throughput: authoritative model-state tracking (real-GGUF attribution, timestamped store, gateway-recreate drift fix) - #182

Merged

AlienWalker1995 merged 9 commits into

mainfrom

feat/throughput-model-state

Aug 20, 2026

Commits on Aug 19, 2026

test: isolate DASHBOARD_DATA_PATH so pytest cannot pollute the live throughput store

Hermes Bot
and
claude
committed
169e672
dashboard: throughput store v2 - timestamped samples, 7-day eviction, clean reset of alias-conflated v1 data

Hermes Bot
and
claude
committed
5c8b6a1
test: assert alias/backend actually captured in service_usage event

Hermes Bot
and
claude
committed
c738128
dashboard: /api/throughput/stats carries the authoritative active model from ops /model-config

Hermes Bot
and
claude
committed
f9a9b7a
dashboard: guard active-model fetch with asyncio.Lock to prevent concurrent ops-controller calls

Hermes Bot
and
claude
committed
ccf65e4
fix(ops-api): recreate model-gateway on any gateway-templated key change, not only ctx

Hermes Bot
and
claude
committed
5945256
model-gateway: throughput samples attribute to the GGUF that actually served, not the alias

Hermes Bot
and
claude
committed
dc07c71
dashboard: throughput hero is the registry's active model - guessing heuristics deleted

Hermes Bot
and
claude
committed
cd7465b
fix(throughput): attribute control-plane state precisely, dedupe alias derivation

Hermes Bot
and
claude
committed
4cd269a