Competence grid for part 5

Use this grid to judge whether you can observe your platform, scale it and support it with a local model. Read each row and solve the task next to it without looking anything up. If it comes easily, you have reached the goal. If not, the link in the last column takes you to the right place. Within the Swiss framework curriculum, part 5 mainly serves the competences B14.2 (monitoring in operations), B13.5 (service levels) and B4.2 (applying current technologies).

I can …	Task	Look up
name the four golden signals	Match each signal with a metric of the web shop.	Metrics with Prometheus
roll out the monitoring stack from Git	Explain where the application monitoring takes its chart and values from and where the Grafana password lives.	Metrics with Prometheus
add targets to Prometheus	Show how the OpenNebula exporters and the web shop become targets.	Metrics with Prometheus
write PromQL queries	Write a query for requests per second by status code.	Metrics with Prometheus
describe the path of the logs	Describe the path of a log line from the pod through Alloy to Loki and explain what Loki indexes.	Logs with Loki
search logs with LogQL	Find the error lines of the web shop and count them per pod.	Logs with Loki
explain SLI, SLO and error budget	Calculate a Saturday's error budget for the web shop SLO.	Alerts and SLOs
justify a burn rate alert	Explain why the alert needs two time windows.	Alerts and SLOs
tell symptoms from causes and route alerts	Classify WebshopErrorBudgetBurn and KubePersistentVolumeFillingUp and explain what group_wait and repeat_interval do.	Alerts and SLOs
work out the number of pods the HPA wants	Calculate how many pods the HPA asks for with three pods at 90% utilization and a target of 60%.	Autoscaling
set up and limit autoscaling	Check whether maxReplicas with a pod's limit and the headroom for a rollout fits into the quota.	Autoscaling
evaluate a load test	Explain why more pods did not bring more throughput in the lab.	Autoscaling
compare autoscaling with OneFlow elasticity	Name two differences between the HPA and OneFlow and explain why the HPA grew faster in the lab.	Autoscaling
run and measure a local model (advanced)	Measure the tokens per second and explain what they depend on.	Advanced tasks
control access to the model (advanced)	Give one member access and show that another has none.	Advanced tasks
review a proposal of the model (advanced)	Assess a proposal of your triage service and decide with reasons.	Advanced tasks
assess the risks of a model in operations (advanced)	Name the safeguards of your triage service and what they protect against.	Advanced tasks
improve the questions to a model (advanced)	Explain how you compare two versions of a prompt fairly.	Advanced tasks

With this part you complete milestone 3. Rows marked as advanced belong to the advanced tasks in chapter 5.5 and are not part of the base. You hand in milestone 3 together with milestone 4 as hand-in 2. Part 6 is about security, backup, service levels and cost. As an advanced task there, you bring a member of your own onto the platform.