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
+2
-1
@@ -59,4 +59,5 @@ COPY tests/ tests/
|
||||
COPY deploy/ deploy/
|
||||
USER mindnet:mindnet
|
||||
ENTRYPOINT []
|
||||
CMD ["python", "-m", "pytest", "-q", "-p", "no:cacheprovider", "tests/test_container_preflight.py"]
|
||||
CMD ["python", "-m", "pytest", "-q", "-p", "no:cacheprovider", \
|
||||
"tests/test_container_preflight.py", "tests/test_container_connectivity.py"]
|
||||
@@ -15,7 +15,7 @@ x-hardening: &hardening
|
||||
context: ..
|
||||
dockerfile: deploy/Dockerfile
|
||||
target: runtime
|
||||
env_file: [env/dev.env]
|
||||
env_file: ["${MINDNET_ENV_FILE:-env/dev.env}"]
|
||||
read_only: true
|
||||
tmpfs: ["/tmp:size=64m,mode=1777"]
|
||||
cap_drop: [ALL]
|
||||
|
||||
@@ -17,7 +17,7 @@ x-hardening: &hardening
|
||||
context: ..
|
||||
dockerfile: deploy/Dockerfile
|
||||
target: runtime
|
||||
env_file: [env/prod.env]
|
||||
env_file: ["${MINDNET_ENV_FILE:-env/prod.env}"]
|
||||
read_only: true
|
||||
tmpfs: ["/tmp:size=64m,mode=1777"]
|
||||
cap_drop: [ALL]
|
||||
|
||||
Executable
+89
@@ -0,0 +1,89 @@
|
||||
#!/usr/bin/env bash
|
||||
# Baut, prüft und startet einen Mindnet-Container-Stack. Genutzt von den
|
||||
# Gitea-Workflows (develop -> dev, main -> prod) und für manuelle Deploys.
|
||||
#
|
||||
# MINDNET_ENV_FILE=/home/llmadmin/docker/mindnet-dev/dev.env deploy/deploy.sh dev
|
||||
#
|
||||
# Ablauf (jeder Schritt bricht ab, bevor der laufende Stack angefasst wird):
|
||||
# 1. Container-Tests im Test-Image, ohne Netzwerk (blockierend)
|
||||
# 2. volle Testsuite ohne Netzwerk (informativ, bekannte Fehler siehe Migrationsplan)
|
||||
# 3. Image bauen, Tag = Umgebung + Commit
|
||||
# 4. preflight --live: Ollama/Qdrant erreichbar, Modelle, 768 Dimensionen, Vault (nur lesend)
|
||||
# 5. up --wait (Healthchecks), danach smoke.py über die veröffentlichten Ports
|
||||
# 6. Prod: bei Fehler Rückkehr zum letzten erfolgreichen Tag
|
||||
#
|
||||
# Die env-Datei liegt auf dem Host außerhalb von Git; in ihrem Verzeichnis
|
||||
# merkt sich das Skript den letzten erfolgreichen Image-Tag (last_good_tag).
|
||||
set -euo pipefail
|
||||
cd "$(dirname "$0")/.."
|
||||
|
||||
ENV="${1:-}"
|
||||
case "$ENV" in
|
||||
dev|prod) ;;
|
||||
*) echo "Aufruf: deploy/deploy.sh dev|prod" >&2; exit 2 ;;
|
||||
esac
|
||||
: "${MINDNET_ENV_FILE:?MINDNET_ENV_FILE auf die env-Datei des Hosts setzen}"
|
||||
[ -f "$MINDNET_ENV_FILE" ] || { echo "env-Datei fehlt: $MINDNET_ENV_FILE" >&2; exit 1; }
|
||||
export MINDNET_ENV_FILE
|
||||
|
||||
SHA="$(git rev-parse --short=12 HEAD)"
|
||||
export MINDNET_IMAGE_TAG="$ENV-$SHA"
|
||||
LAST_GOOD="$(dirname "$MINDNET_ENV_FILE")/last_good_tag"
|
||||
|
||||
if [ "$ENV" = dev ]; then
|
||||
PREFIX=mindnet_dev
|
||||
RUN_DEPS=() # startet die eigene Dev-Qdrant für den Preflight mit
|
||||
SMOKE_ARGS=()
|
||||
else
|
||||
PREFIX="$(sed -n 's/^COLLECTION_PREFIX=//p' "$MINDNET_ENV_FILE" | tail -n 1)"
|
||||
RUN_DEPS=(--no-deps) # Prod hat keine eigene Qdrant
|
||||
SMOKE_ARGS=(--require-data)
|
||||
fi
|
||||
|
||||
dc() { docker compose -f "deploy/compose.$ENV.yml" --env-file "$MINDNET_ENV_FILE" "$@"; }
|
||||
step() { printf '\n=== %s\n' "$*"; }
|
||||
show_logs() { dc ps || true; dc logs --tail 150 api ui || true; }
|
||||
published() { dc port "$1" "$2" | head -n 1 | sed 's/^0\.0\.0\.0:/127.0.0.1:/'; }
|
||||
|
||||
smoke() {
|
||||
python3 deploy/smoke.py --api "http://$(published api 8000)" --ui "http://$(published ui 8501)" \
|
||||
--prefix "$PREFIX" "${SMOKE_ARGS[@]}"
|
||||
}
|
||||
|
||||
step "1/5 Container-Tests (ohne Netzwerk)"
|
||||
docker build -f deploy/Dockerfile --target test -t "mindnet:test-$SHA" .
|
||||
docker run --rm --network none "mindnet:test-$SHA"
|
||||
|
||||
step "2/5 Volle Testsuite (informativ, ohne Netzwerk)"
|
||||
if ! docker run --rm --network none "mindnet:test-$SHA" \
|
||||
python -m pytest tests -q -p no:cacheprovider --continue-on-collection-errors --basetemp=/tmp/pytest; then
|
||||
echo "WARNUNG: Testsuite nicht grün. Bekannte Fehler: docs/05_Development/05_docker_migration_plan.md"
|
||||
fi
|
||||
docker image rm "mindnet:test-$SHA" >/dev/null || true
|
||||
|
||||
step "3/5 Image bauen ($MINDNET_IMAGE_TAG)"
|
||||
dc build api ui
|
||||
|
||||
step "4/5 Preflight mit Erreichbarkeit (nur lesend)"
|
||||
dc run --rm "${RUN_DEPS[@]}" api preflight --live
|
||||
|
||||
step "5/5 Start ($MINDNET_IMAGE_TAG) und Smoke"
|
||||
if dc up -d --wait --no-build && smoke; then
|
||||
echo "$MINDNET_IMAGE_TAG" > "$LAST_GOOD"
|
||||
echo "=== mindnet-$ENV läuft mit $MINDNET_IMAGE_TAG"
|
||||
exit 0
|
||||
fi
|
||||
|
||||
show_logs
|
||||
PREV="$(cat "$LAST_GOOD" 2>/dev/null || true)"
|
||||
if [ "$ENV" = prod ] && [ -n "$PREV" ] && [ "$PREV" != "$MINDNET_IMAGE_TAG" ]; then
|
||||
step "Rollback auf $PREV"
|
||||
export MINDNET_IMAGE_TAG="$PREV"
|
||||
if dc up -d --wait --no-build && smoke; then
|
||||
echo "Rollback erfolgreich; Deploy von $ENV-$SHA ist trotzdem fehlgeschlagen."
|
||||
else
|
||||
show_logs
|
||||
echo "ROLLBACK FEHLGESCHLAGEN, manueller Eingriff nötig." >&2
|
||||
fi
|
||||
fi
|
||||
exit 1
|
||||
+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()
|
||||
|
||||
|
||||
|
||||
@@ -0,0 +1,31 @@
|
||||
# VORLAGE, NICHT AKTIV: liegt absichtlich nicht unter .gitea/workflows/.
|
||||
# Aktivierung erst mit Freigabe (Gate G3, Dev-Start auf dem Runner-Host):
|
||||
# git mv deploy/gitea-workflows/deploy-dev.yml .gitea/workflows/deploy-dev.yml
|
||||
# Danach deployt jeder Push auf develop den Dev-Stack (Muster wie Kansho).
|
||||
# Ablauf und Abbruchbedingungen: deploy/deploy.sh.
|
||||
|
||||
name: Deploy mindnet DEV (Container)
|
||||
|
||||
on:
|
||||
push:
|
||||
branches: [develop]
|
||||
workflow_dispatch:
|
||||
|
||||
concurrency:
|
||||
group: deploy-mindnet-dev
|
||||
cancel-in-progress: false
|
||||
|
||||
jobs:
|
||||
deploy:
|
||||
# Runner auf dem Zielhost (heute llm-node). Bei einem Umzug auf einen anderen
|
||||
# Mini-PC nur das Label und den Pfad der env-Datei anpassen.
|
||||
runs-on: linux_host
|
||||
env:
|
||||
# Liegt auf dem Host, außerhalb von Git (Vorlage: deploy/env/dev.env.example).
|
||||
MINDNET_ENV_FILE: /home/llmadmin/docker/mindnet-dev/dev.env
|
||||
steps:
|
||||
- name: Checkout
|
||||
uses: actions/checkout@v4
|
||||
|
||||
- name: Test, Build, Preflight, Start, Smoke
|
||||
run: bash deploy/deploy.sh dev
|
||||
@@ -1,57 +0,0 @@
|
||||
# VORLAGE – NICHT AKTIV. Liegt absichtlich außerhalb von .gitea/workflows/.
|
||||
# Aktivierung (Kopie nach .gitea/workflows/deploy-dev.yml) erst nach Freigabe,
|
||||
# da der Runner `linux_host` auf llm-node (= Produktionsserver) läuft.
|
||||
#
|
||||
# Muster wie Kansho, aber nur manuell auslösbar (workflow_dispatch), kein Push-Trigger.
|
||||
# Ein PROD-Workflow für Container existiert bewusst noch nicht; der bestehende
|
||||
# .gitea/workflows/deploy.yml (systemd, Push auf main) bleibt unverändert.
|
||||
|
||||
name: Deploy mindnet DEV (Container)
|
||||
|
||||
on:
|
||||
workflow_dispatch:
|
||||
|
||||
concurrency:
|
||||
group: deploy-mindnet-dev
|
||||
cancel-in-progress: false
|
||||
|
||||
jobs:
|
||||
deploy:
|
||||
runs-on: linux_host
|
||||
env:
|
||||
TARGET: /home/llmadmin/docker/mindnet-dev
|
||||
COMPOSE: docker compose -f deploy/compose.dev.yml --env-file deploy/env/dev.env
|
||||
steps:
|
||||
- name: Checkout develop
|
||||
run: |
|
||||
set -euo pipefail
|
||||
REPO="http://192.168.2.144:3000/Lars/mindnet.git"
|
||||
mkdir -p "$TARGET" && cd "$TARGET"
|
||||
if [ ! -d .git ]; then
|
||||
git clone -b develop "$REPO" .
|
||||
else
|
||||
git fetch origin develop
|
||||
git checkout develop
|
||||
git reset --hard origin/develop
|
||||
fi
|
||||
test -f deploy/env/dev.env || { echo "deploy/env/dev.env fehlt auf dem Host"; exit 1; }
|
||||
|
||||
- name: Build (Tag = Commit)
|
||||
run: |
|
||||
set -euo pipefail
|
||||
cd "$TARGET"
|
||||
export MINDNET_IMAGE_TAG="dev-$(git rev-parse --short HEAD)"
|
||||
$COMPOSE build --pull
|
||||
$COMPOSE run --rm --no-deps api preflight
|
||||
|
||||
- name: Start und Health
|
||||
run: |
|
||||
set -euo pipefail
|
||||
cd "$TARGET"
|
||||
export MINDNET_IMAGE_TAG="dev-$(git rev-parse --short HEAD)"
|
||||
if ! $COMPOSE up -d --wait; then
|
||||
$COMPOSE logs api --tail 150 || true
|
||||
$COMPOSE ps || true
|
||||
exit 1
|
||||
fi
|
||||
curl -sf "http://127.0.0.1:${MINDNET_API_PORT:-8012}/healthz"
|
||||
@@ -0,0 +1,33 @@
|
||||
# VORLAGE, NICHT AKTIV: liegt absichtlich nicht unter .gitea/workflows/.
|
||||
# Aktivierung erst bei der Umschaltung (Gate G5, eigene Freigabe), und zwar im
|
||||
# selben Commit, der .gitea/workflows/deploy.yml (systemd) entfernt:
|
||||
# git rm .gitea/workflows/deploy.yml
|
||||
# git mv deploy/gitea-workflows/deploy-prod.yml .gitea/workflows/deploy-prod.yml
|
||||
# Sonst deployen bei jedem Push auf main beide Wege.
|
||||
#
|
||||
# Prod nur über Merge-Commit develop -> main (git merge --no-ff develop), wie bei Kansho.
|
||||
# Ablauf, Abbruchbedingungen und Rollback: deploy/deploy.sh.
|
||||
|
||||
name: Deploy mindnet PROD (Container)
|
||||
|
||||
on:
|
||||
push:
|
||||
branches: [main]
|
||||
workflow_dispatch:
|
||||
|
||||
concurrency:
|
||||
# Gleiche Gruppe wie der systemd-Workflow: nie zwei Prod-Deploys gleichzeitig.
|
||||
group: deploy-mindnet
|
||||
cancel-in-progress: false
|
||||
|
||||
jobs:
|
||||
deploy:
|
||||
runs-on: linux_host
|
||||
env:
|
||||
MINDNET_ENV_FILE: /home/llmadmin/docker/mindnet/prod.env
|
||||
steps:
|
||||
- name: Checkout
|
||||
uses: actions/checkout@v4
|
||||
|
||||
- name: Test, Build, Preflight, Start, Smoke (Rollback bei Fehler)
|
||||
run: bash deploy/deploy.sh prod
|
||||
+113
@@ -0,0 +1,113 @@
|
||||
"""
|
||||
FILE: deploy/smoke.py
|
||||
DESCRIPTION: Nicht-schreibender Smoke-Test gegen einen laufenden Mindnet-Stack,
|
||||
aus Sicht des Hosts über die veröffentlichten Ports. Nur Standardbibliothek,
|
||||
läuft daher direkt auf dem Runner (python3), ohne venv.
|
||||
- API /healthz: Status ok, erwartetes Collection-Präfix
|
||||
- API /admin/stats: Vektorgröße 768; API erreicht Qdrant (Collection-Counts)
|
||||
- UI /_stcore/health
|
||||
--require-data (Prod): alle Collections vorhanden und Notes nicht leer.
|
||||
VERSION: 1.0.0
|
||||
STATUS: Active
|
||||
USAGE: smoke.py --api http://127.0.0.1:8012 --ui http://127.0.0.1:8512 --prefix mindnet_dev [--require-data]
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import sys
|
||||
import time
|
||||
import urllib.error
|
||||
import urllib.request
|
||||
|
||||
EXPECTED_VECTOR_SIZE = 768
|
||||
|
||||
|
||||
class SmokeError(RuntimeError):
|
||||
pass
|
||||
|
||||
|
||||
def _get(url: str, timeout: float = 15) -> bytes:
|
||||
try:
|
||||
with urllib.request.urlopen(url, timeout=timeout) as res:
|
||||
return res.read()
|
||||
except urllib.error.HTTPError as e:
|
||||
raise SmokeError(f"{url}: HTTP {e.code}") from None
|
||||
except (urllib.error.URLError, OSError) as e:
|
||||
raise SmokeError(f"{url} nicht erreichbar ({getattr(e, 'reason', e)})") from None
|
||||
|
||||
|
||||
def _get_json(url: str) -> dict:
|
||||
try:
|
||||
return json.loads(_get(url))
|
||||
except ValueError:
|
||||
raise SmokeError(f"{url} liefert kein JSON") from None
|
||||
|
||||
|
||||
def check_api(api: str, prefix: str, require_data: bool) -> list[str]:
|
||||
"""Gibt Warnungen zurück; harte Fehler als SmokeError."""
|
||||
health = _get_json(f"{api}/healthz")
|
||||
if health.get("status") != "ok":
|
||||
raise SmokeError(f"/healthz meldet {health.get('status')!r}")
|
||||
if health.get("prefix") != prefix:
|
||||
raise SmokeError(f"API nutzt Präfix {health.get('prefix')!r}, erwartet {prefix!r}")
|
||||
|
||||
stats = _get_json(f"{api}/admin/stats")
|
||||
size = (stats.get("config") or {}).get("vector_size")
|
||||
if size != EXPECTED_VECTOR_SIZE:
|
||||
raise SmokeError(f"API-Vektorgröße {size}, erwartet {EXPECTED_VECTOR_SIZE}")
|
||||
|
||||
warnings = []
|
||||
collections = stats.get("collections") or {}
|
||||
for kind in ("notes", "chunks", "edges"):
|
||||
entry = collections.get(kind) or {}
|
||||
count = entry.get("count", -1)
|
||||
if count < 0:
|
||||
# -1: Collection fehlt oder Qdrant ist aus dem API-Container nicht erreichbar.
|
||||
msg = f"Collection {entry.get('name', kind)!r} nicht lesbar (fehlt oder Qdrant nicht erreichbar)"
|
||||
if require_data:
|
||||
raise SmokeError(msg)
|
||||
warnings.append(msg)
|
||||
if require_data and (collections.get("notes") or {}).get("count", 0) <= 0:
|
||||
raise SmokeError("Notes-Collection ist leer")
|
||||
return warnings
|
||||
|
||||
|
||||
def check_ui(ui: str) -> None:
|
||||
body = _get(f"{ui}/_stcore/health").decode("utf-8", "replace").strip()
|
||||
if body != "ok":
|
||||
raise SmokeError(f"UI-Health meldet {body!r}")
|
||||
|
||||
|
||||
def run(api: str, ui: str, prefix: str, require_data: bool = False, attempts: int = 1, delay: float = 5) -> int:
|
||||
api, ui = api.rstrip("/"), ui.rstrip("/")
|
||||
for attempt in range(1, attempts + 1):
|
||||
try:
|
||||
warnings = check_api(api, prefix, require_data)
|
||||
check_ui(ui)
|
||||
except SmokeError as e:
|
||||
if attempt == attempts:
|
||||
print(f"smoke: FEHLER: {e}", file=sys.stderr)
|
||||
return 1
|
||||
time.sleep(delay)
|
||||
continue
|
||||
for w in warnings:
|
||||
print(f"smoke: WARNUNG: {w}")
|
||||
print(f"smoke: OK (api={api}, ui={ui}, prefix={prefix})")
|
||||
return 0
|
||||
return 1
|
||||
|
||||
|
||||
def main(argv: list[str] | None = None) -> int:
|
||||
parser = argparse.ArgumentParser(description=__doc__.split("USAGE:")[0])
|
||||
parser.add_argument("--api", required=True)
|
||||
parser.add_argument("--ui", required=True)
|
||||
parser.add_argument("--prefix", required=True)
|
||||
parser.add_argument("--require-data", action="store_true")
|
||||
parser.add_argument("--attempts", type=int, default=3)
|
||||
args = parser.parse_args(argv)
|
||||
return run(args.api, args.ui, args.prefix, args.require_data, args.attempts)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
sys.exit(main())
|
||||
Reference in new issue
Block a user