v2.3.1 — P0: 消费超时 + 背压 + 信号处理 + 离线降级

Performance Release v2.3.1

Highlights

• 2636fd4 — CronEngine ThreadPool + Audit Batched Async + LLM Cache/Concurrent

P0-3: CronEngine ThreadPool

• ThreadPoolExecutor(8 workers) + run_jobs_parallel() concurrent execution
• 2x speedup for parallel cron jobs (2 jobs: 0.2s → 0.1s)
• Per-job _running flag prevents duplicate concurrent execution
• Auto-register handler in register_job()

P1-4: Audit Batched Async Writes

• Module-level _batch_buffer (deque maxlen=1000)
• Background flush daemon: 50 entries OR 2s interval
• executemany() batch INSERT (1 DB tx vs N writes)
• start_batch_daemon() / stop_batch_daemon() lifecycle

P1-5: LLM Cache + Concurrent

• cloud/brain/llm_cache.py — LRU+TTL cache (200 entries, 3600s TTL)
  • SHA256 key of (system, user, model, temperature)
  • Optional disk persistence to /opt/clawshell/data/llm_cache.json
  • Stats: hits/misses/evictions/hit_rate
• cloud/brain/llm_concurrent.py — ThreadPoolExecutor(4 workers)
  • chat_batch() parallel queries with Future resolution
  • chat() single with cache lookup
  • Avoids redundant API calls for repeated queries

P2-6: TF-IDF (verified no change needed)

• Already on-demand compute at search time (not full rebuild on insert)
• O(n) per query with O(1) term_index lookup

Previous v2.3.1 P0 Fixes

• 173b29f — P0: claim timeout + backpressure + signal handler + offline mode

Verification (ECS Production)

• Engines: 20/20 active
• Tests: 620/621 pass (1 pre-existing task expiry test, unrelated)
• MemOS Key: validated (HTTP 200 from /search/memory endpoint)

Required API Keys

• MEMOS_CLOUD_API_KEY (mpg-*) — https://memos.cloud
• DEEPSEEK_API_KEY (sk-*) — https://platform.deepseek.com/api_keys

Total Diff

4 files changed, 360 insertions(+), 8 deletions(-)