neat spade
hi,
when we setup an instance to have 2 replicas, how do we get the resource related to 1 of the rep, hm I can't find the info from the docs though
for a simple scenario, we have a /get-users to returns a list of in-memory users array in one instance, and /add-user to add an entry to the array
so hitting the add endpoint one time, and get just returns the values of one random replica, either an empty array, or array of 1 val added previously
mild sigilBOT
prime oceanBOT
rustic heart
You’ll want to use a database if you’re running multiple replicas of a service.
odd beacon
that's 2 for 2 on the use a database solution
rustic heart
If you’re expecting to be able to hit a certain replica you’re misunderstanding what replicas are.
neat spade
ok,
yea we are using sse, we store the response obj from client for sending the event-stream over
I can't find the info yet but I think we can't save that into a db...
but thanks for the info though
odd beacon
I also think it's worth asking, are you sure you need any replicas?
rustic heart
Sounds like a good use case for redis.
Redis pubsub with channels based on the users connection ID would likely be fine for this.
neat spade
yea, i can't run any benchmark right now, but I just need to have a rep ready for there are many connections,
we used short polling previously and on peak we had 5-6k ppl concurrently
odd beacon
are you a part of a railway team?
rustic heart
Even with that amount why do you think you need a replica? I’ve had multiple projects with 50k clients on a single instance.
Replicas are for scaling load. If your single service can keep up with the load a replica isn’t gonna really do much but add complexity.
neat spade
oh, is it on a normal 8gb ram instance for 50k clients,
I can hop to a team but was just using the hobby plan initially
i meant i can turn to the team eventually
rustic heart
Depends on the load. In the case of the 50k clients we were using a 1GB digital ocean box.
People way too often see a big number and instantly think they need to scale without actually seeing if they’re hitting any kind of limit.
If you’re doing this all in memory and move to using a database you may find you don’t need the replica after all since the memory footprint could very likely drop down to almost nothing.
neat spade
i see, thanks for info. let me try the benchmark for now
ah i think the **res **we store in memory is a stream object, i remember we were having issue serialize the object on db query, to use that obj to stream the info back from nodejs => client for server sent event
really appreciate
rustic heart
What’re you sending to the client that’s a stream?
If it’s a file then that’s not memory intensive. If it’s something from another client then look at pub/sub.
neat spade
ah ok
rustic heart
just having a client connected uses very few resources, copying a stream to a connection is also really cheap in terms of resources so unless you're doing a bunch of computational work you'll likely see a big decrease in memory usage once this is all moved to a db.
neat spade
ok yea its not having any big computation, we have other servers already doing the math, we just have this server alone to send / notify data back to client when data is published from a redis when a channel / topic matches