Subscriptions: Can a minimum-once guarantee be provided in the face of e.g. network problems and downtime?

[pmorch]

We're considering putting graphql-mesh in front of RabbitMQ for $manyReasons.

Basically, if we just have

RabbitMQ <-- Consumer

then RabbitMQ provides a minimum-once guarantee. If the subscriber is down for a (short) while, e.g. due to network problems, maintenance, upgrades, you name it, any events that occur during downtime will be waiting for them once they re-subscribe (maxiumum queue length permitting).

If we instead have

RabbitMQ <-- graphql-mesh <-- Consumer

the minimum-once guarantee will hold for the left-most connection. But if the GraphQL consumer experiences e.g. network problems, will any events missed during inevitable downtime be waiting when they re-subscribe?

Because if events are lost during connection downtime, both ends need to implement costly re-sync mechanisms if this is be used in real server-to-server implementations.

If it perhaps isn't provided out-of-the-box, what options do we have for providing such an end-to-end minimum-once guarantee?

I'm aware that this question perhaps reveals some basic misunderstanding of GraphQL and its possibilities.