💪Exercise💪 — introduction round: what you run, and what it does not answer
Opening round, before any theory. Everyone answers the same five questions about their own setup. Two minutes per person, no slides, no preparation.
Goal
Name one question about your own system you cannot answer today. That sentence is what the rest of the course maps onto.
Prerequisites
None. This is about your own environment, not the workshop cluster.
💪Exercise💪 — the round
Your turn, five questions, in this order:
| # | Question | Comment |
|---|---|---|
| 1 | What do you have? | Systems running right now, by name: Zabbix, ELK, Datadog, nothing for traces. Not “we have monitoring”. |
| 2 | What do you use it for? | What each one answers on a normal day. If you only open it during an incident, say so. |
| 3 | What works there? | The one thing you would keep if you rebuilt the stack tomorrow. |
| 4 | What does not work? | The tool that is installed and nobody opens, or the one everybody fights. |
| 5 | What can you not answer? | One concrete question about your own system: “why did checkout get slow last Tuesday”. |
If you go blank on question 1
Walk these six. They are questions, not tool categories: “metrics: Zabbix” and “metrics: Prometheus” answer different ones.
- Is it alive? Zabbix, Nagios, health checks, uptime probes
- How much of it is there, and how fast is it? Prometheus, application metrics
- What happened? ELK, Loki, Splunk, files on disk
- Which path did this request take? Tempo, Jaeger, commercial APM
- Which function is burning CPU? Pyroscope, profilers, nothing
- Can an AI agent reach any of this? MCP servers, nothing
The last two questions are the point
You already own tooling and still signed up for this course. Whatever completes questions 4 and 5 is the gap. A tool nobody opens is not missing coverage, it is a rollout that stalled, and that costs more.
Your Bingo “cheatsheet”:
Reason a tool is not used:
- Nobody can read the query syntax
- The data is there, but you cannot line two signals up
- Instrumentation reached 3 services out of 40 and stopped
- Retention was cut to three days because it cost too much
- Names are inconsistent, so nobody trusts the data
- The team that rolled it out no longer exists