Customer has ~80–90 clusters and are evaluating a centralized Harvest + Prometheus + Grafana architecture.
Trying to avoid a per-cluster deployment model if possible.
Trying to understand:
• Practical limits per Harvest instance (clusters, polling load, etc.)
• Whether it’s better to scale vertically (single instance) or horizontally (multiple pollers / instances)
• Recommended patterns for distributing pollers (per region, per Prometheus target group, etc.)
• Any known bottlenecks ?
#Harvest Scaling
magic mango
marble river
hi @magic mango others may chime in also but I'd add a few points. Harvest is very flexible in deployment options. I would install Prometheus/Grafana separately from the Harvest pollers. The pollers scale well both vertically and horizontally. On average, you need 100-200 MB per poller, but this depends on which collectors you enable, number of ONTAP objects collected, etc.
VictoriaMetrics (VM) uses less memory than Prometheus so VM may be a better fit at high scale. You can also use Harvest's VictoriaMerics exporter to push the metrics to VM. The advantage that has over the PrometheusExporter is the PrometheusExporter needs to keep the metrics in-memory waiting for Prometheus to scrape them. The VM exporter can directly push them
In talking with other customers, 80-90 clusters is no big deal. There are quite a few customers monitoring that many and more.
magic mango
Thanks Chris!! I sent along the info and will try and read up and look for a lab on demand. If a customer really needed expertise and had PS hours, would NetApp engage for consulting?
marble river
yes, PS has helped customers with Harvest in the past
craggy sable
Not with deployment, but I have done several PS hours training sessions and Custom kiosk dashboards.
pseudo flare
I'm going to leech off this chat because I have s imilar type of question. If we want more features of what a license copy of Grafana offers, particularly customer dashboards and being able to email those custom dashboards. Is it easiest to just leave Nabox in place and stand up a separate Grafana only, then point it at the NABox metrics datasource?Would this provide separate dashboards/users without duplicating Harvest collection?
orchid salmon
That is right @pseudo flare . You can connect an external Grafana with the NABox Victoria Metrics database.
pseudo flare
Is there a write up for that on the Nabox documentation? If not, I can find some on Google I’m sure.
orchid salmon
@wild phoenix
wild phoenix
Depending on the version you can either enable guest access to Victoria metrics or generate a token for the Victoria metrics scope
The endpoint would be https://yournabox/vm/
pseudo flare
Got it, thanks!