Feature: Load balancer could not distinguish outdated supervisors

The load balancer could not check if a supervisor is running the latest release using the status of a GET response.

This is useful for the healthcheck of the Traefik Proxy to ensure that backend CRNs are running the latest version.

packaging/aleph-vm/DEBIAN/control

@@ -3,6 +3,6 @@ Version: 0.1.8
 Architecture: all
 Maintainer: Aleph.im
 Description: Aleph.im VM execution engine
-Depends: python3,python3-pip,python3-aiohttp,python3-msgpack,python3-aiodns,python3-sqlalchemy,python3-setproctitle,redis,python3-aioredis,python3-psutil,sudo,acl,curl,systemd-container,squashfs-tools,debootstrap
+Depends: python3,python3-pip,python3-aiohttp,python3-msgpack,python3-aiodns,python3-sqlalchemy,python3-setproctitle,redis,python3-aioredis,python3-psutil,sudo,acl,curl,systemd-container,squashfs-tools,debootstrap,python3-packaging
 Section: aleph-im
 Priority: Extra

vm_supervisor/supervisor.py

@@ -23,7 +23,7 @@
     about_executions,
     about_config,
     status_check_fastapi,
-    about_execution_records,
+    about_execution_records, status_check_version,
 )
 
 logger = logging.getLogger(__name__)
@@ -51,6 +51,7 @@ async def server_version_middleware(
         web.get("/about/executions/records", about_execution_records),
         web.get("/about/config", about_config),
         web.get("/status/check/fastapi", status_check_fastapi),
+        web.get("/status/check/version", status_check_version),
         web.route("*", "/vm/{ref}{suffix:.*}", run_code_from_path),
         web.route("*", "/{suffix:.*}", run_code_from_hostname),
     ]

vm_supervisor/views.py

@@ -2,7 +2,8 @@
 import logging
 import os.path
 from string import Template
-from typing import Awaitable
+from typing import Awaitable, Optional
+from packaging.version import Version, InvalidVersion
 
 import aiodns
 import aiohttp
@@ -141,3 +142,26 @@ async def status_check_fastapi(request: web.Request):
             "persistent_storage": await status.check_persistent_storage(session),
         }
         return web.json_response(result, status=200 if all(result.values()) else 503)
+
+
+async def status_check_version(request: web.Request):
+    """Check if the software is running a version equal or newer than the given one"""
+    reference_str: Optional[str] = request.query.get("reference")
+    if not reference_str:
+        raise web.HTTPBadRequest(text="Query field '?reference=` must be specified")
+    try:
+        reference = Version(reference_str)
+    except InvalidVersion as error:
+        raise web.HTTPBadRequest(text=error.args[0])
+
+    try:
+        current = Version(get_version_from_git())
+    except InvalidVersion as error:
+        raise web.HTTPServiceUnavailable(text=error.args[0])
+
+    if current >= reference:
+        return web.Response(
+            status=200, text=f"Up-to-date: version {current} >= {reference}"
+        )
+    else:
+        return web.HTTPForbidden(text=f"Outdated: version {current} < {reference}")