🖼️ See the drawing + lab 🏠 Course home 🌿 Branch on GitHub ✏️ View source

🖼️ आकृती आणि labThe drawing + lab पूर्ण पानावर उघडा ↗Open full page ↗

🙋 धडा 07 — Health probes: "जागी आहेस का? तयार आहेस का?"

📍 तुम्ही इथे आहात: 13 पैकी धडा 07 · मागचा: lesson-06-configmaps-secrets · पुढचा: lesson-08-resources

📦 या ब्रँचमध्ये काय आहे

धडे 01–06, आणि: liveness आणि readiness probes — Kubernetes निरोगी pod आणि zombie यातला फरक कसा ओळखतो. खरी file:

k8s/deployment.yaml — livenessProbe (/healthz) आणि readinessProbe (/readyz)

🧒 5 वर्षांच्या मुलाला समजावल्यासारखे

शिक्षिका वर्गात फिरतात आणि प्रत्येक मुलाला दोन वेगवेगळे प्रश्न विचारतात:

"जागा आहेस का?" 😴 — एखादे मूल बाकावर डोके टेकून झोपले असेल आणि लागोपाठ अनेक वेळा उत्तर देत नसेल, तर शिक्षिका त्याला नर्सकडे पाठवतात आणि ताजेतवाने मूल आणतात. (कडक शाळा. पण खूप परिणामकारक.) → हा liveness probe: उत्तर नाही = container restart करा.
"प्रश्नांची उत्तरे द्यायला तयार आहेस का?" 🙋 — PE च्या तासावरून नुकतेच परत आलेले, अजून बुटाच्या नाड्या बांधत असलेले मूल जागे आहे पण तयार नाही. मूल हात वर करेपर्यंत शिक्षिका त्याला प्रश्न विचारतच नाहीत. → हा readiness probe: तयार नाही = तुला traffic नाही (धडा 04 मधली Service तुला वगळते), पण restart पण नाही.

दोन प्रश्न, दोन अगदी वेगळे परिणाम. या दोघांची सरमिसळ करणे ही Kubernetes मधली सर्वात नेहमीची चूक आहे!

🗺️ आकृती

flowchart TB
    K["kubelet - the teacher<br/>asks every few seconds"]
    K -->|"GET /healthz"| L{"alive?"}
    L -->|"yes"| ok1["😊 leave it alone"]
    L -->|"no, several times"| restart["🔄 RESTART container<br/>same pod, fresh process"]
    K -->|"GET /readyz"| R{"ready?"}
    R -->|"yes"| ok2["✅ Service sends traffic"]
    R -->|"no"| skip["⏸️ taken OUT of the Service list<br/>no traffic, NO restart"]

❓ काय

Liveness probe — "process वाचवता येणार नाही इतकी अडकली आहे का?" पुन्हा पुन्हा fail झाला → kubelet container restart करतो. Deadlocks आणि गोठलेल्या apps वरचा उपाय.
Readiness probe — "तू आत्ता requests ना सेवा देऊ शकतोस का?" Fail झाला → तो पुन्हा पास होईपर्यंत pod ला Service च्या endpoint यादीतून काढले जाते. "अजून सुरू होत आहे" आणि "माझा database थोड्या वेळासाठी पोहोचत नाही" यावरचा उपाय.
Probes httpGet (आपला), tcpSocket, किंवा command exec करणारे असू शकतात. Knobs: initialDelaySeconds (सुरू झाल्यानंतरचा सवलतीचा वेळ), periodSeconds (किती वेळाने), failureThreshold (कृती करण्याआधी किती वेळा चुकू द्यायचे).
खूप हळू सुरू होणाऱ्यांसाठी startup probe पण असतो — इथे त्याची गरज नाही.

🤔 का

Probes नसतील तर Kubernetes ला फक्त exit होणारा container लक्षात येतो. जिवंत-पण-गोठलेल्या process ला कायम traffic मिळत राहील. Zero-downtime rollouts (धडा 11) पण probes मुळेच चालतात: नव्या pod ला traffic फक्त /readyz ने हो म्हटल्यावरच मिळते — म्हणून सुरूच न होऊ शकणाऱ्या खराब build ला एकही request मिळत नाही. Self-healing (धडा 03) + probes = cluster स्वतःला स्वतः चालवतो.

🔧 कसे (या repo मध्ये)

k8s/deployment.yaml मध्ये:

livenessProbe:
  httpGet: { path: /healthz, port: 3000 }
  initialDelaySeconds: 5     # let it boot before judging
  periodSeconds: 10          # ask every 10s
readinessProbe:
  httpGet: { path: /readyz, port: 3000 }
  initialDelaySeconds: 5
  periodSeconds: 5           # ask more often — traffic depends on it

App दोन्ही endpoints apps/school-api/src/server.js मध्ये लागू करते: /healthz = "process प्रतिसाद देते तरी आहे"; /readyz = "आणि माझे DB connection चालते". Ingress (धडा 10) AWS load balancer च्या checks साठी /healthz च पुन्हा वापरतो.

🧪 करून पाहा

kubectl -n school describe pod -l app=school-api | grep -A3 -i liveness

# Watch readiness gate traffic:
kubectl -n school get pods -w    # READY column: 0/1 until /readyz passes, then 1/1

# See restarts caused by liveness (RESTARTS column):
kubectl -n school get pods
# Try it: kubectl exec into a pod and kill the node process —
# the container restarts in place, pod name stays the same:
kubectl -n school exec deploy/school-api -- sh -c 'kill 1'
kubectl -n school get pods       # RESTARTS just went up by 1 🎉

🔧 या धड्यासाठी kubectl

kubectl describe pod (Events) · kubectl get pods -w

🔗 यावर आधारित: 🪪 AWS L15 (ALB health check = readiness)

✅ तपासा — तुम्हाला काय दिसायला हवे

/readyz बिघडवा → pod Running राहतो पण Service च्या endpoints मधून बाहेर जातो; /healthz बिघडवा → RESTARTS वाढतो.

🧹 साफसफाई

local cluster: kubectl delete -f k8s/ (किंवा तुम्ही apply केलेली file) — EKS वर चालू राहिलेल्या प्रत्येक गोष्टीचे तासाप्रमाणे बिल लागते

⚠️ नेहमीच्या चुका

एकाच probe कडून दोन्ही कामे करवून घेणे — readiness traffic अडवतो, liveness restart करतो
initialDelaySeconds/startup probe शिवाय हळू सुरू होणाऱ्या app वर liveness probe → restart चे चक्र

⏭️ पुढे

निरोगी pods, छान — पण प्रत्येक जण किती CPU आणि memory खाऊ शकतो? Requests आणि limits — जेवणाचे वाटे.

git checkout lesson-08-resources

🙋 Lesson 07 — Health probes: "Are you awake? Are you ready?"

📍 You are here: Lesson 07 of 13 · Previous: lesson-06-configmaps-secrets · Next: lesson-08-resources

📦 What's in this branch

Lessons 01–06, plus: liveness and readiness probes — how Kubernetes tells a healthy pod from a zombie. Real file:

k8s/deployment.yaml — livenessProbe (/healthz) and readinessProbe (/readyz)

🧒 Explain like I'm 5

The teacher walks around the class and asks each kid two different questions:

"Are you awake?" 😴 — If a kid fell asleep face-down on the desk and doesn't answer several times in a row, the teacher sends them to the nurse and brings in a fresh kid. (Harsh school. Very effective.) → That's the liveness probe: no answer = restart the container.
"Are you ready to answer questions?" 🙋 — A kid who just came back from PE, still tying shoelaces, is awake but not ready. The teacher simply doesn't call on them until they raise their hand. → That's the readiness probe: not ready = no traffic for you (the Service from lesson 04 skips you), but no restart either.

Two questions, two very different consequences. Mixing them up is one of the most common Kubernetes mistakes!

🗺️ Diagram

flowchart TB
    K["kubelet - the teacher<br/>asks every few seconds"]
    K -->|"GET /healthz"| L{"alive?"}
    L -->|"yes"| ok1["😊 leave it alone"]
    L -->|"no, several times"| restart["🔄 RESTART container<br/>same pod, fresh process"]
    K -->|"GET /readyz"| R{"ready?"}
    R -->|"yes"| ok2["✅ Service sends traffic"]
    R -->|"no"| skip["⏸️ taken OUT of the Service list<br/>no traffic, NO restart"]

❓ What

Liveness probe — "is the process stuck beyond saving?" Fails repeatedly → kubelet restarts the container. Cure for deadlocks and frozen apps.
Readiness probe — "can you serve requests right now?" Fails → pod is removed from the Service's endpoint list until it passes again. Cure for "still starting up" and "my database is briefly unreachable".
Probes can be httpGet (ours), tcpSocket, or exec a command. Knobs: initialDelaySeconds (grace period after start), periodSeconds (how often), failureThreshold (how many misses before acting).
There's also a startup probe for very slow starters — not needed here.

🤔 Why

Without probes, Kubernetes only notices a container that exits. A process that's alive-but-frozen would keep receiving traffic forever. Probes are also what makes zero-downtime rollouts (lesson 11) work: a new pod gets traffic only after /readyz says yes — so a bad build that can't start never receives a single request. Self-healing (lesson 03) + probes = the cluster runs itself.

🔧 How (in this repo)

In k8s/deployment.yaml:

livenessProbe:
  httpGet: { path: /healthz, port: 3000 }
  initialDelaySeconds: 5     # let it boot before judging
  periodSeconds: 10          # ask every 10s
readinessProbe:
  httpGet: { path: /readyz, port: 3000 }
  initialDelaySeconds: 5
  periodSeconds: 5           # ask more often — traffic depends on it

The app implements both endpoints in apps/school-api/src/server.js: /healthz = "process responds at all"; /readyz = "and my DB connection works". The Ingress (lesson 10) reuses /healthz for the AWS load balancer's checks.

🧪 Try it

kubectl -n school describe pod -l app=school-api | grep -A3 -i liveness

# Watch readiness gate traffic:
kubectl -n school get pods -w    # READY column: 0/1 until /readyz passes, then 1/1

# See restarts caused by liveness (RESTARTS column):
kubectl -n school get pods
# Try it: kubectl exec into a pod and kill the node process —
# the container restarts in place, pod name stays the same:
kubectl -n school exec deploy/school-api -- sh -c 'kill 1'
kubectl -n school get pods       # RESTARTS just went up by 1 🎉

🔧 kubectl for this lesson

kubectl describe pod (Events) · kubectl get pods -w

🔗 Builds on: 🪪 AWS L15 (ALB health check = readiness)

✅ Verify — what you should see

break /readyz → the pod stays Running but leaves the Service's endpoints; break /healthz → RESTARTS increments.

🧹 Clean up

local cluster: kubectl delete -f k8s/ (or the file you applied) — on EKS, anything left running bills by the hour

⚠️ Common mistakes

one probe doing both jobs — readiness gates traffic, liveness restarts
a liveness probe on a slow-starting app with no initialDelaySeconds/startup probe → restart loop

⏭️ Next

Healthy pods, great — but how much CPU and memory may each one eat? Requests & limits — lunch portions.

git checkout lesson-08-resources

← Previousconfigmaps secrets Next →resources

This page is the lesson's README from the lesson-07-health-probes branch, shown here so the whole School stays on one site. Code files open on GitHub at the same branch.