Flashcard Deck — har module ke must-remember Q→A¶
Roz 5 minute. Yehi ek aadat sabse zyada lifetime retention deti hai.¶
Kaise use karein (active recall): 1. Sirf left column (Sawaal) dekho — right column ko haath/kaagaz se dhak lo. 2. Memory se jawab bolo (ya likho). Retrieve karo, re-read mat karo. 3. Phir right column kholo aur check karo. Galat nikla? Us card pe ⭐ lagao — kal usse pehle karo. 4. Spaced schedule: naya module → Day 1, Day 3, Day 7, Day 30. (See 00-INDEX §0.1.)
🇮🇳 Hinglish intuition: Flashcard = memory ka dumbbell. Ek baar uthana nahi, baar-baar badhte gap pe uthana muscle (yaaddaasht) banata hai.
Anki users: har module table ko copy karke pipe (
|) separator se import kar sakte ho — Front = Sawaal, Back = Jawab.
Deck size: ~190 cards across M0–M9 + the connected-system (SYS) + both capstones. Jump: M0 · M1 · M2 · M3 · M4 · M5 · M6 · SYS · M8 · M9 · CAP1 · CAP2
M0 — Foundations¶
Teaching chapter: 01-M0-foundations.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| DevOps kya solve karta hai? | Dev (change chahiye) vs Ops (stability chahiye) ki wall todta — automation + Git se Speed, Safety, Repeatability. |
| 3 pillars kaunse hain? | Speed (jaldi ship), Safety (galti se bacho / rollback), Repeatability (har baar same result). |
| Provisioning vs Configuration? | Provisioning = khaali machine banao (Terraform). Configuration = us machine ke andar software install/setup (Ansible). |
| Packaging vs Orchestration? | Packaging = app + dependencies ek dabbe me band (Docker). Orchestration = bahut dabbe reliably chalao + heal (Kubernetes). |
| 4 layers aur unke tools? | Provisioning=Terraform, Configuration=Ansible, Packaging=Docker, Orchestration=Kubernetes. Mantra: T→A→D→K. |
| Stateful vs Stateless? | Stateful = data andar rakhta (maaro to data gaya) — DB. Stateless = kuch nahi rakhta (maaro, same replacement aa jaata) — API. |
| "App jo DB use karti hai woh stateful hai" — sahi? | Galat. App khud stateless hai; state DB me hai. App disposable, DB nahi. (Classic interview trap.) |
| Pets vs Cattle? | Pet = naam-wala, pyaar se sambhaala, mare to dukh (special server). Cattle = numbered, koi bhi ek jaisa, mare to naya (disposable pod). |
| Two loops kya hain? | Outer/setup (Terraform+Ansible, kabhi-kabhi, Pets) + Inner/delivery (push→CI→Argo→K8s, har push, Cattle). Dono cluster pe milte hain. |
| Error-type → layer reflex? | Error ka type location batata hai: UNREACHABLE=network, Permission denied=SSH key, OOMKilled=RAM, CrashLoopBackOff=app/logs. |
| Idempotency vs Reconciliation? | Idempotency = ek operation dobara chalao, kuch naya nahi (switch). Reconciliation = loop lagataar desired vs current fix karta (thermostat). |
| Junior vs senior — asli fark? | Commands Google/AI de dega. Engineer DESIGN + DECIDE karta — "kaise banaoge, kyun" — wahi interview me poochte hain. |
| Monolith vs microservices? | Monolith = ek bada app (simple start, deploy/scale tight). Microservices = chhote independent services (flexible, par network/ops complexity badhti). |
M1 — Terraform¶
Teaching chapter: 02-M1-terraform.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| IaC kya hai? | Text files me infrastructure describe karna — Git me, reviewable, reproducible, auditable. No ClickOps. |
| Declarative vs imperative — ek line? | Declarative = "kya chahiye" batao (destination); imperative = "kaise karo" likho (steps). Terraform declarative hai. |
terraform plan kya karta hai? |
Safe dry-run — code + state + real AWS compare karta, kuch badalta nahi. Bill dekhna pehle, payment baad me. |
terraform apply ke baad? |
Real API calls, real cost, state update. Payment ho gayi — wapas nahi. |
| tfstate file kya hai, kyun critical? | Terraform ki diary 📔 — kya banaya, real cloud IDs. Gayi to TF andha — sab naya banata, duplicates. |
| tfstate S3 me kyun, laptop pe kyun nahi? | Laptop = sirf tumhare paas; teammate ka TF andha; simultaneous apply = state corrupt. S3 = shared almari + DynamoDB = taala 🔒. |
| Drift kya hai, recover kaise? | Reality badli, state intact. plan drift dikhata; apply reality ko code pe wapas laata. |
| Lost-state vs drift (trap!)? | Lost-state = diary jal gayi → TF zero sochta → sab rebuild → duplicates/orphans. Drift = diary intact, reality badli → apply se theek. Alag recovery. |
| Orphaned resource, fix? | Cloud me exist karta, state me nahi — bina maalik ki gaay. Bill aa raha. Fix: terraform import aws_instance.web i-0abc.... |
| Module = kya, kaise? | Black-box: inputs=variables, outputs=outputs, andar hidden. Root modules ko wire karta — modules aapas me directly baat nahi karte. |
| Dev vs prod me alag tfstate key kyun? | destroy dev me = sirf dev ka state; prod untouched. Ek shared key = dev destroy se prod offline. |
Provisioner remote-exec kyun avoid? |
Sirf creation pe chalta, re-apply pe nahi — not idempotent; failures state me invisible. Use Ansible instead. |
skip_final_snapshot = true prod me kyun deadly? |
Accidental destroy = DB permanently gone, no snapshot, no undo. Prod me hamesha false. |
| Idempotent apply — hook? | count = 3 code 100 baar apply = 3 servers. Light switch: sau baar ON dabao, ON hi rehta — 100x brighter nahi. |
M2 — Ansible¶
Teaching chapter: 03-M2-ansible.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Ansible kya hai — ek line? | Agentless, push-based configuration management. SSH + Python se desired state enforce karta, safely repeatable. |
| Agentless ka matlab? | Managed nodes pe koi agent install nahi — sirf SSH + Python. Plumber kaam kare, apni copy chhode bina nikal jaaye. |
| Push vs pull — Ansible kahan? | Ansible = push (tum trigger karo, immediate SSH). Argo CD = pull (agent poll karta). Push = on-demand; pull = continuous. |
| Inventory file ka kaam? | "Kahan" — managed nodes ki list + groups. Playbook = "kya karo"; inventory = "kahan karo". |
ok vs changed (interview)? |
ok = already correct, kuch nahi kiya. changed = Ansible ne fix kiya. 2nd run me changed=0 = convergence = idempotency ka proof. |
| Module vs shell — kyun module? | Module smart: pehle check, phir act. Shell andha: blindly chalata (echo >> file har run append = corruption). Shell tabhi jab module na ho, creates:/when: se guard. |
| Handler kya, kyun regular task nahi? | Handler sirf changed pe fire karta (notify 🔔 se). Config same → no restart → no downtime. Regular task har run restart = needless downtime. |
--check kya karta hai? |
Dry-run — kya WOULD change, kuch badalta nahi. Ansible ka terraform plan. |
kubeadm init pe creates: guard kyun? |
"Yeh file exist karti hai to kaam ho chuka — skip". Bina guard re-run = doosra kubeadm init cluster corrupt kar sakta. |
| 3-playbook K8s order kyun? | common → master → workers. Dependency chain: containerd pehle, phir kubeadm init, phir workers join. Order enforce karo. |
| EKS/RDS pe Ansible kyun kam? | EKS = no master EC2 to SSH; RDS = managed, no host. Ansible ka job evaporate. Rakho: bastion config, fleet ops, on-prem. |
| ansible-vault kya karta hai? | Secrets file AES-256 encrypt. Git me plaintext creds kabhi nahi. Runtime --ask-vault-pass se decrypt. |
| Convergence — definition + proof? | System desired state reach karke hold kar raha. Proof: 2nd run = changed=0. Golden interview answer. |
M3 — Docker¶
Teaching chapter: 04-M3-docker.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Container kya hai — kernel level? | Linux process, isolated via namespaces (private views) + cgroups (CPU/RAM limits). Host kernel share; no guest OS. |
| Container vs VM — 3 differences? | Container: ms start, MBs, 100s/host, shared kernel. VM: minutes, GBs, ~10/host, own kernel. Flat vs makaan. |
| Image vs container — hook? | Image = recipe (read-only blueprint, docker build). Container = pakwaan (running instance, docker run). Ek image → kai containers. |
| Layer kya hai — hook? | Har Dockerfile instruction = ek immutable layer. Cache hit = instant. Save-point 💾. |
| Cache invalidation kab? | Koi instruction/input change → us layer pe cache miss → neeche ke saare layers rebuild. Top-down only, upar kabhi nahi. |
| Layer-order trick? | Dependencies (rarely change) upar, code (roz badalta) neeche. Code edit → sirf last COPY rebuild; pip install cached. 2 min → 3 sec. |
| Build context + .dockerignore kyun? | Context = daemon ko bheji directory (docker build .). .dockerignore se node_modules/, .git/, .env exclude — chhota context, secrets safe. |
| Multi-stage build kyun? | Build stage = bada toolchain; runtime stage = sirf output + prod deps. Build stage discard → image 60-90% chhota, attack surface kam. |
latest tag trap? |
Mutable label — jo last push kare pointer move. 2 nodes alag time pull → different images; rollback impossible. Use git SHA: myapp:abc1234. |
| RUN vs CMD vs ENTRYPOINT? | RUN = build-time (layer). CMD = runtime default (override-able). ENTRYPOINT = fixed executable. ENTRYPOINT + CMD = fixed + swappable args. |
| ImagePullBackOff — diagnose? | kubectl describe pod → Events. (1) tag nahi → manifest unknown; (2) auth → pull access denied → imagePullSecret/ECR IAM; (3) network → timeout → NAT/SG. |
| Compose DNS — kaise? | Docker embedded DNS service names → container IPs. web → db:5432, IP yaad nahi karna. K8s Service ka same idea. |
docker history kab? |
Har layer + size + command dekhne ke liye — bloated image debug ya audit. |
M4 — Kubernetes Core¶
Teaching chapter: 05-M4-kubernetes-core.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Reconciliation loop kya karta? | Desired vs current ka fark dhoondh ke band karta — 24/7. Chowkidar jo hamesha ginta. |
| Pod vs Container? | Pod = tiffin 🍱 (shared network+storage). Container = andar ki dish (app process). Ek pod me ek/zyada containers. |
| Bare pod prod me kyun nahi? | Bare pod ke upar koi desired state nahi — crash = K8s naya nahi banata. Hamesha Deployment. |
| Deployment→ReplicaSet→Pod — tum kisse baat karte ho? | Sirf Deployment (scale, rollback). ReplicaSet+Pod K8s khud manage karta. |
| Service kyun zaroori? | Pod IP har baar badle; Service = stable virtual IP + DNS jo kabhi nahi badalti. Fixed phone number ☎️. |
| Labels ke 2 use? | (1) Traffic routing: Service selector → pod label. (2) Placement: pod nodeSelector → node label. |
| Readiness fail = ? | Pod endpoints se hata, traffic band — kill NAHI. Pass hone pe wapas add. Traffic signal 🚦. |
| Liveness fail = ? | Pod kill + restart — stuck/deadlock hatao. Pulse check 💓. |
| Running vs Ready? | Running = process alive. Ready = readiness pass, traffic milega. READY 0/1 = alive par probe fail → koi traffic nahi. |
| Master pe app pods kyun nahi? | NoSchedule taint 🚷. App pods master overload karein → control plane crash → poora cluster mute. |
| Pod Pending, CPU/RAM free — 3 wajah? | (1) IP pool khatam, (2) 110-pod cap hit, (3) nodeSelector/taint mismatch. Dekho: kubectl describe pod Events. |
| Deployment vs StatefulSet? | Deployment = stateless cattle 🐄 (koi bhi pod same). StatefulSet = stateful pet 🐶 (stable naam pod-0/1, apna PV). DB → RDS behtar. |
| PersistentVolume kya? | Pod ke bahar ki disk — locker 🔒 jo pod maarne pe bhi zinda. Naya pod same PV se judta, data safe. |
| k3s vs kubeadm vs EKS? | k3s = free-tier/studio flat. kubeadm = khud-banaya ghar. EKS = managed flat. |
imagePullPolicy trap? |
Mutable tag (:latest/:prod) + IfNotPresent = node purani cached image chalata rahega, deploy "hua" par pod purana. Fix: immutable SHA tag. |
restartPolicy — Deployment vs Job? |
Deployment/StatefulSet = Always (hamesha restart). Job/CronJob = OnFailure/Never (success pe ruk jao). Isliye web pod wapas aata, complete Job nahi. |
| Service ke 4 port fields? | containerPort (app sun raha) = targetPort (Service kis pod-port pe bheje) ← port (Service ka apna port) ← nodePort (bahar ka gate, NodePort only). targetPort ≠ containerPort = #1 "connection refused" bug. |
| initContainer vs sidecar? | initContainer = main se pehle chal ke exit (wait-for-DB, migration); fail = pod Init:0/1 atka. Sidecar = main ke saath chalta rehta (log shipper, proxy). |
M5 — Sizing & Cost¶
Teaching chapter: 06-M5-sizing-and-cost.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Instance family kaise choose? | Pehle "bhookh": balanced=M, CPU=C, RAM=R, extreme RAM=X, IO=I, GPU=P/G, spiky-dev=T. Family galat to koi size nahi bachata. |
m6g.xlarge decode? |
m=general |
| T-series CPU-credit trap? | Idle pe credits earn, burst pe spend. Credits=0 → baseline cap (~10-20%) pe lock — app slow, no error. Dekho CloudWatch CPUCreditBalance. |
| exit 137 = ? | OOMKilled — RAM limit cross. describe pod confirm; top pod actual dekho; memory.limits badhao ya R-family. |
| CPU throttle kaise pehchano? | Silent slow API, no error, no restart. CPU limit cross → process slow par alive. Limits badhao ya C-family. |
| requests vs limits? | request = reserved seat (scheduler placement). limit = deewar (CPU cross→throttle alive; RAM cross→OOMKilled/137). Dono zaroori prod me. |
| Fragmentation → Pending kyun? | Total free kaafi, par koi ek node poori request fit nahi karta. Bus me seats hain par saath nahi. CA naya node add karega. |
| On-Demand vs Reserved vs Spot? | On-Demand=dev/unpredictable. Reserved/Savings=24/7 baseline (40-70% off). Spot=stateless only (70-90% off, 2-min warning — DB pe kabhi nahi). |
| HPA kya scale karta? | Pod count — CPU%/custom metric pe. Nodes nahi. |
| Cluster Autoscaler kab? | Jab pod Pending (koi node fit nahi) → naya node. Chain: HPA → pods Pending → CA → node → schedule. |
| VPA kya, conflict? | Pod ke requests vertically resize (restart hota). HPA ke saath CPU-target conflict = oscillation. VPA recommendation-mode me rakho. |
| Headroom kyun? | 20-30% buffer: spike se pehle HPA react, rolling update ke liye, OS/kubelet overhead. 100% bhara = Pending. |
| DB sizing rule? | Primary prod DB kabhi K8s pod me nahi — RDS/Aurora. Backups/failover/patching AWS sambhaale. |
M6 — CI/CD¶
Teaching chapter: 07-M6-cicd.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| CI vs Delivery vs Deployment (3 line)? | CI = test+build+package har push. Delivery = artifact hamesha prod-ready, human approve. Deployment = koi gate nahi, auto prod. |
| Runner kya, kyun ephemeral? | Fresh VM jo ek job chalata phir khatam. Koi prior state nahi → deterministic. Isliye har run deps install (ya cache). |
github.sha tag kyun, latest nahi? |
Immutable — ek SHA hamesha same image. Rollback exact. latest mutable = rollback ambiguous. |
needs: kya karta? |
Job dependency — build-push tab chale jab test pass. Test fail = build-push cancel. |
| Manifest-update pattern? | CI cluster ko chhuta nahi — sirf k8s/deployment.yaml me tag update + Git commit. Argo andar se apply karta. CI ke paas cluster creds nahi. |
| CI infinite loop + 3 fixes? | CI manifest commit → woh CI trigger → loop. Fix: (1) paths: filter, (2) [skip ci] in message, (3) GITHUB_TOKEN (bot commit re-trigger nahi karta). |
contents: write kyun, dono jagah? |
TOKEN default read-only. YAML me permissions: contents: write AND repo Settings→Actions→Read+write. Ek bhi miss = "Permission denied". |
| Matrix build kya karta? | Ek job def → N parallel runners. Wall-clock = sabse slow service (sum nahi). 4+6+3 min sequential vs 6 min parallel. |
if: condition kab? |
Push jobs ko PR pe rokne: if: github.ref == 'refs/heads/main'. PR pe sirf test, push nahi. |
CI se seedha kubectl apply kyun nahi? |
CI ke paas cluster creds honge — compromised runner = prod access. Manifest-update me blast radius = sirf Git repo. |
uses: vs run:? |
uses: = tested marketplace action (edge cases handle). run: = khud ka shell. Checkout/auth/login ke liye uses: prefer. |
| Test gate kya karta? | Fail = kuch ship nahi hota. Broken code image hi nahi banti. |
SYS — The Connected System¶
Teaching chapter: 09-connected-system.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Do loops + kahan milte hain? | Outer (Terraform+Ansible, infra, Pets, rarely) + Inner (CI+Docker+Argo+K8s, delivery, Cattle, every push). Milte hain Kubernetes cluster pe. |
| Bridge 1 + failure? | Terraform→Ansible: terraform output IPs → inventory.ini. Toot: wrong IP → UNREACHABLE. |
| Bridge 2? | Ansible→K8s: 3 playbooks (common→init→join) → live cluster. Toot: no containerd/wrong token → join fail. |
| Bridge 3? | Docker→Pod: image ECR push, kubelet pull. Toot: bad tag/no auth → ImagePullBackOff. |
| Bridge 4? | git push→CI: on: push: trigger. Toot: wrong branch filter → kabhi nahi chalta. |
| Bridge 5 (⭐) kya + kyun star? | CI→Git manifest: CI tag update + commit, cluster seedha nahi chhuta. Interviewers yahi probe karte. Isliye CI ko cluster creds nahi chahiye. |
| Bridge 6? | Git→Argo→K8s: Argo changed manifest pull + apply → rolling update. Toot: OutOfSync never syncs → purana version. |
| Bridge 7? | Pod→Service→User: EndpointSlice ready pods route; NodePort/Ingress expose. Toot: no ready pods → 503. |
| Bridge 8? | Pod→RDS: psycopg2 → rds:5432 env-var host. Toot: wrong SG/password → timeout/auth error. |
| Thread 1 — Reconciliation? | Loop desired vs current compare karke drive karta — hamesha. TF/Ansible manually; K8s/Argo continuously. |
| Thread 2 — State outside? | Durable state (tfstate, DB) compute se bahar → pods/servers Cattle. Kill+replace freely, no data loss. |
| Thread 3 — Preview before apply? | terraform plan, ansible --check, kubectl --dry-run, CI test gate. Bill dekho phir payment. |
| Thread 4 — Push vs Pull? | Push (Ansible/Actions): initiator ke paas target creds. Pull (Argo): agent andar se Git read, creds bahar nahi. Git = neutral middle. |
| Thread 5 — Idempotency? | Same op dobara = same result, no duplicates. apply 2x = same VPC. Switch, counter nahi. |
| Blast-radius boundary? | Har layer ka alag owner. Design karte waqt: "kis owner ko batana? Yeh box fail = kya toot ta?" = architecture review ka 80%. |
M8 — Observability & SRE¶
Teaching chapter: 10-M8-observability-sre.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| 3 pillars + har ek ka sawaal? | Metrics (Prometheus): kitna/kitne/kitni-baar over time? Logs (Loki): is ek request me exactly kya hua? Traces (OTel+Jaeger): time kahan/kis span me gaya? |
| Prometheus pull model? | Prometheus khud app ka /metrics scrape karta har ~15s — push nahi. Same pull idea jaise Argo CD. |
| Counter vs Gauge? | Counter sirf badhta, restart pe reset → hamesha rate() se query. Gauge upar-neeche (memory, queue) → raw value theek. |
| Histogram latency ke liye kyun? | Buckets store karta (<100ms, <500ms) → p50/p95/p99 nikal sakte. Average se p99 nahi milta. |
| Cardinality explosion kya? | High-cardinality label (user_id, order_id) → millions of series → Prometheus RAM explode → OOMKilled. Labels bounded rakho. |
| SLI / SLO / SLA? | SLI = jo measure karo. SLO = internal target (stricter). SLA = external contract (penalty). SLO hamesha SLA se strict (buffer). |
| Error budget + governance? | 100% − SLO. Budget bacha → risky features ship. Budget khatam → sirf reliability. Number politics ko replace karta. |
| 4 Golden Signals? | Latency, Traffic, Errors, Saturation. Yeh 4 = almost any service ki health picture. |
| Alert on symptoms vs causes? | Cause: CPU>80% (user ko feel nahi = fatigue). Symptom: error_rate>1% for 5m (user suffer). Symptom pe page; cause dashboard pe silent. |
for: 5m threshold se important kyun? |
Threshold = kab trigger; for: = kab page. 30s spike self-resolve — bina for: 3am fake page. Real incidents persist. |
| Structured logging kyun? | JSON + trace_id har line → queryable, trace se join, no code change during incident. print() = 3am regex hell. |
| p99 vs average? | Average tail chhupata (99% fast, 1% dying = average theek dikhega). p99 real user pain dikhata. |
M9 — Advanced K8s Internals¶
Teaching chapter: 11-M9-advanced-k8s-internals.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Startup probe kya karta? | Boot ke dauran liveness+readiness ko hold karta; slow-start apps ko CrashLoopBackOff se bachata. |
| Readiness fail consequence? | Pod EndpointSlice se remove — traffic band, pod alive, restart NAHI. |
| Liveness fail consequence? | Container restart (kill) — SIGTERM phir SIGKILL. |
| Liveness footgun kya? | Liveness me DB/external check — DB slow → saare pods cascade restart. DB check readiness me daalo. Smoke detector jo poori building khali kara de. |
| QoS Guaranteed kaise? | requests == limits (CPU aur memory dono, dono set). Last to evict. |
| QoS BestEffort kab? | Koi requests/limits set nahi. Memory pressure pe pehle evict. Titanic lifeboat priority sabse neeche. |
| Graceful shutdown race + fix? | SIGTERM aur EndpointSlice-removal async — SIGTERM pehle aa sakta jab requests abhi aa rahe. Fix: preStop: sleep 5. |
| HPA formula? | ceil(currentReplicas × currentMetric / targetMetric). ceil = round up. |
| HPA scale-down slow kyun? | 300s stabilization window — transient drop pe pod hataane se flapping rokta. |
| HPA unknown/50% kab? | metrics-server nahi, ya pods me requests set nahi. |
| CoreDNS FQDN format? | <service>.<namespace>.svc.cluster.local → ClusterIP resolve. |
| nginx Ingress 404 — pehla check? | Request ka Host: header Ingress rule ke host: se match karta? curl -H 'Host: ...' se test. |
| EndpointSlice kya karta? | Ready pods ka live list; kube-proxy isse padh ke iptables/IPVS rules banata. |
| iptables vs IPVS kab? | 500+ Services pe: iptables O(n), IPVS O(1) hash — measurably faster at scale. |
| kube-proxy vs Calico — kaam? | kube-proxy = Service networking (ClusterIP → sahi pod). Calico/CNI = Pod networking (pod↔pod, IPAM, NetworkPolicy). Dono chahiye. |
| kube-proxy packet forward karta? | Nahi. Sirf iptables/IPVS rules program karta; kernel har packet forward karta. "Rule ek baar likha, kernel har baar chalata." |
| kaun sa K8s component NAHI? | containerd (CRI) aur Calico (CNI) pluggable third-party hain — K8s sirf use karta. kube-proxy asli K8s component hai. Cilium/eBPF to kube-proxy ko hi replace kar sakta. |
| RBAC blast radius? | Pod cluster-admin + compromise = attacker ko full cluster (Secrets, pod create, data delete). Least privilege do. |
CAP1 — Capstone: URL Shortener¶
Chapter: 12-capstone-url-shortener.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Kaunsi 3 APIs? | POST /shorten → {short}; GET /{code} → 302 redirect; GET /health → {status:ok}. |
| State pods me kyun nahi? | Pod delete = data gone. RDS = managed, private subnet, survives pod deaths. App stateless = cattle. |
| S3 backend + DynamoDB lock? | S3 = remote tfstate (team-safe); DynamoDB = lock (ek waqt ek apply). |
CI me github.sha kyun? |
Immutable, traceable. Rollback exact (git revert → purana SHA → Argo same deploy). latest = ambiguous. |
| CI infinite loop kaise rokein? | paths: ['app/**'] filter + [skip ci] tag — dono layers. |
Argo selfHeal: true? |
Manual kubectl change detect karke ~3 min me Git state restore. |
type: LoadBalancer self-managed pe pending kyun? |
No cloud-controller-manager → koi ELB provision nahi. Fix: NodePort (ya MetalLB). |
publicly_accessible = false RDS benefit? |
RDS ka public IP nahi; internet se unreachable even if SG galat. Defense-in-depth. |
| Rollback kaise? | git revert <bad-commit> → Argo new SHA detect → rolling update to old image. No manual kubectl. |
terraform destroy VPC pe hang — cause? |
K8s LoadBalancer ne AWS ELB banaya jo TF track nahi karta. Fix: K8s resources pehle delete. |
| kubeadm idempotency Ansible me? | args: creates: /etc/kubernetes/admin.conf — file exist = task skip. |
| Setup loop vs delivery loop? | Setup = terraform+ansible (ek baar, slow). Delivery = push→CI→Argo (har push, automated). |
| Self-managed vs EKS — honest jawab? | Self-managed = internals seekhe (etcd, kubeadm, CNI). Prod me EKS — control-plane HA AWS manage karta. |
CAP2 — Capstone: MicroShop¶
Chapter: 13-capstone-microshop.md
| Sawaal (Front) | Jawab (Back) |
|---|---|
| Inter-service call kaise? | order-api → http://catalog-api:8000 via K8s Service DNS; CoreDNS short name resolve same namespace me. |
| CoreDNS FQDN cross-namespace kyun? | Short name sirf same namespace. Alag namespace: catalog-api.default.svc.cluster.local. |
| Matrix CI 3 services parallel? | strategy.matrix.service: [catalog-api, order-api, frontend] — 3 runners parallel; ~90s vs ~4 min sequential. |
needs: update-manifests me kya guarantee? |
Sirf tab chale jab SAARI matrix jobs pass — partial manifest update se inter-service contract break rokta. |
| Cache-aside (Redis)? | Read → Redis check; miss → Postgres; result Redis me TTL pe store. Write on miss only. |
| Redis pod kill = cache data? | Sab wipe (no PVC). App next miss pe Postgres fallback. Lab ke liye acceptable. |
| App-of-apps kab better? | Multiple teams independent deploy — per-service rollback/sync/RBAC chahiye tab. |
for_each toset(...) se? |
Set me listed ECR repos bante; ek string add = ek naya repo, baaki unchanged. |
| Star-moment test command? | kubectl exec -it deploy/order-api -- curl http://catalog-api:8000/products/p1 — response = inter-service DNS kaam kar raha. |
| Redis fail pe try/except? | DB se serve (fallback), cache miss treat. Users ko 500 nahi. |
| Per-service Golden Signals kyun? | Frontend slow = catalog lag ya frontend bug? Per-service latency/error bina root cause guess. |
| catalog-api order-api se pehle deploy kyun? | order-api startup pe catalog call karta; catalog ready nahi to order readiness fail. |
Deck khatam. Roz ek module + ⭐-wale purane cards. 30 din me → 09 ka blank-page test do: 2 loops + 8 bridges + 5 threads bina dekhe.
Back to 00-INDEX · Full glossary + reflex tables → 16-reference-appendix