Path 05 / Learning syllabus
Observability
Find out why a report is late by following metrics, logs and traces through the service. Then repair a gap in the evidence so the next investigator can see what happened.
Explore the syllabus
Your course project
An evidence-backed investigation
Investigate slow Dispatch exports, first with a synthetic evidence packet and later with an instrumented local service. Build a dashboard around the export journey, follow a request into background work and diagnose a deliberate telemetry gap. Deliver a reproducible investigation and one tested instrumentation improvement.
Before you begin
Start where it makes sense.
For engineers who understand a basic service but are new to telemetry. Start with supplied evidence; later lessons use a disposable local service and collector, with Linux basics as preparation. No production account is required.
Already know the foundations? Jump to any lesson in the syllabus.
Start with the supplied examples. Each exercise explains what you need, what to try and how to check the result; software-based exercises use a disposable practice environment.
The ordered path
What you will learn
Follow the sequence to build the course project, or open the lesson that answers your question.
- 01BeginnerLesson · 6 min read
Start with a question, then find evidence
Define the symptom, affected population and time window before interpreting a chart. Decide what observation would help next.
What you will try
Use the opening metric, log and trace to choose a discriminating check and explain why the evidence does not yet prove a cause.
- 02BeginnerLesson · 4 min read
Map what the telemetry needs to explain
Follow submission, queue time, worker execution and download. Identify which stages existing signals cover and which are invisible.
What you will try
Annotate the export journey with a question and required signal at each stage, including background completion.
- 03BeginnerLesson · 4 min read
Read metrics and calculate the right rate
Distinguish counters, gauges and distributions. Keep windows and populations aligned when calculating error fractions, latency and throughput.
What you will try
Calculate an export failure fraction from counts, compare it by region and show how a fleet average can hide the problem.
- 04IntermediateLesson · 4 min read
Write logs you can investigate with
Give events clear names, timestamps, service identity and useful correlation fields. Decide what information must be excluded or redacted.
What you will try
Rewrite an ambiguous failure event and use the revised fields to find the affected export without recording its sensitive contents.
- 05BeginnerLesson · 4 min read
Read a trace across concurrent work
Follow spans, parent relationships, links and overlapping operations. Distinguish elapsed time from the sum of work and recognize unexplained gaps.
What you will try
Compare two export traces and locate the extra time without adding overlapping child durations together.
- 06IntermediateLesson · 4 min read
Collect the first signals from a local service
Connect instrumentation, export, a collector and its debug destination. Verify each step with a known request before interpreting the data.
What you will try
Run the supplied lab, send a labelled synthetic request and verify its service identity, event and trace at the intended destination.
- 07IntermediateLesson · 4 min read
Connect a request to its background job
Carry useful context across process and queue boundaries. Link request identifiers, log events and traces while showing where correlation is unavailable.
What you will try
Follow one export from submission into its worker, then locate and repair a deliberate break in context propagation.
- 08IntermediateLesson · 4 min read
Build a dashboard around the user journey
Choose panels that distinguish demand, completion, errors and waiting work. Make time range, units and population visible.
What you will try
Create a view of export completion and queue age, then find a region-specific delay that disappears in the aggregate.
- 09IntermediateLesson · 4 min read
Detect a failure and test the alert end to end
Translate the service’s agreed response objective into a signal and routing rule. Verify what a responder receives, including missing or delayed data.
What you will try
Exercise the alert with a controlled lab failure and check its scope, timing, owner and link to the first useful investigation.
- 10IntermediateLesson · 4 min read
Control telemetry volume without losing the question
Evaluate label cardinality, sampling, retention and sensitive fields against the investigations the service needs to support.
What you will try
Compare counts with sampled traces, remove an unbounded label and document which questions the revised collection can no longer answer.
- 11IntermediateLesson · 4 min read
Investigate a failure in the telemetry pipeline
Distinguish a healthy service from missing evidence. Follow collection, buffering and export failures before trusting an empty dashboard.
What you will try
Break the lab’s export path, use independent requests to check the service and identify where its telemetry stopped.
- 12IntermediateLesson · 4 min read
Complete an investigation another engineer can repeat
Combine the symptom, queries, timeline and decisive checks into an explanation with clear limits. Test one improvement to the evidence.
What you will try
Investigate the staged export slowdown, present competing explanations and verify that one instrumentation change exposes a previously missing step.
