Hacktoberfest 2026: le issue che i maintainer hanno segnato per ottobre, aperte e adatte ai principianti. Sfoglia le issue Hacktoberfest

[GUI] Monitor: the Conversation cache card shows no parked conversations or memory for a native model

Aperta Adatta ai principianti
  • Qualcosa non va?
#4,205 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

I maintainer di solito rispondono entro 3 giorni

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
2/5
Tempo stimato
1-3 ore
Idoneità per principianti
75/100
Tipo di issue
Bug
Chiarezza
Specificata chiaramente
Stato di attività
Attiva
Stack tecnologico
python, typescript
Ambito
desktop, observability

Direzione di ricerca

Leggi apps/host/src/lib/native-monitor.ts e apps/host/src/screens/Monitor.tsx, poi confronta nativeMonitorData() con la mappatura in apps/host/server/monitor-dashboard.mjs. Con un modello nativo caricato, confronta i campi conversation_cache dell’istanza in GET /metrics con la scheda Monitor. Il lavoro è completato quando la scheda mostra le conversazioni parcheggiate, la memoria rispetto al budget e i conteggi di parcheggi e ripristini ricavati dalle metriche native.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

needs-format

Area：[GUI] (the desktop app)
Desktop app, Terminal or source?：Desktop app (Windows)
Reproduction status：Reproduced locally
Strata version： Strata 0.2.5 · engine 0.1.50
Graphics card, memory and system：RTX3070 16GB MOD · 80 GB RAM · Windows 11
Model, size and settings：Qwen3.8-Flash-Next IQ3_S · 256K context · one GPU · no --parallel

What happened

Problem: In the desktop app's Monitor, the "Conversation cache" card never shows the
"Parked conversations" and "Their memory (RAM)" bars, or the "Parked / restored" row, while a
native model is loaded. The card only shows the two lines it computes from the request list
("7 of 11 requests reused prompt tokens", "Follow-up requests can reuse the previous prompt's
cached tokens"). So it looks as if the conversation cache holds nothing in RAM.

Expected: the parked conversations, their RAM against the budget, and the parks/restores,
the same as the card shows when the dashboard comes from the server's /metrics
(apps/host/server/monitor-dashboard.mjs).

Evidence that the engine does park (from the attached screenshot): in "Recent requests" a
373-token request with Reused 0 (22:39:16) is followed by a 16,477-token request with
Reused 16,446 (22:40:06), and a 373-token request with Reused 0 (22:41:33) is followed by a
52,989-token request with Reused 51,339 (22:54:58). No --parallel, so there are no batch slots.

Suspected cause (unconfirmed static analysis: I read the code, I did not debug it):

  • apps/host/src/screens/Monitor.tsx: with native instances the screen uses
    nativeMonitorData(native, ...) instead of the stored monitor state.
  • apps/host/src/lib/native-monitor.ts, nativeMonitorData(): it maps live, engine, requests,
    totals and hardware from the instance's /metrics, but not conversation_cache. The returned
    MonitorState has no conversationCache, so cache?.enabled in Monitor.tsx is undefined and
    the bars and the parked/restored row are not rendered.
  • The server does send it: serve/server.py metrics() returns "conversation_cache"
    (conversation_cache_view: enabled, slots, parked, bytes, budget_mib, parks, restores, ...),
    and apps/host/server/monitor-dashboard.mjs maps it to conversationCache for the other path.
  • main has the same native-monitor.ts (no "conversation" in it) as of 2026-10-26.

What would verify it: with a native model loaded, compare GET /metrics on the instance's port
(conversation_cache.parked / bytes / enabled) with what the Monitor card shows.

Related, but not the same: #596 (the card), #3033 (stale numbers after a refused park),
#3039 (double-counted parks), #4076 (other blank Monitor tiles on Windows).

Smallest steps
  1. Desktop app on Windows, load Qwen3.8-Flash-Next IQ3_S (native model, one GPU, no --parallel).
  2. Send a long chat (tens of thousands of tokens), then a short unrelated request, then continue the long chat.
  3. Open Monitor and look at the Conversation cache card.
Lingua principale
C++
Stelle
11.6k
Fork
1k
Nessuna PR unita negli ultimi 30g

Preparare l'ambiente

Questo progetto non fornisce container di sviluppo, Dockerfile né guida per i contributori, quindi l'ambiente è a tuo carico: parti dal suo README e consulta la nostra guida al primo contributo per i passaggi generali.

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di Niko1221/Strata

Tutte le issue di Niko1221/Strata

Issue simili

Altre issue su C++

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.