perf(agent,gateway): back cancel-wait polls off from 1ms to 25ms (re-review #5)

The fence-cancel poll loops (sync host wait in conversation_compression,
async hygiene wait in gateway/run) spun at 1kHz while the worker held
the fence through its lock-setup window — which rides SessionDB write
patience and can last seconds. 25ms keeps sub-tick cancel latency
without the spin.
This commit is contained in:
Teknium
2026-08-02 15:33:17 -07:00
parent 0628e33347
commit 06bdc48c1f
2 changed files with 12 additions and 2 deletions
+6 -1
View File
@@ -972,7 +972,12 @@ def run_compress_context_with_progress_timeout(
break
cancelled = fence.try_cancel_before_commit()
if cancelled is None:
time.sleep(0.001)
# Round-2 #5: the fence is only held transiently here (lock
# setup / cancel admission — an in-flight commit is caught by
# the commit_in_flight check above), but that window rides
# SessionDB write patience and can last seconds. 25ms keeps
# sub-tick latency without a 1kHz spin.
time.sleep(0.025)
if not cancelled:
# Pre-commit ceiling already elapsed, but begin_commit() won the
# race. Waiting is intentional: SessionDB mutation cannot be
+6 -1
View File
@@ -16747,7 +16747,12 @@ class GatewayRunner(GatewayAuthorizationMixin, GatewayKanbanWatchersMixin, Gatew
_hyg_commit_fence.try_cancel_before_commit()
)
if _cancelled is None:
await asyncio.sleep(0.001)
# Round-2 #5: transient
# lock-setup windows ride
# write patience for seconds;
# 25ms keeps sub-tick latency
# without 1kHz spin.
await asyncio.sleep(0.025)
if not _cancelled:
# The worker crossed the commit boundary just
# before the timeout. The fence poll waited for