1
0
mirror of synced 2024-11-24 06:20:10 +01:00
artemis/docs/migrating.md

4.4 KiB

Migrating from an older build of ARTEMiS

If you haven't updated artemis in a while, you may find that configuration options have moved, been renamed, or no longer exist. This document exists to help migrate from legacy versions of artemis to newer builds.

Dependancies

Make sure your dependiences are up to date with what's required to run artemis. A simple pip install -r requirements.txt will get you up to date.

Database

Database migration is required if you are using a version of artemis that still uses the old custom-rolled database versioning system (raw SQL scripts). Artemis now uses alembic to manage database versioning, and you will need to move to this new system.

BEFORE DOING ANY DATABASE WORK, ALWAYS MAKE SURE YOU HAVE FUNCTIONAL, UP-TO-DATE BACKUPS!!

For almost all situations, simply running python dbutils.py migrate will do the job. This will upgrade you to the latest version of the old system, move you over to alembic, then upgrade you to the newest alembic version. If you encounter any errors or data loss, you should report this as a bug to our issue tracker.

Configuration

Configuration management is the sewage cleaning of the sysadmin world. It sucks and nobody likes to do it, but it needs to be done or everyone ends up in deep shit. This section will walk through what configuration options have changed, and how to set them properly.

core.yaml

title->hostname is now server->hostname. This hostname is what gets sent to clients in response to auth requests, so it should be both accessable from whereever the client is, and point properly to the title server.

With the move to starlette and uvicorn, different services now run as seperate USGI applications. billing->standalone and allnet->standalone are flags that determine weather the service runs as a stand-alone service, on it's own seperate port, or as a part of the whole application. For example, setting billing->standalone to True will cause a seperate instance of the billing server to spin up listening on 8443 with SSL using the certs listed in the config file. Setting it to False will just allow the main server to also serve /request/ and assumes that something is standing in front of it proxying 8443 SSL to whatever server->port is set to.

Beforehand, if server->is_develop was False, the server assumed that there was a proxy standing in front of it, proxying requests to proper channels. This was, in hindsight, a very dumb assumption. Now, server->is_using_proxy is what flags the server as having nginx or another proxy in front of it. The effects of setting this to true are somewhat game-dependant, but generally artemis will use the port listed in server->proxy_port (and server->proxy_port_ssl for SSL connections, as defined by the games) instead of server->port. If set to 0, server->proxy_port will default to what server->port (and server->proxy_port_ssl will default to 443) make sure to set them accordingly. Note that some title servers have their own needs and specify their own specific ports. Refer to game_specific_info.md for more infomation. (For example, pokken requires SSL using an older, weaker certificate, and always requires the port to be sent even if it's port 443)

index.py's args have changed. You can now override what port the title server listens on with -p and tell the server to use ssl with -s.

Rather then having a standalone config variable, the frontend is a seperate wsgi app entirely. Having enable be True will launch it on the port specified in the config file. Otherwise, the fontend will not run.

title->reboot_start_time/reboot_end_time allow you to specify when the games should be told network maintanence is happening. It's exact implementation depends on the game. Do note that many games will beave unexpectly if reboot_end_time is not 07:00.

If you wish to make use of aimedb's SegaAuthId system to better protect the few title servers that actually use it, set aimedb->id_secret to base64-encoded random bytes (32 is a good length) using something like openssl rand -base64 64. If you intend to use the frontend, the same thing must be done for frontend->secret or you won't be able to log in.

mucha's only option is now just log level.

aimedb now has it's own listen_address field, in case you want to proxy everything but aimedb, so it can still listen on 0.0.0.0 instead of 127.0.0.1.