Pavel Khusainov

alvelcom @ GitHub Twitter

Using the moving average and exponential smoothing to compute typical daily profile graph in prometheus.

Prometheus alerts can be flopping sometimes, that is really annoying when you get multiple notifications at 3 AM. Or when you manually assign an incident to someone else and with a new alert you need to assign this person again. This post looks into two ways how this situation can be improved.

The study compares different network modes of CoreOS's flannel. It shows that most solutions perform somewhat worse than the bare host network, IPVLan performed slightly worse, and finally VXLAN that performed considerably worse.