The same NetFlow v9 export into Akvorado and into GoFlow2 with Loki: byte totals and record counts matched exactly, while memory was about 2.4 GiB against 146 MiB. Where the light stack runs out: ifIndex-only filters, no classification, slow long ranges.
From "the WAN is slow" to one IP in six steps with Akvorado and Cisco IOS XE NetFlow v9. Along the way: why a 28.75 Mbps peak was really about 5 Mbps, and why querying both routers counted 858 MB of ICMP that was really 429 MB.
NetFlow v9 from two Cisco IOS XE 17.18.2 routers into Akvorado with SNMPv3 enrichment, classifiers, Sankey, and top talkers. ClickHouse died with exit 132 on a CPU without AVX2 until pinned to 26.3, and the router image had no cache timeout command.
Turning a Grafana 13 NOC dashboard into a 1920x1080 wall screen: five large stat tiles with solid backgrounds, a problems list filtered to severity 2 and up, and basic bar gauges, because gradient mode colored every bar red. A real HTTP outage showed 93.2% in red.
One Grafana 13 dashboard with Zabbix problems, Prometheus SNMP and probe metrics, and Loki syslog. A real fault tested it: both pollers lost acc-sw1 at once, which put the problem on the switch. Also: why a Zabbix problem count tile fails, and filtering probe noise.
Prometheus blackbox exporter 0.29.1 probing Cisco IOS XE routers, a switch, and a web server with ICMP, SSH banner, and HTTP checks in Grafana. A two-minute web outage dropped HTTP availability to 84.48% while SNMP and ICMP still looked fine.
Grafana 13 dashboards moved into Git with file provisioning and generated per device by an Ansible Jinja2 template. One template fix to the SNMP job match and rate window repaired every device dashboard. Gotchas: escaping {{ifName}}, fileglob, silent JSON failures.
node_exporter 1.12.1 on an Ubuntu 24.04 server, scraped by the same Prometheus as the Cisco routers. PromQL for CPU, load against core count, MemAvailable and disk, plus a textfile-collector metric for hours since the last successful backup.
Flexible NetFlow from Cisco IOS XE 17.18.2 as IPFIX into GoFlow2, Alloy and Loki, with LogQL top talkers in Grafana 13. 41 flow records covering 1.92 GB in 25 minutes, why flow bars spike with a 30-minute active timeout, and the query step that stops double counting.
Grafana 13 alerting for Cisco IOS XE WAN failures: two multi-dimensional rules, a policy tree grouped by router, a mute timing, a ticket silence, and an email template that fixes resolved alerts reporting "0 BGP peers". Tested by shutting a WAN link twice.
Telegraf 1.40.3 against Cisco IOS XE 17.18.2: the image has no gNMI server, so the subscription got connection refused. 10-second SNMP through Telegraf was the fallback, and it caught a 42 Mb/s burst that 60-second polling showed as 14 Mb/s.
IOS XE 17.18.2 syslog into Loki 3.7.9 through Alloy 1.21.3. Alloy rejected Cisco RFC 3164 with "expecting a sequence", so the routers send RFC 5424. LogQL for flaps, config changes with usernames, and OSPF and BGP events, next to Prometheus metrics.