Compress tool outputs, logs, files, and RAG chunks before they reach the LLM. 20% fewer tokens for coding agents, 60-95% fewer tokens for JSON, same answers. Library, proxy, MCP server.
Python agentaianthropicclaude-codecompressioncontext-engineeringcontext-windowcursorfastapilangchainllmmcpopenaiprompt-engineeringproxypythonragtoken-optimizationtokenstypescript
fix(proxy): stop get_recent deep-copying dropped message payloads (#3613)
gglucass44b8e7ba877a
diff --git a/headroom/proxy/request_logger.py b/headroom/proxy/request_logger.pyindex 2f5ad7cda..89a6fe9b6 100644--- a/headroom/proxy/request_logger.py+++ b/headroom/proxy/request_logger.py@@ -25,7 +25,8 @@ import jsonimport loggingimport sysfrom collections import deque-from dataclasses import asdict+from copy import deepcopy+from dataclasses import asdict, fieldsfrom pathlib import Pathfrom threading import Lockfrom typing import TYPE_CHECKING, Any@@ -82,6 +83,10 @@ def redact_image_base64(payload: Any) -> Any:return result.value+# Payload fields `get_recent` never returns and therefore must never walk.+_HEAVY_FIELDS = frozenset({"request_messages", "compressed_messages", "response_content"})++class RequestLogger:"""Log requests to JSONL file.@@ -147,12 +152,16 @@ class RequestLogger:"""Get recent log entries (without request/compressed messages and response_content)."""# Convert deque to list for slicing (deque doesn't support slicing)entries = list(self._logs)[-n:]+ # Not asdict(): that deep-copies every field BEFORE the heavy ones are+ # dropped, so each entry cost a full walk of its request_messages and+ # compressed_messages (~3.4 ms for a 400 KB Claude Code transcript).+ # /stats calls this with n=10_000 synchronously on the event loop, so+ # a full deque made every /stats build take ~30 s and a dashboard+ # polling it starved /v1/messages. The retained fields are still+ # deep-copied, so callers cannot reach nested containers (tags,+ # savings_breakdown) that the in-memory log owns.return [- {- k: v- for k, v in asdict(e).items()- if k not in ("request_messages", "compressed_messages", "response_content")- }+ {f.name: deepcopy(getattr(e, f.name)) for f in fields(e) if f.name not in _HEAVY_FIELDS}for e in entries]diff --git a/tests/test_proxy/test_request_logger.py b/tests/test_proxy/test_request_logger.pyindex b9d140baf..a2ca71218 100644--- a/tests/test_proxy/test_request_logger.py+++ b/tests/test_proxy/test_request_logger.py@@ -98,3 +98,37 @@ def test_get_memory_stats_accounts_for_compressed_messages():stats = logger.get_memory_stats()assert stats.entry_count == 1assert stats.size_bytes > 0+++def test_get_recent_never_walks_message_payloads():+ """`/stats` calls get_recent(10_000) on the event loop; the payloads it+ drops must not be traversed first (asdict deep-copied them, ~30 s per call+ on a full deque). A leaf that refuses to be deep-copied proves the walk+ is gone."""++ class _NoCopy:+ def __deepcopy__(self, memo):+ raise AssertionError("get_recent walked a message payload")++ logger = RequestLogger(log_file=None, log_full_messages=True)+ logger.log(+ _entry(+ request_messages=[{"role": "user", "content": _NoCopy()}],+ compressed_messages=[{"role": "user", "content": _NoCopy()}],+ tags={"agent": "codex", "meta": {"depth": 1}},+ transforms_applied=["smart_crusher"],+ savings_breakdown=[{"tokens": 60}],+ )+ )++ recent = logger.get_recent(10)+ assert recent[0]["tags"] == {"agent": "codex", "meta": {"depth": 1}}+ assert recent[0]["transforms_applied"] == ["smart_crusher"]+ # Still copies, not aliases, of the entry's own containers, all the way+ # down: mutating a nested value must not change the next /stats result.+ recent[0]["tags"]["agent"] = "x"+ recent[0]["tags"]["meta"]["depth"] = 99+ recent[0]["savings_breakdown"][0]["tokens"] = 0+ again = logger.get_recent(10)[0]+ assert again["tags"] == {"agent": "codex", "meta": {"depth": 1}}+ assert again["savings_breakdown"] == [{"tokens": 60}]