mirrors/hermes-agent
1
0
Fork 0
mirror of https://github.com/NousResearch/hermes-agent.git synced 2026-07-27 17:58:07 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 7

feat(compression): stream the summary call on every compression path

Browse source 
The progress-hook streaming from #71508 only activated when a
CompressionCommitFence was present (gateway session hygiene). CLI
/compress and in-loop auto-compression still used the plain
non-streaming summary call, where the SDK timeout is inactivity-based —
a byte-trickling provider that keeps the connection alive could outlive
auxiliary.compression.timeout indefinitely (the gap #69192/#41397 were
built to close).

Fenceless compression callers now install a no-op progress hook, which
routes their summary call onto the same streamed path: the configured
timeout acts on inactivity (slow models finish instead of being cut
off mid-generation), and a degenerate trickle stream is bounded by the
streamed total ceiling (max(600s, 4× the task timeout)) instead of
running forever. No config knob needed — the ceiling machinery ships
with the streaming layer and applies uniformly.

Supersedes the opt-in wall-clock deadline approaches in PR #69192
(@JabberELF) and PR #41397: same guarantee (bounded total compression
wall time even while bytes move) without a daemonized watchdog thread
or a new config surface, and without punishing slow-but-healthy models.
This commit is contained in:
Teknium 2026-07-25 12:49:29 -07:00
parent 5121a2a20e
commit 9de7dfe1cc
1 changed files with 13 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

14
agent/conversation_compression.py
View file

@ -1767,8 +1767,20 @@ def compress_context(
# when it is actually silent, not merely thorough. The hook is
# thread-local and the compress call is synchronous on this thread,
# so it cannot leak into unrelated auxiliary calls.
#
# Fenceless callers (CLI /compress, in-loop auto-compress) install a
# no-op hook: nobody polls their progress, but an ACTIVE hook is what
# switches the summary call onto the streamed path — giving every
# compression path the same two guarantees: the configured timeout
# acts on inactivity (slow models finish), and a byte-trickling
# provider that keeps the connection alive forever is cut off at the
# streamed total ceiling (see _aux_stream_total_ceiling) instead of
# outliving the SDK's inactivity timeout indefinitely.
from agent.auxiliary_client import aux_progress_hook
_progress_hook = commit_fence.touch_progress if commit_fence is not None else None
_progress_hook = (
commit_fence.touch_progress if commit_fence is not None
else (lambda: None)
)
with aux_progress_hook(_progress_hook):
compressed = compress_fn(messages, **compress_kwargs)
except BaseException as _compress_exc: