Sticky Loadbalancing

[jcbjoe]

Hi!

Is there anyway to use cookies to specify which worker to route specific requests to? The reason this is important for us is because we store state in a local dictionary which is much faster to access than going to Redis or Mongo.

Currently, I'm running 4 separate instances of Granian and use HAProxy for sticky routing. However some of the Granian processes seem to be running on the same cores.

Thanks!
Joe

[gi0baro]

Is there anyway to use cookies to specify which worker to route specific requests to? The reason this is important for us is because we store state in a local dictionary which is much faster to access than going to Redis or Mongo.

Currently, I'm running 4 separate instances of Granian and use HAProxy for sticky routing.

No, there's currently no support for that. Using a proxy server like HAProxy or similar is the correct way to go for this.

However some of the Granian processes seem to be running on the same cores.

Granian runs 2 different loops (at least in ASGI and RSGI protocols), so you might want to run NCpu/2 workers.

[jcbjoe]

Hi @gi0baro,

Thanks for the update, Is there any plans to implement sticky load balancing?

Also is there any downsides to running 4 separate instances of Granian on different ports?

Thanks,
Joe

[gi0baro]

Is there any plans to implement sticky load balancing?

No, it feels a bit out of context for the project at the moment.

Also is there any downsides to running 4 separate instances of Granian on different ports?

Not that I'm aware of.