Hello everyone.
Quick question
Let’s say that our production Atlas Server is M50 with its own replica set of multiple electable nodes.
We are going to start playing with an analytic node but for now, we want to start with the lowest tier possible and go up if necessary. In this case, the lowest tier we can choose in production is M20.
Do we need to worry about anything if we use a low-performance node?
Any effects on the primary node or replication lag?
Here is my non-scientific take on your issue.
An a-node (analytic node) replicates all the data. It has to be able to handle at least the same write load as the other rs-nodes (replica set nodes).
I want to argue that it needs more.
One advantage of an a-node is the ability to have extra indices tailored to your AUC (analytic use-cases). These extra indices have to be updated when writes happen. This means that your a-node might needs to handle an higher load than rs-nodes during writes. Extra indices also means more disk space to store them. Extra indices might also require more RAM otherwise the indices have to be read from disk.
If any AUC happens during heavy write normal traffic use-case, the a-node will need more CPU than your other rs-nodes to keep up with replication and your AUC.
So, my gut feeling is that an a-node requires more resources than rs-nodes. Except may be if your normal use-cases are mostly reads.
You could start with smaller and enable auto-scaling and see.
Please share your findings, especially if they differ from my gut feeling. I like when my gut feeling errs me as it justify the extra work needed to have the facts right.