High storage cost due to too many IOpS in Azure

[GaugeTheory]

Hello everybody,
I have started use RabbitMQ operator in Azure AKS cluster to deliver my messages. I have also enabled persistent storage to keep my messages in a PVC for data redundancy. In my previous setup I was using bitnami RabbitMQ without operator and without PVC

After a month of usage, I realised 20GB PVC for the RabbitMQ pods cost too much. Disk size costs few Euros/month, but disk operations costs ~100Euros/month. I suspect this is due to IOpS to disk. I see that RabbitMQ PVCs get ~200 IOpS. This is also Message rate I get per second. That means every single messages get written into disk.

My question is more generic one. Is there a way to reduce IOpS to disk? I don't need to write every messages to disk at all, it just needs to write to disk when the messages in memory reaches some size (i.e.1GB) for disaster recovery. I want nothing to be written into disk as long as size of the messages in memory is below a certain threshold.

I have also tried quorum queues, which has 512MB WAL size by default. I have hoped this would reduce IOpS since WAL would never reach 512 MB in normal operations. Still I see similar IOpS as before

https://blog.rabbitmq.com/posts/2020/04/quorum-queues-and-why-disks-matter/

If there is no option to reduce IOpS, i think I will have to get rid of PVCs for RabbitMQs. Is it also possible to disable PVC with RabbitMQ operator?

Or maybe there is a better StorageClass option that fits better with high IOpS cost wise in Azure? I could also pay a little for disk operations but definitely not 100x the disk cost. Or maybe local disk of the VM?

Thanks for any help

[mkuratczyk]

If you are ok with just not having a PVC, you can set storage size to 0: https://www.rabbitmq.com/kubernetes/operator/using-operator.html#persistence