David/HERMES-GIELIE
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

docs(#4908): correct cache-invalidation comment per Opus gate — TTL is the backstop for externally-driven changes

Browse source 
Opus SHIP-WITH-FIXES (no code change required): the comment/docstring overstated
that the session-list-change listener covers ALL structural mutations. In-app
mutations fire it, but externally-driven changes (scheduled cron completion,
external CLI writes) don't — for those the 30s streaming TTL is the load-bearing
backstop (≤30s bounded, self-healing lag). Comment-only accuracy fix; logic
unchanged (both gates: perf fix correct, freeze effective, staleness bounded).
This commit is contained in:
nesquena-hermes 2026-06-25 21:51:07 +00:00 committed by nesquena-hermes
parent b87a605306
commit 7791370d49
2 changed files with 15 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
api/models.py
View file

@ -4256,8 +4256,11 @@ def _cli_sessions_streaming_freeze_marker():
a stream starts/stops (so the just-finished turn's rows are picked up promptly).
A streaming session's own CLI/cron title/count is not what this projection
returns (the streaming session is overlaid live by the route layer), and any
structural mutation invalidates the cache directly via
``clear_cli_sessions_cache``, so nothing user-visible lags under the freeze. (#4842)
in-app structural mutation invalidates the cache directly via
``clear_cli_sessions_cache``. Externally-driven changes that don't fire that
listener (a scheduled cron completing, an external CLI writing rows) surface
within one streaming-TTL window (≤30s) rather than instantly — a bounded,
self-healing lag that is the deliberate latency/CPU trade-off of the freeze. (#4842)
"""
try:
active = _active_stream_ids()

15
api/routes.py
View file

@ -208,11 +208,16 @@ def _on_session_list_changed(profile: str | None = None) -> None:
# #4842: also drop the inner CLI/cron projection cache. While a turn streams
# that cache is frozen on a stable streaming marker (so per-token message
# writes don't bust it), which means it no longer self-invalidates via the
# state.db content fingerprint mid-stream. Structural mutations (cron
# completion, new/renamed/archived sessions, attention) DO fire this
# listener — and never fire per streamed token — so clearing here restores
# prompt freshness for real changes without reintroducing the per-poll
# rebuild the freeze removed.
# state.db content fingerprint mid-stream. In-app structural mutations
# (session create/rename/archive/delete/branch/pin/move/import, attention)
# fire this listener — and never fire per streamed token — so clearing here
# restores prompt freshness for those without reintroducing the per-poll
# rebuild the freeze removed. Note: externally-driven changes that do NOT go
# through this listener (a scheduled cron job completing, or an external CLI
# writing rows directly) are not cleared here mid-stream; for those the 30s
# streaming TTL is the backstop — they surface within one streaming-TTL
# window (≤30s) rather than instantly. That bound is the deliberate
# latency/CPU trade-off of the freeze.
try:
from api.models import clear_cli_sessions_cache
clear_cli_sessions_cache()