mirror of
https://github.com/bytedance/deer-flow.git
synced 2026-11-08 03:30:34 +00:00
3f1438d4a0
* fix(channels): publish Discord and WeChat JSON state atomically * fix(channels): adopt the shared atomic publisher for both state writers * test(channels): pin the cleanup-failure direction too * test(channels): keep a platform-independent guard on the publish call shape * style(tests): apply ruff format to the channel encoding test
149 lines
5.2 KiB
Python
149 lines
5.2 KiB
Python
"""Channel JSON state must be published whole, never truncated in place.
|
|
|
|
``DiscordChannel._persist_thread_mappings`` and ``WechatChannel._save_state``
|
|
write state a crash could otherwise truncate: the Discord mapping would fail to
|
|
load and the WeChat cursor buffer would be lost, re-reading from the start. Both
|
|
now stage a sibling and replace it, and a stage that cannot be discarded is
|
|
reported instead of replacing the outcome.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import json
|
|
import threading
|
|
from pathlib import Path
|
|
|
|
from app.channels.discord import DiscordChannel
|
|
from app.channels.wechat import WechatChannel
|
|
|
|
|
|
def _discord(tmp_path):
|
|
channel = DiscordChannel.__new__(DiscordChannel)
|
|
channel._thread_store_lock = threading.Lock()
|
|
channel._active_threads = {"a": {"thread_id": "t1"}}
|
|
channel._thread_store_path = tmp_path / "state" / "discord-threads.json"
|
|
return channel
|
|
|
|
|
|
def _wechat(tmp_path):
|
|
channel = WechatChannel.__new__(WechatChannel)
|
|
channel._cursor_path = tmp_path / "state" / "wechat-cursor.json"
|
|
channel._get_updates_buf = "cursor-1"
|
|
return channel
|
|
|
|
|
|
def _fail_staged_write(monkeypatch):
|
|
"""Fail the staged write the way a full disk would, before any rename.
|
|
|
|
With the shared helper this is the earliest injectable point: the bytes are
|
|
written to a sibling, fsync fails, and the rename never happens. The point of
|
|
the test is that the live file is untouched by that failure.
|
|
"""
|
|
|
|
def failing_fsync(_fd):
|
|
raise OSError("disk full")
|
|
|
|
monkeypatch.setattr("deerflow.utils.atomic_write.os.fsync", failing_fsync)
|
|
|
|
|
|
def _no_sidecar(target: Path) -> bool:
|
|
prefix = f".{target.name}."
|
|
return [p.name for p in target.parent.iterdir() if p.name.startswith(prefix)] == []
|
|
|
|
|
|
def test_discord_state_is_replaced_whole(tmp_path):
|
|
channel = _discord(tmp_path)
|
|
target = tmp_path / "state" / "discord-threads.json"
|
|
|
|
channel._persist_thread_mappings()
|
|
|
|
assert json.loads(target.read_text(encoding="utf-8")) == {"a": {"thread_id": "t1"}}
|
|
assert _no_sidecar(target)
|
|
|
|
|
|
def test_discord_failed_staged_write_keeps_the_previous_mapping(tmp_path, monkeypatch):
|
|
channel = _discord(tmp_path)
|
|
target = tmp_path / "state" / "discord-threads.json"
|
|
target.parent.mkdir(parents=True)
|
|
target.write_text('{"kept": true}', encoding="utf-8")
|
|
_fail_staged_write(monkeypatch)
|
|
|
|
channel._persist_thread_mappings()
|
|
|
|
assert target.read_text(encoding="utf-8") == '{"kept": true}'
|
|
assert json.loads(target.read_text(encoding="utf-8")) == {"kept": True}
|
|
assert _no_sidecar(target)
|
|
|
|
|
|
def test_wechat_state_is_replaced_whole(tmp_path):
|
|
channel = _wechat(tmp_path)
|
|
target = tmp_path / "state" / "wechat-cursor.json"
|
|
|
|
channel._save_state()
|
|
|
|
assert json.loads(target.read_text(encoding="utf-8")) == {"get_updates_buf": "cursor-1"}
|
|
assert _no_sidecar(target)
|
|
|
|
|
|
def test_wechat_failed_staged_write_keeps_the_previous_cursor(tmp_path, monkeypatch):
|
|
channel = _wechat(tmp_path)
|
|
target = tmp_path / "state" / "wechat-cursor.json"
|
|
target.parent.mkdir(parents=True)
|
|
target.write_text('{"get_updates_buf": "kept"}', encoding="utf-8")
|
|
_fail_staged_write(monkeypatch)
|
|
|
|
channel._save_state()
|
|
|
|
assert json.loads(target.read_text(encoding="utf-8")) == {"get_updates_buf": "kept"}
|
|
assert _no_sidecar(target)
|
|
|
|
|
|
def _fail_staged_write_and_cleanup(monkeypatch):
|
|
"""Fail the staged write, and fail discarding the staged sibling too."""
|
|
|
|
def failing_fsync(_fd):
|
|
raise OSError("disk full")
|
|
|
|
real_unlink = Path.unlink
|
|
|
|
def failing_unlink(self, *args, **kwargs):
|
|
if self.name.startswith(".") and self.name.endswith(".tmp"):
|
|
raise PermissionError("staging is still open")
|
|
return real_unlink(self, *args, **kwargs)
|
|
|
|
monkeypatch.setattr("deerflow.utils.atomic_write.os.fsync", failing_fsync)
|
|
monkeypatch.setattr(Path, "unlink", failing_unlink)
|
|
|
|
|
|
def test_discord_failed_cleanup_still_reports_and_keeps_the_outcome(tmp_path, monkeypatch, caplog):
|
|
"""A discard that fails must not escape, and both failures stay visible."""
|
|
channel = _discord(tmp_path)
|
|
target = tmp_path / "state" / "discord-threads.json"
|
|
target.parent.mkdir(parents=True)
|
|
target.write_text('{"kept": true}', encoding="utf-8")
|
|
_fail_staged_write_and_cleanup(monkeypatch)
|
|
caplog.clear()
|
|
|
|
with caplog.at_level("WARNING"):
|
|
channel._persist_thread_mappings()
|
|
|
|
assert target.read_text(encoding="utf-8") == '{"kept": true}'
|
|
assert "failed to persist thread mappings" in caplog.text
|
|
assert "Failed to remove temporary output file" in caplog.text
|
|
|
|
|
|
def test_wechat_failed_cleanup_still_reports_and_keeps_the_outcome(tmp_path, monkeypatch, caplog):
|
|
channel = _wechat(tmp_path)
|
|
target = tmp_path / "state" / "wechat-cursor.json"
|
|
target.parent.mkdir(parents=True)
|
|
target.write_text('{"get_updates_buf": "kept"}', encoding="utf-8")
|
|
_fail_staged_write_and_cleanup(monkeypatch)
|
|
caplog.clear()
|
|
|
|
with caplog.at_level("WARNING"):
|
|
channel._save_state()
|
|
|
|
assert target.read_text(encoding="utf-8") == '{"get_updates_buf": "kept"}'
|
|
assert "failed to persist cursor state" in caplog.text
|
|
assert "Failed to remove temporary output file" in caplog.text
|