The day starts with still remembering where all the rooms are, but too tired to get up for the first session. Oh well.
TL;DR, pack your columns by data size, large to small, don't use legacy types like currency and json.
For now, run postgres_exporter, but caveats abound, and postgres metrics were designed for an era of human DBAs running queries. Querying for metrics over SQL means... you contend with the main application for resources. There may be ongoing work to expose metrics from a native background process over prometheus format.
There's a shiny toy, for monitoring high volumes of traffic, and you can script in lua for fancy custom rules.
Acronyms I didn't quite understand, mostly reserve capacity if you want guaranteed bandwidth.
So you want to hole punch through 2 way NAT with TCP? libp2p have a protocol called [DCUtR]: probe the other endpoint through a relay, half the RTT and send out connections from both side simultaneously. There's a 70% success rate. UDP is better/faster.
CNIs are are now always configured in lists, and each takes config and the results of the previous output, and should output results to be passed through to the next in chain.
Another talk by Kris Nova. More or less an in person version of the Hachyderm outage postmortem. Plus some news about future governance. Will democracy work? Nivenly is now a foundation to be a steward of hachyderm and aurae, so I guess we'll see.
Fancy eBPF tooling that understands containers (vs the base tools from BCC / iovisor). Integrates with Kubernetes.
Namespace your metrics with service/component names for clarity, follow OpenMetrics reccommendations on type/suffic naming.