marker),
|
|
headers=_headers(),
|
|
)
|
|
|
|
assert resp.status_code == 200, resp.text
|
|
assert resp.headers["content-type"].startswith("text/event-stream")
|
|
assert b"message_start" in resp.content
|
|
# Caching SSE bytes under a key with no `stream` component is what
|
|
# served a stream to a buffered caller in the first place.
|
|
assert proxy.cache is not None
|
|
assert len(proxy.cache._cache) == 0, "an unparseable body must never be cached"
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
("marker_kind", "expect_buffered"),
|
|
[
|
|
("owned", True),
|
|
("none", False),
|
|
("foreign", False),
|
|
],
|
|
)
|
|
def test_buffering_is_gated_on_a_redeemable_marker(
|
|
marker_kind: str, expect_buffered: bool, ccr_marker: str
|
|
) -> None:
|
|
"""A resident ``headroom_retrieve`` is not on its own a reason to buffer (#3071).
|
|
|
|
The tool is injected once and kept resident so the tools array stays
|
|
byte-stable for the prompt cache. Buffering on its presence alone meant
|
|
every later streaming turn of a sticky session lost incremental delivery —
|
|
time-to-first-byte became the whole generation. Retrieval can only expand a
|
|
marker that is in the outgoing body *and* redeemable now, so that is what
|
|
the wire-format decision keys on.
|
|
"""
|
|
marker = {
|
|
"owned": ccr_marker,
|
|
"none": None,
|
|
# Correct shape, not ours: adopting it would send the model to a
|
|
# retrieval that is guaranteed to miss (#2836).
|
|
"foreign": "deadbeefcafe",
|
|
}[marker_kind]
|
|
|
|
calls: dict[str, object] = {}
|
|
|
|
async def fake_stream_response(url, headers, body, *args, **kwargs): # noqa: ANN001
|
|
calls["stream_body"] = body
|
|
return fastapi.responses.StreamingResponse(iter([SSE_BODY]), media_type="text/event-stream")
|
|
|
|
async def fake_retry(method, url, headers, req_body, *args, **kwargs): # noqa: ANN001
|
|
calls["buffered_body"] = json.loads(json.dumps(req_body))
|
|
return httpx.Response(
|
|
200,
|
|
json={
|
|
"id": "msg_1",
|
|
"type": "message",
|
|
"role": "assistant",
|
|
"model": "claude-sonnet-4-20250514",
|
|
"content": [{"type": "text", "text": "ok"}],
|
|
"stop_reason": "end_turn",
|
|
"usage": {"input_tokens": 10, "output_tokens": 5},
|
|
},
|
|
headers={"content-type": "application/json"},
|
|
)
|
|
|
|
app = create_app(_config())
|
|
with TestClient(app) as client:
|
|
client.app.state.proxy._stream_response = fake_stream_response
|
|
client.app.state.proxy._retry_request = fake_retry
|
|
resp = client.post(
|
|
"/v1/messages",
|
|
json=_body(with_thinking=False, marker=marker),
|
|
headers=_headers(),
|
|
)
|
|
|
|
assert resp.status_code == 200, resp.text
|
|
if expect_buffered:
|
|
assert "buffered_body" in calls, "a redeemable marker must still buffer"
|
|
assert calls["buffered_body"]["stream"] is False
|
|
else:
|
|
assert "stream_body" in calls, "nothing to retrieve — the client must keep streaming"
|
|
assert "buffered_body" not in calls
|
|
assert calls["stream_body"]["stream"] is True
|
|
|
|
|
|
def test_cache_hit_never_replays_a_foreign_content_type() -> None:
|
|
"""A cache entry cannot hand a caller a wire format it did not ask for."""
|
|
body = {
|
|
"model": "claude-sonnet-4-20250514",
|
|
"max_tokens": 64,
|
|
"stream": False,
|
|
"messages": [{"role": "user", "content": "hi"}],
|
|
}
|
|
payload = json.dumps(
|
|
{
|
|
"id": "msg_1",
|
|
"type": "message",
|
|
"role": "assistant",
|
|
"model": "claude-sonnet-4-20250514",
|
|
"content": [{"type": "text", "text": "cached"}],
|
|
"stop_reason": "end_turn",
|
|
"usage": {"input_tokens": 1, "output_tokens": 1},
|
|
}
|
|
).encode()
|
|
|
|
async def fail_retry(*args, **kwargs): # noqa: ANN001, ANN002, ANN003
|
|
raise AssertionError("upstream must not be called on a cache hit")
|
|
|
|
app = create_app(_config())
|
|
with TestClient(app) as client:
|
|
proxy = client.app.state.proxy
|
|
proxy._retry_request = fail_retry
|
|
key = proxy.cache._compute_key(
|
|
body["messages"],
|
|
body["model"],
|
|
system=None,
|
|
tools=None,
|
|
tool_choice=None,
|
|
temperature=None,
|
|
top_p=None,
|
|
top_k=None,
|
|
max_tokens=64,
|
|
stop=None,
|
|
thinking=None,
|
|
output_config=None,
|
|
)
|
|
proxy.cache._cache[key] = CacheEntry(
|
|
response_body=payload,
|
|
response_headers={"content-type": "text/event-stream"},
|
|
created_at=datetime.now(),
|
|
ttl_seconds=3600,
|
|
)
|
|
|
|
resp = client.post("/v1/messages", json=body, headers=_headers())
|
|
|
|
assert resp.status_code == 200, resp.text
|
|
assert resp.headers["content-type"].startswith("application/json")
|
|
assert resp.json()["content"][0]["text"] == "cached"
|