Member OP
I am buy yesterday a monitor tool from Vultr and is terible lah. Alert for eferything. CPU 1% up xixixi alert. Disk read normal xixixi alert. My phone never stop. I want someting smart. Not ML hype just... not stupid. Very tired. Pirst world problem I know but I pay so I want good. What you use that not cry wolf efery minute?
My old script was just check check efery 5min and that was better than this "AI" garbage. Xixixi
Member
Ow to say... I am using the CloudCone monitoring and is... how to say... acceptable. The «intelligent alerting» is not so intelligent. I drop the h on purpose sometimes, is habit. But the real solution I pind is very simple: hysteresis. You set threshold at 80% for alert ON, 60% for alert OFF. No flap. No spam. The Machine Learning is just... Oui, I say oui, is overkill for most.
Random Words because why not.
I've tested twelve monitoring solutions in the last three years. Uptime percentages: 99.97% average across my fleet. Alert fatigue is real. The ML-based ones promise the world and deliver 40% false positive rates in my experience. Status pages become meaningless when everything is "degraded."
What actually works: hysteresis, as mentioned. Also: require two consecutive failures before alerting. Exponential backoff on flapping checks. The simple math outperforms neural nets for infrastructure blips.
To be precise: many of these services process your metrics in jurisdictions with inadequate protections. The hysteresis approach has another advantage—it runs locally, no data leaves your network.
I have used this configuration for three years without alert fatigue. It is deterministic, auditable, and privacy-preserving. The German approach, you might say.
[code]ALERT_CPU_HIGH=80
ALERT_CPU_CLEAR=60
REQUIRED_FAILS=2
FLAP_BACKOFF_SEC=300
# Only alert if:
# - metric > ALERT_CPU_HIGH
# - AND stays there for REQUIRED_FAILS checks
# - AND was previously below ALERT_CPU_CLEAR
Member
This is very good!!! I am make a test with this config!!! The ML solutions are very expensive and alert for eferything!!! My provider is InterServer and they have basic monitor but I will add this hysteresis in my local script!!! Is very smart!!! Thank you @Fritz48!!!
Network okay. Monitoring worse. Perkele. Hysteresis fix. 80/60. Done. No more 3am wake. Simple. Why vendors not do this. Greed. Perkele.
