fix(cli): open the session store through the state.db registry, not a bare SessionDB()

The classic CLI froze for ~0.7-2s between the banner and the first prompt. py-spy +
strace on real PTY startups showed the main/REPL threads inside
refuse_deleted_wal_generation -> _iter_proc_fd_targets: a second full SessionDB open.

_init_session_store built a bare SessionDB(); a moment later the goal/loop/heartbeat
managers acquired the same state.db through hermes_state_registry from the REPL thread,
which is a different handle, so the whole open ran again — including the /proc-wide
deleted-WAL sidecar scan (~4.4k readlinks). Each readlink drops and re-takes the GIL
while the startup threads (plugin discovery, MCP, skill sync, banner git) are busy, so an
11ms scan stretched to 1.3s per pass, and the second pass landed exactly where the
prompt should have appeared.

Route the CLI's handle (init + the two re-open sites) through the registry so every
in-process consumer shares one writer. One scan per startup; live A/B on the same box,
interleaved x6: banner->prompt gap 0.37s mean -> 0.15s mean (plain), 1.77s -> 0.39s
under strace. The registry release path replaces close(), so /quit, /snapshot restore
and /handoff keep their semantics.
This commit is contained in:
Teknium
2026-09-12 00:13:17 -07:00
parent e8016a18c1
commit 876e444e4e
4 changed files with 47 additions and 6 deletions

8
cli.py
View File

@@ -2835,8 +2835,12 @@ class HermesCLI(CLIProcessNotificationsMixin, CLIAgentSetupMixin, CLICommandsMix
self._session_db = None
self._session_db_unavailable = False
try:
from hermes_state import SessionDB
self._session_db = SessionDB()
# Registry handle, not a bare SessionDB(): goals/loops/heartbeat acquire the same
# path a moment later from the REPL thread, and a second writer repeats the full
# open (the /proc-wide deleted-WAL scan, ~4k readlinks) while the render thread
# holds the GIL — that repeat was the post-banner freeze before the first prompt.
from hermes_state_registry import acquire
self._session_db = acquire()
except Exception as e:
# Without a store the transcript is NOT persisted while the chat looks healthy,
# so surface it prominently rather than only logging.

View File

@@ -513,8 +513,8 @@ class CLIAgentSetupMixin:
logger=logger, single_query=getattr(self, "_single_query_mode", False))
if self._session_db is None:
try:
from hermes_state import SessionDB
self._session_db = SessionDB()
from hermes_state_registry import acquire
self._session_db = acquire()
except Exception as e:
logger.warning("SQLite session store not available — session will NOT be indexed: %s", e)
if (

View File

@@ -1176,8 +1176,8 @@ class CLICommandsMixin:
return _cp(" Agent is busy. Wait for the current turn to finish, then retry /handoff.")
if not self._session_db:
with suppress(Exception):
from hermes_state import SessionDB
self._session_db = SessionDB()
from hermes_state_registry import acquire
self._session_db = acquire()
if not self._session_db:
return _cp(_db_unavailable_line())
# Ensure the session row exists (an empty session has flushed nothing yet): the gateway

View File

@@ -0,0 +1,37 @@
"""The CLI's session store must be the registry's shared handle for state.db.
``_init_session_store`` used to construct a bare ``SessionDB()``. The goal/loop/heartbeat
managers acquire the same path through ``hermes_state_registry`` from the REPL thread a
moment later, so a bare handle meant a SECOND full open — including the /proc-wide
deleted-WAL sidecar scan (~4k readlinks, each a GIL round-trip against the busy startup
threads) — which showed up as the post-banner freeze before the first prompt.
"""
from types import SimpleNamespace
import hermes_cli.goals as goals
from cli import HermesCLI
def test_cli_session_store_is_the_registry_handle_goals_reuse(monkeypatch):
import hermes_state_registry
monkeypatch.setattr(goals, "_DB_CACHE", {})
constructed = []
real_open = hermes_state_registry._open_session_db
def recording_open(path):
constructed.append(path)
return real_open(path)
monkeypatch.setattr(hermes_state_registry, "_open_session_db", recording_open)
cli = SimpleNamespace()
try:
HermesCLI._init_session_store(cli)
assert cli._session_db is not None and not cli._session_db_unavailable
# goals/loops/heartbeat go through the registry: same object, no second open.
assert goals._get_session_db() is cli._session_db
assert len(constructed) == 1
finally:
hermes_state_registry.close_all()