Add reachability checks and develop/main container pipeline (inactive)
- preflight --live: Ollama/Qdrant reachable from the container, required Ollama models present, 768 dimensions; clear abort instead of traceback. Static check rejects unset or container-local service URLs. - deploy/smoke.py: non-writing post-start check via published ports. - deploy/deploy.sh: test (no network), build, preflight, up --wait, smoke; prod rolls back to the last good image tag. - Workflow templates deploy-dev.yml (push develop) and deploy-prod.yml (push main) following the Kansho pattern; not under .gitea/workflows. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
1 parent
7a225ba2b6
commit
8f9f83aa79
14 files changed
+728
-90
No files matched your search
+87
-20
@@ -7,7 +7,10 @@ DESCRIPTION: Container-Entrypoint für Mindnet (API, UI, Preflight).
|
||||
- explizite Umgebung und Collection-Präfix (Dev nie auf Prod-Präfix)
|
||||
- Vault: vorhanden, Sentinel lesbar (NAS gemountet), schreibgeschützt
|
||||
- kein Cloud-LLM-Key ohne ausdrückliche Freigabe
|
||||
VERSION: 1.0.0
|
||||
- Dienst-URLs gesetzt und nicht auf den Container selbst (localhost)
|
||||
Mit 'preflight --live' zusätzlich (nur lesend): Ollama und Qdrant
|
||||
erreichbar, benötigte Modelle vorhanden, Dimensionen stimmen.
|
||||
VERSION: 1.1.0
|
||||
STATUS: Active
|
||||
USAGE: entrypoint.py api | ui | preflight [--live]
|
||||
"""
|
||||
@@ -17,6 +20,7 @@ import json
|
||||
import os
|
||||
import sys
|
||||
import urllib.error
|
||||
import urllib.parse
|
||||
import urllib.request
|
||||
from pathlib import Path
|
||||
|
||||
@@ -24,6 +28,8 @@ import yaml
|
||||
|
||||
CLOUD_KEYS = ("OPENROUTER_API_KEY", "GOOGLE_API_KEY")
|
||||
CLOUD_EGRESS_ACK = "MINDNET_ALLOW_CLOUD_EGRESS"
|
||||
SERVICE_URLS = ("MINDNET_OLLAMA_URL", "QDRANT_URL")
|
||||
LOOPBACK_HOSTS = {"localhost", "127.0.0.1", "::1", "0.0.0.0"}
|
||||
|
||||
|
||||
class PreflightError(RuntimeError):
|
||||
@@ -49,6 +55,16 @@ def check_environment() -> str:
|
||||
return env
|
||||
|
||||
|
||||
def _load_profiles(profiles_path: str | None = None) -> dict:
|
||||
# Gleicher Pfad wie in der App (Settings kennt keinen Override dafür).
|
||||
path = Path(profiles_path or "config/llm_profiles.yaml")
|
||||
try:
|
||||
data = yaml.safe_load(path.read_text(encoding="utf-8")) or {}
|
||||
except (OSError, yaml.YAMLError) as e:
|
||||
raise PreflightError(f"LLM-Profile nicht lesbar: {path} ({type(e).__name__}).") from None
|
||||
return data.get("profiles") or {}
|
||||
|
||||
|
||||
def check_embeddings(profiles_path: str | None = None) -> None:
|
||||
expected_dim = int(os.getenv("MINDNET_EXPECTED_VECTOR_DIM", "768"))
|
||||
expected_model = os.getenv("MINDNET_EXPECTED_EMBEDDING_MODEL", "nomic-embed-text")
|
||||
@@ -63,13 +79,7 @@ def check_embeddings(profiles_path: str | None = None) -> None:
|
||||
raise PreflightError(f"MINDNET_EMBEDDING_MODEL='{env_model}' weicht von '{expected_model}' ab.")
|
||||
|
||||
# Das Profil hat Vorrang vor der Umgebung; fehlt es, fällt der Client auf das Chat-Modell zurück.
|
||||
# Gleicher Pfad wie in der App (Settings kennt keinen Override dafür).
|
||||
path = Path(profiles_path or "config/llm_profiles.yaml")
|
||||
try:
|
||||
data = yaml.safe_load(path.read_text(encoding="utf-8")) or {}
|
||||
except (OSError, yaml.YAMLError) as e:
|
||||
raise PreflightError(f"LLM-Profile nicht lesbar: {path} ({type(e).__name__}).") from None
|
||||
profile = (data.get("profiles") or {}).get("embedding_expert") or {}
|
||||
profile = _load_profiles(profiles_path).get("embedding_expert") or {}
|
||||
if profile.get("provider", "ollama") != "ollama" or profile.get("model") != expected_model:
|
||||
raise PreflightError(f"Profil embedding_expert muss ollama/{expected_model} sein.")
|
||||
if int(profile.get("dimensions") or 0) != expected_dim:
|
||||
@@ -109,36 +119,92 @@ def check_vault() -> None:
|
||||
raise PreflightError(f"Vault '{root}' ist nicht schreibgeschützt gemountet (':ro' erforderlich).")
|
||||
|
||||
|
||||
def check_live() -> None:
|
||||
"""Optionale Laufzeitprüfung: tatsächliche Embedding- und Collection-Dimension (nur lesend)."""
|
||||
def check_service_urls() -> None:
|
||||
for name in SERVICE_URLS:
|
||||
url = os.getenv(name, "").strip()
|
||||
if not url:
|
||||
# Die App fiele sonst still auf http://127.0.0.1:... zurück, also auf den Container selbst.
|
||||
raise PreflightError(f"{name} muss explizit gesetzt sein.")
|
||||
if urllib.parse.urlsplit(url).hostname in LOOPBACK_HOSTS:
|
||||
raise PreflightError(
|
||||
f"{name}='{url}' zeigt im Container auf den Container selbst. "
|
||||
"Host-Dienste über host.docker.internal ansprechen."
|
||||
)
|
||||
|
||||
|
||||
def _fetch_json(url: str, payload: dict | None = None, headers: dict | None = None, timeout: float = 10):
|
||||
"""GET (oder POST mit payload). Netzwerkfehler werden zu PreflightError, HTTPError bleibt erhalten."""
|
||||
req = urllib.request.Request(
|
||||
url, method="POST" if payload is not None else "GET",
|
||||
data=json.dumps(payload).encode() if payload is not None else None,
|
||||
headers={"Content-Type": "application/json", **(headers or {})},
|
||||
)
|
||||
try:
|
||||
with urllib.request.urlopen(req, timeout=timeout) as res:
|
||||
return json.load(res)
|
||||
except urllib.error.HTTPError:
|
||||
raise
|
||||
except (urllib.error.URLError, OSError) as e:
|
||||
hint = ""
|
||||
if urllib.parse.urlsplit(url).hostname == "host.docker.internal":
|
||||
hint = " Lauscht der Dienst auf dem Host nur an 127.0.0.1?"
|
||||
raise PreflightError(f"{url} nicht erreichbar ({getattr(e, 'reason', e)}).{hint}") from None
|
||||
except ValueError:
|
||||
raise PreflightError(f"{url} liefert kein JSON.") from None
|
||||
|
||||
|
||||
def required_ollama_models(profiles_path: str | None = None) -> set[str]:
|
||||
models = {p["model"] for p in _load_profiles(profiles_path).values()
|
||||
if isinstance(p, dict) and p.get("provider") == "ollama" and p.get("model")}
|
||||
# Lokales Chat-Modell: Endpunkt der Fallback-Kette ohne Cloud-Keys (Review R4).
|
||||
if os.getenv("MINDNET_LLM_MODEL", "").strip():
|
||||
models.add(os.environ["MINDNET_LLM_MODEL"].strip())
|
||||
return models
|
||||
|
||||
|
||||
def _has_model(required: str, available: set[str]) -> bool:
|
||||
return required in available or (":" not in required and f"{required}:latest" in available)
|
||||
|
||||
|
||||
def check_live(profiles_path: str | None = None) -> None:
|
||||
"""Laufzeitprüfung (nur lesend): Erreichbarkeit, Modelle, Embedding- und Collection-Dimension."""
|
||||
expected_dim = int(os.getenv("MINDNET_EXPECTED_VECTOR_DIM", "768"))
|
||||
ollama = os.getenv("MINDNET_OLLAMA_URL", "").rstrip("/")
|
||||
model = os.getenv("MINDNET_EXPECTED_EMBEDDING_MODEL", "nomic-embed-text")
|
||||
req = urllib.request.Request(
|
||||
f"{ollama}/api/embeddings", method="POST",
|
||||
data=json.dumps({"model": model, "prompt": "preflight"}).encode(),
|
||||
headers={"Content-Type": "application/json"},
|
||||
)
|
||||
with urllib.request.urlopen(req, timeout=30) as res:
|
||||
got = len(json.load(res).get("embedding") or [])
|
||||
|
||||
tags = _fetch_json(f"{ollama}/api/tags")
|
||||
available = {m.get("name") for m in tags.get("models") or []}
|
||||
missing = sorted(m for m in required_ollama_models(profiles_path) if not _has_model(m, available))
|
||||
if missing:
|
||||
raise PreflightError(f"Ollama ({ollama}) fehlen die Modelle: {', '.join(missing)}.")
|
||||
|
||||
embedding = _fetch_json(f"{ollama}/api/embeddings", {"model": model, "prompt": "preflight"}, timeout=30)
|
||||
got = len(embedding.get("embedding") or [])
|
||||
if got != expected_dim:
|
||||
raise PreflightError(f"Ollama liefert {got} Dimensionen für '{model}', erwartet {expected_dim}.")
|
||||
print(f"preflight: Ollama erreichbar ({ollama}), Modelle vorhanden, '{model}' = {got} Dimensionen.")
|
||||
|
||||
qdrant = os.getenv("QDRANT_URL", "").rstrip("/")
|
||||
headers = {"api-key": os.environ["QDRANT_API_KEY"]} if os.getenv("QDRANT_API_KEY") else {}
|
||||
try:
|
||||
_fetch_json(f"{qdrant}/collections", headers=headers)
|
||||
except urllib.error.HTTPError as e:
|
||||
raise PreflightError(f"Qdrant ({qdrant}) antwortet mit HTTP {e.code}.") from None
|
||||
prefix = os.getenv("COLLECTION_PREFIX", "")
|
||||
for suffix in ("notes", "chunks"):
|
||||
name = f"{prefix}_{suffix}"
|
||||
try:
|
||||
with urllib.request.urlopen(f"{qdrant}/collections/{name}", timeout=10) as res:
|
||||
vectors = json.load(res)["result"]["config"]["params"]["vectors"]
|
||||
info = _fetch_json(f"{qdrant}/collections/{name}", headers=headers)
|
||||
except urllib.error.HTTPError as e:
|
||||
if e.code == 404:
|
||||
print(f"preflight: Collection '{name}' existiert noch nicht.")
|
||||
continue
|
||||
raise
|
||||
raise PreflightError(f"Qdrant-Collection '{name}': HTTP {e.code}.") from None
|
||||
vectors = info["result"]["config"]["params"]["vectors"]
|
||||
sizes = [vectors["size"]] if "size" in vectors else [v["size"] for v in vectors.values()]
|
||||
if any(s != expected_dim for s in sizes):
|
||||
raise PreflightError(f"Collection '{name}' hat Dimension {sizes}, erwartet {expected_dim}.")
|
||||
print(f"preflight: Qdrant erreichbar ({qdrant}), Collections '{prefix}_*' passen.")
|
||||
|
||||
|
||||
def run_checks(mode: str) -> None:
|
||||
@@ -146,6 +212,7 @@ def run_checks(mode: str) -> None:
|
||||
check_embeddings()
|
||||
check_cloud_egress()
|
||||
if mode in {"api", "preflight"}:
|
||||
check_service_urls()
|
||||
check_vault()
|
||||
|
||||
|
||||
|
||||
Reference in new issue
Block a user