Self-evolving Context Database for AI Agents. Unify Agent Memory, Knowledge RAG and Skills.
Python agent-contextagent-memoryagent-pluginsagentic-ragcontext-databasedsh-pluginknowledge-basememorymemory-managementself-evolvingvector-database
"""Pending health server for the OpenViking Docker entrypoint.
While the container is waiting for ``ov.conf`` to appear, the entrypoint
runs this tiny HTTP server on the same port the real OpenViking server
will bind. It answers *every* request — `/`, `/health`, anything — with
the same 503 JSON payload describing what's wrong and how to fix it, so
operators and agents probing the container can self-discover the issue
without having to read ``docker logs``.
This is intentionally undocumented in the user guide: it's a 防呆 (poka-yoke
/ fool-proofing) fallback that only runs when something has already gone
wrong (no ov.conf and no OPENVIKING_CONF_CONTENT). The happy path never
sees it, and pointing users at it would imply it's a deployment mode
worth depending on rather than a safety net. The response body itself is
self-explanatory when someone does hit it.
"""
from __future__ import annotations
import http.server
import json
import os
import socketserver
import sys
_DEFAULT_HOST = "0.0.0.0"
_DEFAULT_PORT = 1933
_DEFAULT_CONFIG_FILE = "/app/.openviking/ov.conf"
def build_payload(config_file: str) -> dict:
"""Return the JSON body served on every route."""
return {
"status": "pending_initialization",
"error": "OpenViking config file not found",
"config_file": config_file,
"fix": [
"mount ~/.openviking on the host to /app/.openviking",
"set OPENVIKING_CONF_CONTENT to the full ov.conf JSON",
"docker exec into this container and run: openviking-server init",
],
}
def make_handler(config_file: str) -> type[http.server.BaseHTTPRequestHandler]:
body = (json.dumps(build_payload(config_file), ensure_ascii=False, indent=2) + "\n").encode(
"utf-8"
)
class Handler(http.server.BaseHTTPRequestHandler):
def _respond(self) -> None:
self.send_response(503)
self.send_header("Content-Type", "application/json; charset=utf-8")
self.send_header("Content-Length", str(len(body)))
self.send_header("Cache-Control", "no-store")
self.end_headers()
if self.command != "HEAD":
self.wfile.write(body)
do_GET = _respond
do_POST = _respond
do_PUT = _respond
do_PATCH = _respond
do_DELETE = _respond
do_HEAD = _respond
do_OPTIONS = _respond
def log_message(self, fmt: str, *args) -> None:
sys.stdout.write("[openviking-pending-health] " + (fmt % args) + "\n")
sys.stdout.flush()
return Handler
class _ReusableServer(socketserver.TCPServer):
allow_reuse_address = True
def serve(host: str, port: int, config_file: str) -> None:
handler = make_handler(config_file)
with _ReusableServer((host, port), handler) as httpd:
sys.stdout.write(
f"[openviking-pending-health] serving on {host}:{port}; waiting for {config_file}\n"
)
sys.stdout.flush()
httpd.serve_forever()
def main() -> int:
host = os.environ.get("OPENVIKING_PENDING_HOST", _DEFAULT_HOST)
port = int(os.environ.get("OPENVIKING_PENDING_PORT", str(_DEFAULT_PORT)))
config_file = os.environ.get("OPENVIKING_CONFIG_FILE", _DEFAULT_CONFIG_FILE)
try:
serve(host, port, config_file)
except KeyboardInterrupt:
return 0
return 0
if __name__ == "__main__":
raise SystemExit(main())