🔀New traffic redirection
Usually, on horizontal scalable models, you should know when to scale up or down. Mostly, in Soketi you should do it based on:
memory - having low memory can be tough for new sockets as it's needed to store new user data on each new connection and actually having memory to store the pointer to the connection
CPU - having low remaining CPU can increase the latency of your connections and mesages
For instances that run in non-worker mode, you can check if they can receive new connections according to a threshold:
This threshold can be tweaked by specifying the HTTP_ACCEPT_TRAFFIC_MEMORY_THRESHOLD, which is the percent of memory at which the /accept-traffic request will return a non-200 OK response. This way, you can configure your infrastructure in such a way that you can route the new traffic to instances that return 200 OK instead of pressuring a random algorithm that will not ensure best available memory instances first.
Alternatively, for Kubernetes there is Network Watcher, which is a sidecar container that will automatically handle the pod labels to redirect traffic to pods that are having enough memory. Network Watcher is not influenced by this particular endpoint
Last updated