Hey everyone, just saw the news about New Relic's new pricing. Switching from per-host to data-ingestion based billing seems to be the trend now, right?
As someone building my first real Grafana/Prometheus stack at work, I'm trying to understand the practical impact. For a small startup with maybe 50 containers and a bunch of custom metrics, would this model actually be more predictable, or does it just shift the worry to cardinality and sampling? 😅
My naive Prometheus config for a service looks like this:
```yaml
scrape_configs:
- job_name: 'node_exporter'
static_configs:
- targets: ['localhost:9100']
```
With this change, I'm wondering if tools like this, where I control the scrape interval and what I collect, become even more attractive for cost control. Any thoughts from those who've compared bills?
Good question. We made this exact switch at my last place, moving from per-host to ingestion-based billing with a different vendor. It can be more predictable, but only if you enforce strict discipline on cardinality. That "bunch of custom metrics" is the wild card.
Your instinct about controlling the scrape config is spot-on. With per-host billing, you might be tempted to scrape everything "just in case." Now, every high-cardinality label you add (like user_id) has a direct, measurable cost. You'll start putting more thought into aggregations and sampling in your exporters.
The config you shared is fine for node_exporter, but for your apps, you'll want to design your metric labels very carefully. I started using Prometheus recording rules to pre-aggregate expensive queries before they ever left the cluster.
Cloud cost nerd. No, I don't use Reserved Instances.
Yes, it becomes way more attractive for cost control. But the real value isn't just in controlling the scrape interval. It's that you're forced to think about your metric *design* before you instrument anything. With ingestion billing, slapping a `user_id` label on every metric is a rookie mistake that will blow up your bill.
Your shift from "per-host" to "cardinality and sampling" worry is exactly right. That's the trade-off. Predictability comes from your team's discipline, not the vendor's model.
For a small startup, I'd stick with your Prometheus stack and use New Relic's free tier for the one or two things you actually need 24/7 eyes on. The moment you start scaling, your custom metrics will make their pricing model scream.