Problem statement
Two hundred pods, zero conventions. A release cannot find the pods it just shipped, the dashboards cannot group anything, and the cleanup script deletes by a hand-written list of names that is wrong within a day. The namespace fleet is a small slice of that mess: four workloads, each labeled a different way. Bring them to one scheme.
About this exercise
A label is a key/value pair in an object's metadata. Nothing in Kubernetes reads a label by itself - labels exist so that other things can find the object by them: selectors, services, dashboards, cleanup jobs. A fleet with no agreed keys is a fleet nothing can query, which is why the scripts fall back on name lists.
Labels are metadata, not spec, so they can be changed on a live object without recreating it. Three operations matter and they are deliberately different:
- Setting a key that is not there yet succeeds quietly.
- Setting a key that already has a value is refused unless the command is told that overwriting is intended. That refusal is the guard rail: it stops a bulk relabel from silently rewriting a value somebody else depends on.
- Removing a key is its own operation, not "setting it to empty". An empty value is still a value, and a selector for the key would still match it.
Labels can also be set at creation time, which is the only way a new object is queryable the moment it exists rather than at some later relabel.
kubectl label --help and kubectl run --help are the reference.
The scheme every object in fleet must carry - exactly these three keys, no others:
app the service's own name checkout, ledger, search, invoice, receipts tier the role it plays one of: web, worker, batch env where it runs one of: prod, staging, dev
squad is a retired key from the old convention. Nothing reads it any more and it must not survive on any object.
How this project is graded
The grader reads the labels off the live objects in fleet and compares them against the scheme, key by key and value by value. It checks that the objects you were not asked to touch still carry exactly what they started with, and that your readback file agrees with the cluster.
What you will build
One new pod, three corrections to live objects, and one readback file:
receipts-web a new pod in fleet, image nginx:alpine, labeled at creation search-indexer env is missing; tier says batch but it is a worker ledger-worker still carries the retired squad keyanswers/labels.txt one line per pod in fleet, pods sorted by name: <pod-name> <key>=<value>,<key>=<value>,... keys within a line sorted alphabetically, no other text
The intended labels:
receipts-web app=receipts tier=web env=prod search-indexer app=search tier=worker env=staging
Step 1 - labeled at birth
Create receipts-web in fleet from image nginx:alpine, carrying exactly the three scheme labels and nothing else. A pod created without labels and relabeled afterwards was unfindable in between - and a creation shortcut that invents its own label leaves a fourth key behind.
created_pod_carries_exactly_the_scheme_labels should pass
Step 2 - correcting live objects
search-indexer is missing env and its tier is wrong. Add the one that is absent and correct the one that is already set - the second is the case the command refuses by default.
relabel_adds_the_missing_scheme_key and overwrite_replaces_an_existing_value should pass
Step 3 - retiring a key
Take squad off ledger-worker. Its other three labels must come through unchanged, and invoice-batch - which you were not asked to touch - must be exactly as it was.
deprecated_key_removed_and_nothing_else_changed should pass
Step 4 - the fleet reads back clean
Every pod in fleet now carries exactly app, tier and env with values from the scheme. Record what the cluster reports into answers/labels.txt, in the format above.
every_object_matches_the_scheme and label_readback_matches_the_cluster should pass
When every test passes
You have:
- One label scheme carried by every object in the namespace, applied at creation and by correcting live objects
- A retired key removed without collateral damage to the labels beside it
- A readback that proves the fleet's labels from the cluster rather than from a spreadsheet
You do not have yet:
- Anything that uses the scheme. Labels only pay off when something queries them, and selectors are the next exercise.
Verify
Click Test after every step.
