"""Context compression worker."""
import structlog
from navi.config import settings
from navi.core.compressor import ContextCompressor, should_compress
from .base import Worker, WorkerContext, WorkerResult
log = structlog.get_logger()
class CompressionWorker(Worker):
"""
Compresses session.context when it approaches the token limit.
session.messages (full display history) is never modified.
The token gate uses the real prompt-token count from the last LLM call
(``ctx.context_tokens``); the compression itself is delegated to
``ContextCompressor.compress_and_save_session`` so the post-turn path gets
the same pipeline as the pre/mid-turn paths: retry on LLM failure,
hard-truncate fallback, the result-None token-budget fallback, the
65%-target safety net, real-baseline clearing, and message-window
archiving. The worker used to call ``compress_context`` directly with its
own save logic — a second, weaker pipeline that silently no-op'ed whenever
the summarizer LLM failed while the context stayed over the threshold.
"""
async def run(self, session, ctx: WorkerContext) -> WorkerResult:
if not settings.context_compression_enabled:
return WorkerResult()
if ctx.context_tokens is None:
return WorkerResult()
if not should_compress(ctx.context_tokens, ctx.max_context_tokens,
settings.context_compression_threshold):
return WorkerResult()
compressor = ContextCompressor()
compressor.set_profile(ctx.profile)
try:
# Mirror the midturn path: a long autonomous turn (one user
# message + many tool iterations) is a single conversational
# turn, so turn-based compression alone finds nothing to
# summarize. The intra-turn fallback lets the worker compress
# it too — without this the post-turn worker always no-ops for
# the navi_code shape.
event = await compressor.compress_and_save_session(
session=session,
session_store=ctx.session_store,
llm=ctx.llm,
model=ctx.model,
temperature=settings.context_summary_temperature,
session_id=ctx.session_id,
reason="postturn",
keep_recent=settings.context_keep_recent,
max_tokens=settings.context_summary_max_tokens,
keep_recent_messages=max(12, settings.context_keep_recent * 2),
)
except Exception:
# compress_and_save_session already retries and hard-truncates
# internally; reaching here means something unexpected broke
# (store failure etc.) — log and leave the session untouched.
log.warning("compression_worker.failed", session_id=ctx.session_id, exc_info=True)
return WorkerResult()
if event is None:
return WorkerResult()
log.info(
"compression_worker.done",
session_id=ctx.session_id,
before=event.messages_before,
after=event.messages_after,
)
return WorkerResult(events=[event])