Skip to content
Register

What do you use for hardware monitoring that isn't Nagios from 2008?

Dedicated Servers by blogfranck 24 replies 2.5K views
blogfranck

Member OP

2 #1

What do you use for hardware monitoring that is not Nagios from 2008?

I have maintained the bare metal fleet for six years. The Nagios configuration grows like vine, untouchable. I seek something modern, but Prometheus ecosystem seems so heavy for single server.

What you use for hosting at small scale? I have ten nodes, not one hundred. Grafana is beautiful but maybe too much for my humble needs. I wish to know sensor temperatures, disk health, memory pressure. Simple alerts, not symphony orchestra.

I await your wisdom.

bellaauc

Member

3-2-1 Believer

#2

Franck, have you tested your restore process for those Nagios configs? The 3-2-1 rule applies to monitoring state too—export before you migrate.

For modern bare metal, I have seen good results with:

  • Prometheus node_exporter (yes, even at small scale)
  • Grafana with a single dashboard
  • Alertmanager to PagerDuty or email
  • BUT: did you test your restore for the new stack too?

The node_exporter footprint is under 30MB RAM. Not heavy at all. Start there, add IPMI later if you need out-of-band sensor data.

3-2-1 or you're already dead
YuriDavid

Member

#3

---
- Nagios alternatives:
- Prometheus:
- node_exporter:
- RAM: ~30MB
- NOTE: not actually heavy at small scale
- alertmanager:
- complexity: medium
- lighter options:
- WARNING: often lack IPMI integration
- WARNING: sensor naming nonstandard across vendors
- IPMI problem:
- 200+ sensor names possible per vendor
- standard exporters: fail on unknown sensors
- custom forks: exist
- @blogfranck:
- recommendation:
- start: node_exporter + Grafana
- iterate: add IPMI when needed
- test: restore process at each stage
---

indentation is not optional
Lag

Member

#4

The IPMI is how to say... the hard part, no?

I have made custom fork of ipmi_exporter, ok, it handles 200+ sensor names, so, the SuperMicro and Dell have different dictionaries, you know!

I publish on my hosting, you can look. The ticket said: "sensor Temp_1 is not found" — ok, now it is found, there!

The code is not pretty, the server is broken sometimes, but the mapping table it grows. 47 vendors now. So, you want I add yours?

it's not lag, it's dramatic tension
lucgone

Member

Le Baguette

#5

47 vendors mapped manually? That sounds like pain

Vive la résistance... électrique
DH

Member

#6

Which kernel, franck? I had sensor issues on 5.4

nuyekigor

Member

#7

Grafana is overkill for ten nodes

blogfranck

Member OP

#8

Kernel 6.1 on most, one still on 5.15. The sensors show in lm-sensors, so kernel is not my blocker.

The blocker is: Nagios config for checks grows by copy-paste. I want to delete it, but I fear what I break.

Lag

Member

#9

The fork is on my Gitea, not GitHub, the connection is slow today. But you want, I send link in PM.

The mapping is JSON now, not hardcode. Before it was pain, yes. Now community send pull request with their SuperMicro X9 or Dell R720, I merge. 47 vendors, maybe 30 from other people.

it's not lag, it's dramatic tension
DH

Member

#10

5.4 had the it87 driver problem, missing chips. 5.10+ better for NCT6775 and similar.

Post a reply

You need an account to reply. Log in or register to join the conversation.

Post reply Preview Save draft