yumoqing/llmage
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
903 Commits 3 Branches 0 Tags
Commit Graph

4 Commits

Author SHA1 Message Date
yumoqing
31c8fde336 fix(balance): replace GETDEL with GET+DEL in Lua scripts for Redis 6.0 compat 
GETDEL requires Redis 6.2+; server runs 6.0.16, so every finalize_balance
and refund_balance call failed silently ('Unknown Redis command called
from Lua script'). Consequences observed under concurrent load test:
- reserve:* keys leaked until 600s TTL instead of being released at finalize
- balance:{org} never reconciled with actual cost (over-deduction by
  max_cost-actual accumulated per call)
- FAILED-request refunds silently lost

GET+DEL inside a Lua script executes atomically (single-threaded),
so the pop semantics are unchanged.
 2026-08-05 12:05:30 +08:00
yumoqing
e3f61f446d feat(balance): extend atomic balance reserve to async+sync+product paths 
- balance.py: module-level redis.asyncio singleton (works in all
  processes without env.redis injection), centralized tpac/self-org
  skip policy, userid param, extend_reserve, invalidate_balance_cache,
  finalize updates max_cost on cold start (no reserve)
- llmclient.py: unified reserve injection in _inference_generator
  (covers all 22 dspy entries); fix luid mismatch (reserve key ==
  llmusage id == finalize key); stream exception refund
- syncinference.py / asyncinference.py: luid reuse + failure refunds
  (submit failure, outer exception, query_task_status FAILED)
- product_interface.py: reserve + refund on product stream/non-stream
- accounting.py: finalize after llm_accounting (was missing for
  direct llmage accounting path -> reserve leaked until TTL)
- init.py: lambda signatures synced
- v1/chat/completions/index.dspy: pass userid to reserve
 2026-08-04 18:35:53 +08:00
yumoqing
a02cd09459 fix: Redis connection error now returns no_redis fallback instead of 429 2026-07-31 15:42:51 +08:00
yumoqing
6351ee9b56 feat: Redis atomic balance reservation — prevent concurrent overspend 
- balance.py: reserve_balance/finalize_balance/refund_balance with Lua
- utils.py: get_model_max_cost from llmusage history, update_model_max_cost
- init.py: register reserve_balance, finalize_balance, refund_balance on env
- chat/completions/index.dspy: reserve before inference, refund on exception
- DB migration: llm.max_cost DECIMAL(10,4) for historical max charge
 2026-07-31 14:57:17 +08:00