feat: ISO-Build-Orchestrierung fuer personalisierte Kunden-ISOs (Phase 4)

Neue iso_builds-Tabelle (Migration 0016) + vier Endpunkte, die
tuxflotte-installer/scripts/build_customer_iso.sh als Hintergrund-Thread
anstossen: POST/GET .../iso-builds (Trigger + Liste), GET
/api/v1/iso-builds/{id} (Status-Poll) und .../download (fertige ISO
ausliefern). Aeltere abgeschlossene Builds derselben Organisation werden
nach einem erfolgreichen neuen Build automatisch aufgeraeumt (Datei +
DB-Zeile) - anodes Root-Dateisystem ist mit 28GB knapp bemessen.

WLAN-Zugangsdaten werden bewusst nicht in iso_builds gespeichert, nur
transient an build_customer_iso.sh durchgereicht.

Lokal end-to-end gegen einen Wegwerf-Postgres-Container verifiziert (alle
Migrationen 0001-0016, echter build_customer_iso.sh-Lauf gegen die
gecachte Mint-ISO, Status-Uebergaenge pending->running->completed,
Download-Endpunkt liefert byte-identische Datei, Cleanup-Logik entfernt
den vorherigen Build einer Organisation nach dem naechsten erfolgreichen
Bau). Testorganisation danach wieder geloescht.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This commit is contained in:
Thomas Stallinger 2026-08-07 00:39:32 +02:00
parent 896c22c537
commit f69ed66267
2 changed files with 268 additions and 0 deletions

241
app.py
View File

@ -5,6 +5,8 @@ import hmac
import json
import os
import secrets
import subprocess
import threading
from uuid import UUID, uuid4
import psycopg
@ -22,6 +24,16 @@ ANSIBLE_CONTENT_REPO = os.environ.get("TUXFLOTTE_ANSIBLE_CONTENT_REPO", "")
ADMIN_TOKEN = os.environ.get("TUXFLOTTE_ADMIN_TOKEN", "")
KUNDENPLATTFORM_TOKEN = os.environ.get("TUXFLOTTE_KUNDENPLATTFORM_TOKEN", "")
AGENT_REPORTABLE_EVENT_TYPES = {"applied", "apply_failed", "removed", "remove_failed"}
# Phase 4 des Self-Service-ISO-Plans (siehe tuxflotte-installer/scripts/
# build_customer_iso.sh). TUXFLOTTE_INSTALLER_DIR ist ein separat auf anode
# geklontes Repo (nicht Teil von provisioning-server selbst), MINT_ISO_PATH
# das einmalig von der offiziellen Downloadseite geholte Quellabbild.
TUXFLOTTE_INSTALLER_DIR = os.environ.get("TUXFLOTTE_INSTALLER_DIR", "/root/tuxflotte-installer")
MINT_ISO_PATH = os.environ.get("TUXFLOTTE_MINT_ISO_PATH", "data/upstream/linuxmint-22.3-cinnamon-64bit.iso")
ISO_BUILD_OUTPUT_DIR = Path(os.environ.get("TUXFLOTTE_ISO_BUILD_OUTPUT_DIR", "data/iso-builds"))
ISO_BUILD_TIMEOUT_SECONDS = 1800
app = FastAPI(title="Provisioning Activation Server", version="0.1.0")
@ -102,6 +114,12 @@ class CreateEnrollmentSessionRequest(BaseModel):
expires_at: str
class CreateIsoBuildRequest(BaseModel):
activation_code: str
wifi_ssid: str | None = None
wifi_psk: str | None = None
def write_log(entry: dict) -> None:
with LOG_PATH.open("a", encoding="utf-8") as f:
f.write(json.dumps(entry, ensure_ascii=False) + "\n")
@ -1106,6 +1124,157 @@ def revoke_enrollment_session(session_id: str):
)
return cur.fetchone() is not None
def iso_build_row_to_dict(row) -> dict:
(build_id, organization_id, status, output_filename, error_message,
created_at, started_at, finished_at) = row
return {
"id": str(build_id),
"organization_id": str(organization_id),
"status": status,
"output_filename": output_filename,
"error_message": error_message,
"created_at": created_at.isoformat(),
"started_at": started_at.isoformat() if started_at else None,
"finished_at": finished_at.isoformat() if finished_at else None,
}
def fetch_iso_build(build_id: str):
with get_database_connection() as conn:
with conn.cursor() as cur:
cur.execute(
"""
SELECT id, organization_id, status, output_filename, error_message,
created_at, started_at, finished_at
FROM iso_builds
WHERE id = %s
""",
(build_id,),
)
row = cur.fetchone()
return iso_build_row_to_dict(row) if row else None
def fetch_iso_builds(organization_id: str):
with get_database_connection() as conn:
with conn.cursor() as cur:
cur.execute(
"""
SELECT id, organization_id, status, output_filename, error_message,
created_at, started_at, finished_at
FROM iso_builds
WHERE organization_id = %s
ORDER BY created_at DESC
""",
(organization_id,),
)
return [iso_build_row_to_dict(row) for row in cur.fetchall()]
def create_iso_build(organization_id: str) -> dict:
build_id = uuid4()
with get_database_connection() as conn:
with conn.cursor() as cur:
cur.execute(
"INSERT INTO iso_builds (id, organization_id) VALUES (%s, %s)",
(build_id, organization_id),
)
return {"id": str(build_id), "status": "pending"}
def update_iso_build(
build_id,
status: str,
output_filename: str | None = None,
error_message: str | None = None,
mark_started: bool = False,
mark_finished: bool = False,
):
with get_database_connection() as conn:
with conn.cursor() as cur:
cur.execute(
f"""
UPDATE iso_builds
SET status = %s,
output_filename = COALESCE(%s, output_filename),
error_message = %s
{", started_at = CURRENT_TIMESTAMP" if mark_started else ""}
{", finished_at = CURRENT_TIMESTAMP" if mark_finished else ""}
WHERE id = %s
""",
(status, output_filename, error_message, build_id),
)
def cleanup_old_iso_builds(organization_id: str, keep_build_id) -> None:
"""
Loescht Ausgabedateien + Zeilen abgeschlossener frueherer Baeulaeufe
derselben Organisation (Speicherplatz auf anode ist begrenzt) - eine neu
erzeugte Kunden-ISO ersetzt die vorherige, ein erneuter Download der
alten ergibt fuer den Kunden ohnehin keinen Sinn mehr.
"""
with get_database_connection() as conn:
with conn.cursor() as cur:
cur.execute(
"""
SELECT id, output_filename FROM iso_builds
WHERE organization_id = %s AND id != %s
AND status IN ('completed', 'failed')
""",
(organization_id, keep_build_id),
)
old_builds = cur.fetchall()
for old_id, output_filename in old_builds:
if output_filename:
(ISO_BUILD_OUTPUT_DIR / output_filename).unlink(missing_ok=True)
cur.execute(
"""
DELETE FROM iso_builds
WHERE organization_id = %s AND id != %s
AND status IN ('completed', 'failed')
""",
(organization_id, keep_build_id),
)
def run_iso_build(build_id, organization_id: str, activation_code: str, wifi_ssid: str | None, wifi_psk: str | None) -> None:
output_path = ISO_BUILD_OUTPUT_DIR / f"{build_id}.iso"
update_iso_build(build_id, status="running", mark_started=True)
try:
result = subprocess.run(
[
"bash",
f"{TUXFLOTTE_INSTALLER_DIR}/scripts/build_customer_iso.sh",
MINT_ISO_PATH,
str(output_path),
activation_code,
wifi_ssid or "",
wifi_psk or "",
],
cwd=TUXFLOTTE_INSTALLER_DIR,
capture_output=True,
text=True,
timeout=ISO_BUILD_TIMEOUT_SECONDS,
)
except subprocess.TimeoutExpired:
output_path.unlink(missing_ok=True)
update_iso_build(build_id, status="failed", error_message="Zeitüberschreitung beim ISO-Bau.", mark_finished=True)
return
except Exception as exc:
output_path.unlink(missing_ok=True)
update_iso_build(build_id, status="failed", error_message=str(exc)[:4000], mark_finished=True)
return
if result.returncode != 0 or not output_path.is_file():
output_path.unlink(missing_ok=True)
error_tail = (result.stderr or result.stdout or "unbekannter Fehler")[-4000:]
update_iso_build(build_id, status="failed", error_message=error_tail, mark_finished=True)
return
update_iso_build(build_id, status="completed", output_filename=output_path.name, mark_finished=True)
cleanup_old_iso_builds(organization_id, build_id)
@app.get("/health")
def health():
return {
@ -1710,6 +1879,78 @@ def revoke_enrollment_session_endpoint(session_id: str, authorization: str | Non
return {"success": True}
@app.get("/api/v1/organizations/{organization_id}/iso-builds")
def list_iso_builds(organization_id: str, authorization: str | None = Header(default=None)):
if not require_service_token(authorization):
return {"success": False, "error": "unauthorized", "message": "Fehlendes oder ungültiges Service-Token."}
return {"success": True, "iso_builds": fetch_iso_builds(organization_id)}
@app.post("/api/v1/organizations/{organization_id}/iso-builds")
def create_iso_build_endpoint(
organization_id: str,
payload: CreateIsoBuildRequest,
authorization: str | None = Header(default=None),
):
if not require_service_token(authorization):
return {"success": False, "error": "unauthorized", "message": "Fehlendes oder ungültiges Service-Token."}
if not payload.activation_code.strip():
return {"success": False, "error": "invalid_request", "message": "activation_code darf nicht leer sein."}
ISO_BUILD_OUTPUT_DIR.mkdir(parents=True, exist_ok=True)
iso_build = create_iso_build(organization_id)
thread = threading.Thread(
target=run_iso_build,
args=(UUID(iso_build["id"]), organization_id, payload.activation_code, payload.wifi_ssid, payload.wifi_psk),
daemon=True,
)
thread.start()
return {"success": True, "iso_build": iso_build}
@app.get("/api/v1/iso-builds/{build_id}")
def get_iso_build_endpoint(build_id: str, authorization: str | None = Header(default=None)):
if not require_service_token(authorization):
return {"success": False, "error": "unauthorized", "message": "Fehlendes oder ungültiges Service-Token."}
iso_build = fetch_iso_build(build_id)
if iso_build is None:
return {"success": False, "error": "not_found", "message": "ISO-Build wurde nicht gefunden."}
return {"success": True, "iso_build": iso_build}
@app.get("/api/v1/iso-builds/{build_id}/download")
def download_iso_build(build_id: str, authorization: str | None = Header(default=None)):
if not require_service_token(authorization):
return {"success": False, "error": "unauthorized", "message": "Fehlendes oder ungültiges Service-Token."}
iso_build = fetch_iso_build(build_id)
if iso_build is None:
return {"success": False, "error": "not_found", "message": "ISO-Build wurde nicht gefunden."}
if iso_build["status"] != "completed" or not iso_build["output_filename"]:
return {"success": False, "error": "not_ready", "message": "Der ISO-Build ist noch nicht abgeschlossen."}
output_path = ISO_BUILD_OUTPUT_DIR / iso_build["output_filename"]
if not output_path.is_file():
return {"success": False, "error": "file_missing", "message": "Die ISO-Datei wurde nicht gefunden."}
return FileResponse(
output_path,
media_type="application/octet-stream",
filename="tuxflotte-installationsmedium.iso",
)
@app.get("/api/v1/devices/{device_id}/auftragskatalog")
def get_device_auftragskatalog(
device_id: str,

View File

@ -0,0 +1,27 @@
BEGIN;
-- Phase 4 des Self-Service-ISO-Plans: Bautracking fuer personalisierte
-- Kunden-ISOs (siehe tuxflotte-installer/scripts/build_customer_iso.sh).
-- Der Bau selbst laeuft als Hintergrund-Thread im provisioning-server
-- (siehe run_iso_build() in app.py) - diese Tabelle ist der Status-
-- Speicher dafuer, den der Trigger-/Status-/Download-Endpoint gemeinsam
-- nutzen. WLAN-Zugangsdaten werden bewusst NICHT hier gespeichert (nur
-- transient waehrend des Baus verwendet, siehe app.py) - persistente,
-- verschluesselte Ablage ist Sache der Kundenplattform (Phase 5).
CREATE TABLE iso_builds (
id UUID PRIMARY KEY,
organization_id UUID NOT NULL
REFERENCES organizations(id)
ON DELETE CASCADE,
status TEXT NOT NULL DEFAULT 'pending'
CHECK (status IN ('pending', 'running', 'completed', 'failed')),
output_filename TEXT,
error_message TEXT,
created_at TIMESTAMPTZ NOT NULL DEFAULT CURRENT_TIMESTAMP,
started_at TIMESTAMPTZ,
finished_at TIMESTAMPTZ
);
CREATE INDEX iso_builds_organization_id_idx ON iso_builds (organization_id);
COMMIT;