@kodekloud: Kubernetes Tricky Interview Question! 11 pods drain off the node without a problem. The twelfth won't budge, and the eviction API keeps saying no. That is the trap in this question, and the answer is a pod disruption budget. Kubernetes runs your apps as pods across a set of servers called nodes, and it will not let a service drop below the number of copies you set. With only one copy left, evicting it breaks that floor, so Kubernetes blocks it on purpose. Reach for a force delete and a stateful app like a database can end up running twice at once, which quietly corrupts data. Would you force delete it, or find the safe fix? Comment below. #kubernetes #kubectl #devops #k8s #poddisruptionbudget #sre #cloudnative #containers #kubernetestips #devopsinterview #techtok #platformengineering
KodeKloud
Region: SG
Thursday 20 August 2026 12:45:52 GMT
Music
Download
Comments
Peter_Daniel :
Built a nomination form (free-text name entry, no validation/autocomplete) for a church event — 7.6k+ submissions came in, and the goal is picking the top 3 nominees per category. Problem: same person shows up as multiple spelling/name variants ("Similoluwa," "Simi Dahunsi," "Dahunsi Similoluwa") splitting their real vote count across entries, which distorts who's actually top 3.
Worked through the fix with AI (Claude): raw data stays untouched, merge decisions live in a separate mapping table (non-destructive + fully reversible). Fuzzy-matching only suggests likely-same-person clusters, never auto-merges — I pick which specific names in a cluster actually combine, avoiding "chaining" errors where unrelated names get lumped together. Short/ambiguous names (≤4 chars, like "Simi") never get auto-matched at all — routed to manual review since that's genuinely too risky to guess algorithmically. Basically human-in-the-loop record linkage. But I am very Curious and open to learn how you'd approach this at this scale 👀
2026-08-21 06:05:51
0
Oscar :
The solution is to scale down the replica to zero and then drain it
2026-08-22 17:54:19
0
Zac Nixon :
You turned an implicit problem (more than one db pod) into an explicit problem
2026-08-20 15:23:34
2
r_lex_99 :
PDBs are a must when using kubernetes for runners. Because runners like those in ADO cannot have their memory drained since they perform processes like checking out. Other things this might effect is memory for when you are running reports. It took us nearly a year to figure it out. The only other way is to use batch nodes which don’t drain or only drain OOO
2026-08-21 22:43:39
0
someonefromthere1 :
how I can mster this.. it's not mentioned in the course
2026-08-22 21:41:25
0
Patrik Los Pappos :
If you run databases in K8s you are asking for trouble. Think twice (or more) before running stateful thingies in K8s for mission critical applications.
2026-08-20 13:22:29
14
Samodharitha Nawamma :
❤️❤️❤️
2026-08-26 04:42:41
0
To see more videos from user @kodekloud, please go to the Tikwm
homepage.