simon
c1ab751d95
- domestic_full.sh / pipeline.sh 新增 --no-report 参数：跳过日报生成， 保持 M1 抓取 + M2→M6 管道不变，用于 12:00/18:00 只抓不生成日报 - crontab: 07:00 全量含日报，12:00/18:00 --no-report（日报每日一次，降低 LLM 费用） - docs/llm-models.md: 各 LLM 调用点分别用的 provider/model 事实清单 - docs/llm-cost-analysis.md: 真实 token 消耗与费用构成、降本建议 - docs/llm-cache-optimization.md: DeepSeek 上下文缓存机制与命中率提升设计 - docs/README.md 增加上述三个文档索引
132 lines
4.9 KiB
Bash
Executable File
132 lines
4.9 KiB
Bash
Executable File
#!/bin/bash
|
||
# =============================================
|
||
# 国内服务器：全链路管道 M2 → M3 → M4 → M5 → M6 → 日报
|
||
# =============================================
|
||
# 用法：
|
||
# ./scripts/pipeline.sh # 全新执行（步骤级不跳过）
|
||
# ./scripts/pipeline.sh --resume # 从中断处继续（跳过已完成步骤）
|
||
# ./scripts/pipeline.sh --no-report # 只跑 M2→M6，跳过日报生成
|
||
# ./scripts/pipeline.sh --no-report --resume
|
||
# 前提：domestic_sync.sh 已完成
|
||
# =============================================
|
||
# 各步骤"跳过已处理文件"（文件级增量，由各 Python 模块自带）：
|
||
# M2 正文提取: data/processed/{source}/{date}/{url_hash}.json 存在则跳过
|
||
# M3 去重: 指纹库 SQLite（data/dedup_fingerprints.sqlite3）判重，重复自动丢弃
|
||
# M4 翻译: data/events/{date}/{url_hash}.json 存在则跳过
|
||
# M5 向量: data/embeddings/{date}/index.json 记录已向量化，跳过
|
||
# M6 入库: Qdrant upsert 幂等（点 id = url_hash，重复写入覆盖，无重复）
|
||
# 日报: MySQL 唯一键 (report_date, intl, "") 幂等覆盖
|
||
# =============================================
|
||
set -euo pipefail
|
||
|
||
SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
|
||
PROJECT_DIR="$(dirname "$SCRIPT_DIR")"
|
||
|
||
cd "$PROJECT_DIR"
|
||
|
||
# ── 参数解析 ──
|
||
RESUME=0
|
||
NO_REPORT=0
|
||
for arg in "$@"; do
|
||
case "$arg" in
|
||
--resume) RESUME=1 ;;
|
||
--no-report) NO_REPORT=1 ;;
|
||
*) echo "未知参数: ${arg}（支持 --resume / --no-report）" >&2; exit 1 ;;
|
||
esac
|
||
done
|
||
|
||
source "$SCRIPT_DIR/_step_state.sh"
|
||
|
||
LOG() { echo "[$(date '+%Y-%m-%d %H:%M:%S')] $*"; }
|
||
|
||
# ── 步骤日志：全部输出实时显示到终端，同时完整写入日志文件 ──
|
||
LOG_DIR="logs"
|
||
mkdir -p "$LOG_DIR"
|
||
STEP_LOG="$LOG_DIR/pipeline_$(date +%Y%m%d_%H%M%S).log"
|
||
: > "$STEP_LOG"
|
||
LOG "══════ 全链路管道开始（resume=${RESUME} no_report=${NO_REPORT}）═══════"
|
||
LOG "步骤日志: ${STEP_LOG}（终端实时显示完整进度）"
|
||
|
||
# ── AI 模型信息读取（供阶段 banner 显性展示供应商/模型）──
|
||
ai_model_info() {
|
||
# $1: 场景名（translation / daily_report）或 "embedding"
|
||
# 输出格式: "供应商 / 模型"
|
||
.venv/bin/python3 -c "
|
||
import sys, yaml
|
||
scene = sys.argv[1]
|
||
raw = yaml.safe_load(open('configs/system.yaml', encoding='utf-8'))
|
||
if scene == 'embedding':
|
||
cfg = raw.get('embedding', {})
|
||
p = cfg.get('provider', '?')
|
||
m = cfg.get('dashscope_model', '?')
|
||
else:
|
||
base = raw.get('llm', {})
|
||
sc = (raw.get('llm_scenes', {}) or {}).get(scene, {})
|
||
merged = {**base, **sc}
|
||
p = merged.get('provider', '?')
|
||
m = merged.get('model') or merged.get(p + '_model', '?')
|
||
print(f'{p} / {m}')
|
||
" "$1"
|
||
}
|
||
|
||
# 加载 .env
|
||
if [ -f .env ]; then
|
||
set -a
|
||
. ./.env
|
||
set +a
|
||
fi
|
||
|
||
# ── M2: 正文提取 ──
|
||
LOG "━━━ M2 正文提取 ━━━"
|
||
step_run M2_extract .venv/bin/python3 -c "
|
||
from extractor.pipeline import process_all_sources
|
||
stats = process_all_sources()
|
||
print(f'M2: {stats[\"total_articles\"]} 篇, {stats[\"elapsed_sec\"]:.0f}s')
|
||
"
|
||
|
||
# ── M3: 去重 ──
|
||
LOG "━━━ M3 三层去重 ━━━"
|
||
step_run M3_dedup .venv/bin/python3 -c "
|
||
from dedup.pipeline import dedup_all_sources
|
||
stats = dedup_all_sources()
|
||
print(f'M3: 唯一 {stats[\"unique\"]}/重复 {stats[\"duplicate\"]}, {stats[\"elapsed_sec\"]:.0f}s')
|
||
"
|
||
|
||
# ── M4: 翻译+事件（AI 大模型）──
|
||
LOG "━━━ M4 翻译+事件抽取（AI 大模型: $(ai_model_info translation)）━━━"
|
||
step_run M4_translate .venv/bin/python3 -c "
|
||
from llm.pipeline import translate_all_deduped
|
||
stats = translate_all_deduped()
|
||
print(f'M4: {stats[\"success\"]}/{stats[\"total\"]} 篇, {stats[\"elapsed_sec\"]:.0f}s')
|
||
"
|
||
|
||
# ── M5: 向量生成（AI 大模型）──
|
||
LOG "━━━ M5 向量生成（AI 大模型: $(ai_model_info embedding)）━━━"
|
||
step_run M5_embed .venv/bin/python3 -c "
|
||
from embedding.pipeline import embed_all_events
|
||
stats = embed_all_events()
|
||
print(f'M5: {stats[\"success\"]}/{stats[\"total\"]} 篇, {stats[\"elapsed_sec\"]:.0f}s')
|
||
"
|
||
|
||
# ── M6: Qdrant 入库 ──
|
||
LOG "━━━ M6 Qdrant 入库 ━━━"
|
||
step_run M6_index .venv/bin/python3 -c "
|
||
from vectorstore.pipeline import ingest_all_embeddings
|
||
stats = ingest_all_embeddings()
|
||
print(f'M6: {stats[\"ingested\"]}/{stats[\"total\"]} 条, {stats[\"elapsed_sec\"]:.0f}s')
|
||
"
|
||
|
||
# ── 日报（AI 大模型）──
|
||
if [ "$NO_REPORT" = "1" ]; then
|
||
LOG "━━━ 日报生成（已跳过，--no-report）━━━"
|
||
else
|
||
LOG "━━━ 日报生成（AI 大模型: $(ai_model_info daily_report)）━━━"
|
||
step_run report .venv/bin/python3 -c "
|
||
from scheduler.reporter import generate_report
|
||
report_id = generate_report()
|
||
print(f'日报: report_id={report_id}' if report_id is not None else '日报: 无数据/失败')
|
||
"
|
||
fi
|
||
|
||
LOG "══════ 全链路管道完成 ✅ ═══════"
|