Saad khan

@saad_devops
21 Followers
206 Following
58 Posts
Linux | Cloud | DevOps | Security • Automation
Github :- https://github.com/saadcnx
Twitter :- https://x.com/saad__devops
Linkedin :- https://www.linkedin.com/in/saad-khan-sysops/
Medium :- https://medium.com/@saadcnx
@slientguylw221 that great

REST isn't dead. But for internal K8s traffic? gRPC wins.

HTTP/1.1 problems:
- Head-of-line blocking
- Connection overhead
- JSON parsing tax

HTTP/2 + gRPC:
- True multiplexing
- ProtoBuf binary (5x smaller)
- 4 streaming patterns

K8s control plane already uses gRPC.

Read more: https://medium.com/@saadcnx/beyond-rest-why-modern-kubernetes-clusters-are-migrating-to-grpc-and-http-2-15d6c48b6641

#Kubernetes #gRPC #DevOps

Beyond REST: Why Modern Kubernetes Clusters are Migrating to gRPC and HTTP/2

If you are building microservices in Kubernetes today and still relying solely on traditional REST over HTTP/1.1 for internal…

Medium

Will AI kill Kubernetes? 🤖❓

NO.

CNCF ecosystem = $200B+ investment

Security, monitoring, service mesh, tooling – ALL built around K8s.

Too big. Too established. Too valuable.

AI will RUN ON K8s, not replace it.

Kubernetes is the operating system of the cloud era. Here to stay. 🏗️

#Kubernetes #GenAI #CloudNative

K8s QoS Classes Explained ⚖️💀

OOM Killer order:

BestEffort (no requests/limits) → First sacrifice
Burstable (requests < limits) → Second
Guaranteed (requests = limits) → Last

Root cause: No requests = scheduler overcommits = node runs out of RAM

Fix:

Set requests = limits

Use LimitRange

Reserve memory for OS (--system-reserved)

medium post :- https://medium.com/@saadcnx/the-silent-app-killer-how-kubernetes-qos-classes-can-secretly-wipe-out-your-nodes-919edae91712

#Kubernetes #DevOps #SRE

The Silent App Killer: How Kubernetes QoS Classes Can Secretly Wipe Out Your Nodes

Every very experienced DevOps engineer has faced it at least once: The infrastructure looks healthy, CPU utilization is fine, resources…

Medium

K8s 5-Minute Timeout Trap ⏰💥

Node dies? K8s waits 300 seconds before evicting pods.

Your app = dead for 5 minutes.

Fix:

tolerationSeconds: 15
Downtime: 5 min → 15 sec

For spot instances:

AWS: Node Termination Handler
GKE: Graceful Node Shutdown (built-in)

Don't let defaults kill your SLA!

Check out my medium post for more details :- https://medium.com/@saadcnx/stop-letting-kubernetes-kill-your-app-the-5-minute-timeout-trap-bc1e344f4c75

#Kubernetes #DevOps #SRE #K8s

🚨 Stop Letting Kubernetes Kill Your App: The 5-Minute Timeout Trap

If you are running Kubernetes in production and haven’t touched your default pod eviction settings, you are sitting on a ticking time bomb.

Medium

Why is GCP different? 🤔

Live Migration: Google patches hardware and updates the host without rebooting your VM. Zero downtime maintenance.

Shared Core: Gmail & YouTube run on the same infrastructure you rent in GCP. Enterprise-grade security by default.

Cloud Service Models simplified:

-IaaS (Infrastructure): Rent the kitchen. Cook yourself. (VMs)

-PaaS (Platform): Kitchen + Chef provided. Just bring the recipe. (App Engine)

-SaaS (Software): Full meal delivered. Just eat. (Gmail)
Responsibility shifts up the stack!

#GCP

5 Pillars of Cloud Computing you must know:

- On-Demand Self-Service (Automation)
- Broad Network Access (Any device)
- Resource Pooling (Carpooling for servers!)
- Rapid Elasticity (Grow/Shrink instantly)
- Measured Service (Pay only for what you use)