Fix run-journal writer-lock cache leak on session delete

`_lock_for` caches one threading.Lock per (dir, file, pid) in the module-global
`_WRITER_LOCKS`, but nothing evicted those entries: `delete_run_journal` rmtree'd
the on-disk `_run_journal/{sid}/` directory yet left the cached lock objects
behind, so a long-lived gateway leaked one entry per deleted run forever.

Delete now evicts every cached lock whose parent directory matches the removed
session (pid-independent) under `_WRITER_LOCKS_GUARD`, leaving unrelated
sessions' locks intact.

Refs #4633, #2097.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

api/run_journal.py

@@ -285,7 +285,22 @@ def delete_run_journal(session_id: str, *, session_dir: Path | None = None) -> b
     if not session_journal_dir.exists():
         return False
     shutil.rmtree(session_journal_dir, ignore_errors=True)
-    return not session_journal_dir.exists()
+    removed = not session_journal_dir.exists()
+    # Evict any writer locks the removed runs left behind. `_lock_for` keys are
+    # ``(str(path.parent), path.name, pid)`` and every run file for this session
+    # lives directly under ``session_journal_dir``, so drop all keys whose parent
+    # dir matches — pid-independent — to keep `_WRITER_LOCKS` from growing forever.
+    # Guard on confirmed removal: `rmtree(ignore_errors=True)` can silently leave
+    # the directory (locked files on Windows, permission transients). If the files
+    # still exist their locks are still live — evicting them would hand a later
+    # `_lock_for` caller a brand-new Lock, breaking mutual exclusion with a writer
+    # still holding the old one.
+    if removed:
+        dir_key = str(session_journal_dir)
+        with _WRITER_LOCKS_GUARD:
+            for key in [k for k in _WRITER_LOCKS if k[0] == dir_key]:
+                del _WRITER_LOCKS[key]
+    return removed
 
 
 def stale_interrupted_event(session_id: str, run_id: str, *, after_seq: int | None = None) -> dict | None:

tests/test_run_journal_writer_lock_evict.py

@@ -0,0 +1,49 @@
+"""Regression tests for #4633/#2097 — deleting a run journal must evict its writer lock.
+
+`_lock_for` lazily creates a `threading.Lock` per ``(dir, file, pid)`` key and
+caches it in the module-global `_WRITER_LOCKS`, but nothing ever removed those
+entries: `delete_run_journal` rmtree'd the on-disk directory yet left the lock
+objects behind, so a long-lived gateway leaked one entry per deleted run
+forever. These tests pin that delete now drops the matching cache entries while
+leaving unrelated sessions' locks intact.
+"""
+import api.run_journal as run_journal
+from api.run_journal import RunJournalWriter, delete_run_journal
+
+
+def _keys_for(session_dir, sid):
+    dir_key = str(session_dir / run_journal.RUN_JOURNAL_DIR_NAME / sid)
+    return [k for k in run_journal._WRITER_LOCKS if k[0] == dir_key]
+
+
+def test_delete_run_journal_evicts_writer_locks(tmp_path):
+    writer = RunJournalWriter("sid-del", "run-1", session_dir=tmp_path)
+    writer.append_sse_event("token", {"text": "hello"})
+    # Constructing the writer + appending populated the per-run lock cache.
+    assert _keys_for(tmp_path, "sid-del")
+
+    assert delete_run_journal("sid-del", session_dir=tmp_path) is True
+
+    # The cached lock(s) for the deleted session are gone — no unbounded climb.
+    assert _keys_for(tmp_path, "sid-del") == []
+
+
+def test_delete_run_journal_keeps_other_sessions_locks(tmp_path):
+    RunJournalWriter("sid-keep", "run-k", session_dir=tmp_path).append_sse_event("token", {"text": "k"})
+    RunJournalWriter("sid-del", "run-d", session_dir=tmp_path).append_sse_event("token", {"text": "d"})
+
+    delete_run_journal("sid-del", session_dir=tmp_path)
+
+    assert _keys_for(tmp_path, "sid-del") == []
+    assert _keys_for(tmp_path, "sid-keep"), "unrelated session's lock must survive"
+
+
+def test_delete_run_journal_noop_leaves_cache_untouched(tmp_path):
+    RunJournalWriter("sid-keep", "run-k", session_dir=tmp_path).append_sse_event("token", {"text": "k"})
+    before = _keys_for(tmp_path, "sid-keep")
+
+    # Missing/invalid ids never touch the cache.
+    assert delete_run_journal("nope", session_dir=tmp_path) is False
+    assert delete_run_journal("../escape", session_dir=tmp_path) is False
+
+    assert _keys_for(tmp_path, "sid-keep") == before