enne2
fcd6b1670e
- embed.py: EMBED_CHAIN (JSON per-nodo {name,url,api,key,timeout_ms}, api
llamacpp|ollama), cooldown 60s sui nodi falliti, validazione dimensione
EMBED_DIM, compatibilità legacy quando la catena è vuota
- state.py: ResilientQdrant — proxy che ritenta i metodi del client Qdrant
su httpx.TransportError (store/search/transienti), errori applicativi
esenti; contatore qdrant_retries in metriche
- metrics: qmem_embed_calls_total + durata per backend
- /v1/version espone embed_nodes; versione 2.10.0
- test: 11 nuovi (chain, cooldown, dim mismatch, legacy, retry transiente) — 58 pass
66 lines
3.1 KiB
Python
66 lines
3.1 KiB
Python
"""Configurazione statica del Memory Gateway letta dall'ambiente."""
|
|
from __future__ import annotations
|
|
|
|
import logging
|
|
import os
|
|
from collections import Counter
|
|
from typing import Any
|
|
|
|
QDRANT_URL = os.environ.get("QDRANT_URL", "http://127.0.0.1:6333")
|
|
QDRANT_API_KEY = os.environ.get("QDRANT_API_KEY", "")
|
|
EMBED_API = os.environ.get("EMBED_API", "ollama")
|
|
EMBED_URL = os.environ.get("EMBED_URL", os.environ.get("OLLAMA_URL", "http://127.0.0.1:11434"))
|
|
EMBED_MODEL = os.environ.get("EMBED_MODEL", "bge-m3")
|
|
EMBED_API_KEY = os.environ.get("EMBED_API_KEY", "")
|
|
EMBED_DIM = int(os.environ.get("EMBED_DIM", "1024"))
|
|
# Catena di fallback per gli embedding (JSON, formato RERANK_CHAIN + campo "api").
|
|
# Vuota → comportamento legacy: endpoint singolo da EMBED_API/EMBED_URL/EMBED_API_KEY.
|
|
EMBED_CHAIN = os.environ.get("EMBED_CHAIN", "")
|
|
EMBED_TIMEOUT_MS = int(os.environ.get("EMBED_TIMEOUT_MS", "30000"))
|
|
EMBED_RETRY_COOLDOWN_S = int(os.environ.get("EMBED_RETRY_COOLDOWN_S", "60"))
|
|
# Retry transiente per le chiamate Qdrant (store/search inclusi)
|
|
QDRANT_RETRIES = int(os.environ.get("QDRANT_RETRIES", "3"))
|
|
COLLECTION = os.environ.get("COLLECTION", "memories")
|
|
API_KEYS: set[str] = {k.strip() for k in os.environ.get("API_KEYS", "").split(",") if k.strip()}
|
|
RATE_LIMIT_PER_MIN = int(os.environ.get("RATE_LIMIT_PER_MIN", "120"))
|
|
MAX_TEXT_LEN = int(os.environ.get("MAX_TEXT_LEN", "8000"))
|
|
|
|
GUARDRAIL_ENABLED = os.environ.get("GUARDRAIL_ENABLED", "true").lower() == "true"
|
|
GUARDRAIL_BLOCK_THRESHOLD = float(os.environ.get("GUARDRAIL_BLOCK_THRESHOLD", "0.85"))
|
|
GUARDRAIL_WARN_THRESHOLD = float(os.environ.get("GUARDRAIL_WARN_THRESHOLD", "0.70"))
|
|
GUARDRAIL_VERSION = "similarity-v1"
|
|
GIT_COMMIT = os.environ.get("GIT_COMMIT", "unknown").strip()
|
|
GATEWAY_VERSION = os.environ.get("GATEWAY_VERSION", "2.10.0").strip()
|
|
|
|
VM_PUSH_URL = os.environ.get("VM_PUSH_URL", "http://host.docker.internal:8428/api/v1/import/prometheus")
|
|
VM_PUSH_INTERVAL = int(os.environ.get("VM_PUSH_INTERVAL", "30"))
|
|
METRICS_ENABLED = os.environ.get("METRICS_ENABLED", "true").lower() == "true"
|
|
|
|
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
|
|
log = logging.getLogger("memory-gateway")
|
|
|
|
SPARSE_VECTOR_NAME = "bm25"
|
|
|
|
# Re-ranking: catena di fallback resiliente (frigate → brain locale).
|
|
# Il default nel codice è OFF; il deploy imposta RERANK_ENABLED=true e la catena.
|
|
RERANK_ENABLED = os.environ.get("RERANK_ENABLED", "false").lower() == "true"
|
|
RERANK_MODEL = os.environ.get("RERANK_MODEL", "bge-reranker-v2-m3")
|
|
RERANK_CANDIDATES = int(os.environ.get("RERANK_CANDIDATES", "16"))
|
|
RERANK_MAX_DOC_CHARS = int(os.environ.get("RERANK_MAX_DOC_CHARS", "800"))
|
|
RERANK_TIMEOUT_MS = int(os.environ.get("RERANK_TIMEOUT_MS", "10000"))
|
|
RERANK_RETRY_COOLDOWN_S = int(os.environ.get("RERANK_RETRY_COOLDOWN_S", "60"))
|
|
RERANK_CHAIN = os.environ.get("RERANK_CHAIN", "")
|
|
_metrics: dict[str, Any] = {
|
|
"requests": Counter(),
|
|
"duration_sum": Counter(),
|
|
"duration_count": Counter(),
|
|
"errors": Counter(),
|
|
"search_queries": 0,
|
|
"search_hits": 0,
|
|
"rerank_calls": Counter(),
|
|
"rerank_duration_sum": Counter(),
|
|
"embed_calls": Counter(),
|
|
"embed_duration_sum": Counter(),
|
|
"qdrant_retries": 0,
|
|
}
|