turnstone/app/services/diagnose/pipeline.py

"""Multi-agent diagnose pipeline orchestrator — Stage 1–5 wiring."""

from __future__ import annotations

import asyncio
import dataclasses
import logging
import os
from collections.abc import AsyncGenerator
from pathlib import Path
from typing import Any

# Optional ML classifier model for Stage 2.
# When empty (default), Stage 2 falls back to pattern_tags then regex.
# Set TURNSTONE_CLASSIFIER_MODEL to a HuggingFace model ID to enable ML classification.
# Recommended: byviz/bylastic_classification_logs (DistilBERT, ~300MB)
_CLASSIFIER_MODEL: str = os.environ.get("TURNSTONE_CLASSIFIER_MODEL", "")

from app.context.retriever import RetrievedContext
from app.services.diagnose.classifier import SeverityClassifier
from app.services.diagnose.hypothesizer import RootCauseHypothesizer
from app.services.diagnose.suppressor import FalsePositiveSuppressor
from app.services.diagnose.synthesizer import SummarySynthesizer
from app.services.diagnose.timeline import TimelineReconstructor
from app.services.search import SearchResult

logger = logging.getLogger(__name__)


async def run_pipeline(
    db_path: Path,
    entries: list[SearchResult],
    ctx: RetrievedContext,
    query: str,
    since: str | None,   # reserved for future range-filtering in stage queries (#29 follow-up)
    until: str | None,   # reserved for future range-filtering in stage queries (#29 follow-up)
    llm_url: str | None,
    llm_model: str | None,
    llm_api_key: str | None,
    tech_level: str = "sysadmin",
    incidents_db_path: Path | None = None,
) -> AsyncGenerator[dict[str, Any], None]:
    """Async generator that runs all 5 pipeline stages and yields SSE event dicts.

    Stages:
      1. TimelineReconstructor  — cluster log entries by time
      2. SeverityClassifier     — annotate clusters with severity
      3. RootCauseHypothesizer  — generate hypotheses via LLM
      4. FalsePositiveSuppressor — rank and suppress known patterns
      5. SummarySynthesizer     — produce a narrative diagnosis

    Yields events in order:
      {"type": "status", "message": "Building timeline…"}
      {"type": "pipeline_stage", "stage": 1, ...}
      {"type": "pipeline_stage", "stage": 2, ...}
      {"type": "pipeline_stage", "stage": 3, ...}
      {"type": "pipeline_stage", "stage": 4, ...}
      {"type": "hypotheses", "data": [...]}
      {"type": "status", "message": "Synthesizing…"}
      {"type": "reasoning", "text": "..."}   — only when synthesis produces text
      {"type": "done"}
    """
    # Stage 1: Timeline reconstruction
    yield {"type": "status", "message": "Building timeline…"}
    try:
        timeline = await asyncio.to_thread(
            TimelineReconstructor().reconstruct, entries
        )
    except Exception as exc:
        logger.exception("Stage 1 (timeline) failed: %s", exc)
        yield {"type": "error", "message": "Pipeline error in stage 1 (timeline)"}
        yield {"type": "done"}
        return
    n_clusters = len(timeline.clusters)
    burst = timeline.burst_count
    yield {
        "type": "pipeline_stage",
        "stage": 1,
        "name": "timeline",
        "message": f"Built {n_clusters} clusters, {burst} bursts",
    }

    # Stage 2: Severity classification
    try:
        classified = await asyncio.to_thread(
            SeverityClassifier(model_id=_CLASSIFIER_MODEL).classify, timeline
        )
    except Exception as exc:
        logger.exception("Stage 2 (classifier) failed: %s", exc)
        yield {"type": "error", "message": "Pipeline error in stage 2 (classifier)"}
        yield {"type": "done"}
        return
    sev_counts: dict[str, int] = {}
    for sev in classified.cluster_severities.values():
        sev_counts[sev] = sev_counts.get(sev, 0) + 1
    counts_str = ", ".join(f"{k}:{v}" for k, v in sorted(sev_counts.items()))
    yield {
        "type": "pipeline_stage",
        "stage": 2,
        "name": "classifier",
        "message": f"{classified.classifier_used} classifier: {counts_str}",
    }

    # Stage 3: Root-cause hypotheses
    try:
        hypotheses = await asyncio.to_thread(
            RootCauseHypothesizer().hypothesize,
            classified,
            ctx,
            query,
            llm_url,
            llm_model,
            llm_api_key,
        )
    except Exception as exc:
        logger.exception("Stage 3 (hypothesizer) failed: %s", exc)
        yield {"type": "error", "message": "Pipeline error in stage 3 (hypothesizer)"}
        yield {"type": "done"}
        return
    yield {
        "type": "pipeline_stage",
        "stage": 3,
        "name": "hypotheses",
        "message": f"{len(hypotheses)} hypotheses generated",
    }

    # Stage 4: False-positive suppression
    _incidents_db = incidents_db_path or db_path
    try:
        ranked = await asyncio.to_thread(
            FalsePositiveSuppressor().suppress, hypotheses, _incidents_db
        )
    except Exception as exc:
        logger.exception("Stage 4 (suppressor) failed: %s", exc)
        yield {"type": "error", "message": "Pipeline error in stage 4 (suppressor)"}
        yield {"type": "done"}
        return
    suppressed = sum(1 for rh in ranked if rh.suppress)
    active = len(ranked) - suppressed
    yield {
        "type": "pipeline_stage",
        "stage": 4,
        "name": "suppressor",
        "message": f"{suppressed} suppressed, {active} active",
    }
    yield {
        "type": "hypotheses",
        "data": [dataclasses.asdict(rh) for rh in ranked],
    }

    # Stage 5: Summary synthesis
    yield {"type": "status", "message": "Synthesizing…"}
    try:
        synthesis_text = await asyncio.to_thread(
            SummarySynthesizer().synthesize,
            ranked,
            timeline,
            ctx,
            query,
            llm_url,
            llm_model,
            llm_api_key,
            tech_level,
        )
    except Exception as exc:
        logger.exception("Stage 5 (synthesizer) failed: %s", exc)
        yield {"type": "error", "message": "Pipeline error in stage 5 (synthesizer)"}
        yield {"type": "done"}
        return
    if synthesis_text:
        yield {"type": "reasoning", "text": synthesis_text}

    yield {"type": "done"}