Why does Steam go down every Tuesday?
As most gamers will know, Steam cloud services routinely go down every Tuesday for maintenance, meaning any games that use Steam matchmaking stop working.
My question is: why? In this day and age we have ways of maintaining services with zero downtime using A/B deployments for example. Even running services that can't be load balanced at layer 7 can be gradually drained so as to not interrupt connections.
Any Valve employees who can shed light on this?
I'm not a Valve employee but I can imagine a rolling update would not be easy considering all these games are constantly connected to the various game servers. They would reconnect immediatelly upon losing the connection and could reach an existing game server running the old software version. Then the rolling upgrade reaches that game server and once again the user is disconnected. And so on.
Blue/green deployments would be better but I'm guessing the scale at which they operate is massive. Back in the old days I managed a fleet of Counter-Strike servers and they were super heavy on CPU/network. Spinning up a complete infrastructure to switch all gamers at once would be too costly, even for Valve.
I doubt anyone at Valve will be at liberty of discussing the specifics you're asking about, so these are my $0.02
Because Valve doesn't have an insane amount of employees and since it is not critical software, the downtime is fine.
I don't know for sure, but I'd speculate that it is more about customer service than technical capability. While it seems like a bad experience to expect downtime every week, it would be worse to be told "No more Tuesday maintenance!", and then have it end up happening anyway because something went wrong. And things do go wrong in this industry. Leaving the maintenance window open gives them more flexibility to get a variety of work done without having to worry about ticking off the gamers because of changing expectations.
Obviously we know what you mean and this is a bit of pedantry, but I think blue-green deployment is the term to use.
Because the janitor keeps plugging the server out to plug in his vacuum.
Unrelated, but another company that used to (still does?) the same thing was Konami with PES from 2017-2020.