Cloud Native Bangalore

Kubernetes Bangalore meetup

Capacity: 660
in-person
Event date
Oct 31, 26
08:30 AM - 01:00 PM IST
Registration is open until Oct 28, 2026 at 11:30 PM IST.
Location
Nokia L5 Manyata Business Park, India
About this event

The Kubernetes Bangalore Meetup is a premier community gathering for cloud-native enthusiasts, developers, SREs, and platform engineers in the city. While specific agendas for 31st October 2026 meetup , are typically announced via the CNCF Cloud Native Bangalore or Meetup.com chapters closer to the date, the following description captures the spirit and typical structure of these high-impact events.

Event Overview: Kubernetes Bangalore Meetup

**Event Venue: ** Nokia L5 Manyata Business Park, L5 and L6 Building, Outer Ring Rd, DadaMastan Layout, Manayata Tech Park, Nagavara, Bengaluru, Karnataka 560045 (https://maps.app.goo.gl/ELD1RwBijvxfhDRW8)

Join the cloud-native vanguard. Whether you are just beginning your journey with container orchestration or are a seasoned pro managing large-scale production clusters, this meetup is your gateway to the latest in the Kubernetes ecosystem.

Bangalore’s K8s community is one of the most vibrant in the world, regularly bringing together experts from industry-leading companies to share real-world case studies, technical deep-dives, and hands-on demos.

Agenda
  1. 8:30 AM - 9:20 AM IST

    Registration & Networking

    in-person
  2. 9:20 AM - 9:30 AM IST

    Welcome Note

    in-person
  3. 9:30 AM - 10:00 AM IST

    Journey to the Centre of Your Node

    in-person

    We write security policies in YAML, but what happens after we apply them? How does a rule travel from a manifest through the control plane and runtime-security agent to enforcement inside the Linux kernel? In this session, we’ll journey to the centre of a Kubernetes node using KubeArmor, a CNCF Sandbox project for cloud-native runtime security, as our guide. We’ll follow a policy from configuration through the relevant Kubernetes components and runtime-security agent, exploring how Linux security mechanisms and eBPF contribute to enforcement. We’ll go through each layer using simple explanations, making the underlying concepts accessible to Kubernetes developers, DevOps engineers, and anyone curious about what happens beneath the abstractions. Along the way, we’ll explore what happens when a workload attempts a restricted operation and how a high-level YAML policy connects to the mechanisms protecting workloads at the kernel level.

    SPEAKERS
  4. 10:00 AM - 10:30 AM IST

    Multi-Node LLM Inference at Scale

    in-person

    Deploying massive open-weights models like Llama 3 405B introduces a structural challenge for Kubernetes. The models simply cannot fit into a single GPU or even a single node. Traditional Kubernetes primitives like Deployments or standalone StatefulSets treat pods as independent replicas, making it difficult to manage tightly coupled, multi-node worker groups that require synchronized lifecycles, model sharding, and dedicated interconnect topologies.

    In this session, we present about the LeaderWorkerSet (LWS) API,a Kubernetes-native operator designed by SIG Apps and WG-Batch to treat groups of heterogeneous pods across multiple hosts as a single logical unit.

    In this talk, we’ll dive into how LWS solves multi-node inference and model replication challenges. We will walk through its two-level StatefulSet architecture, explore how it handles gang-scheduling and failure recovery, and demonstrate how it enables disaggregated serving by splitting the compute-heavy prefill stage from the memory-heavy decode stage across specialized hardware pools.

    Whether you are building internal AI platform tooling or optimizing model serving infrastructure on OpenShift and Kubernetes, this session will give you a practical blueprint for running distributed LLM workloads efficiently.

    SPEAKERS
  5. 10:30 AM - 11:00 AM IST

    API-Driven Cluster Lifecycle Across Heterogeneous Environments

    in-person
    SPEAKERS
  6. 11:00 AM - 11:30 AM IST

    Tea Break

    in-person
  7. 11:30 AM - 12:00 PM IST

    Right-Sized, Right Now: Smarter Node Autoscaling with Karpenter

    in-person

    Most Kubernetes clusters still scale through node groups that were sized ahead of time. That means you either overprovision and pay for idle capacity, or underprovision and wait minutes for new nodes. Karpenter takes a different approach. It's an open-source node lifecycle manager under Kubernetes SIG Autoscaling. It watches pods that can't be scheduled, launches nodes that fit what those pods actually need, and keeps shrinking and consolidating the cluster as demand changes.

    This talk explains how Karpenter works: NodePools, bin-packing, spot and on-demand selection, consolidation, and disruption budgets that keep production workloads safe. We'll also look at its provider model, which lets the same core run on different cloud providers. A short live demo shows nodes launching from zero and then consolidating. You'll leave knowing when Karpenter beats Cluster Autoscaler and how to adopt it safely.

    Key takeaways

    • How Karpenter differs in design from Cluster Autoscaler
    • How to configure NodePools, disruption budgets and consolidation safely
    • How the provider model makes Karpenter work across clouds and on-prem
    SPEAKERS
  8. 12:00 PM - 12:30 PM IST

    Stop Overpaying and Under-Guarding Your AI Agents

    in-person

    AI agents and inference workloads break the assumptions most platform teams built their infra around: usage is bursty, access patterns are unpredictable, and a misbehaving agent can have real blast radius. Most teams only discover this the hard way — an overprovisioned always-on deployment burning budget, or an agent with more access than it should ever have needed.

    This talk walks through three CNCF projects already in many teams' stacks — KEDA, OpenCost, and OPA/Kyverno — and how to combine them to solve two concrete problems: controlling cost for spiky AI/agent workloads, and enforcing safe boundaries around what an agent can actually touch. We'll look at real scaling configs, a live cost-attribution view, and an actual admission policy — no new tooling required, just using what CNCF already gives you, wired together with intent.

    Key takeaway: You don't need a custom platform to run AI agents safely and cheaply in prod — KEDA, OpenCost, and OPA/Kyverno, used together, already get you there.

    SPEAKERS
  9. 12:30 PM - 1:00 PM IST

    Understanding Kubernetes Internals: Adventures in optimizing static pods

    in-person

    Static pods in Kubernetes are a very important part of the cluster bootstrapping puzzle. In clusters bootstrapped with kubeadm, the control plane of the cluster itself is run as static pods. In this talk I will go through a specific edge that existed in the lifecycle of static pods and how I fixed it upstream. This talk would be useful for folks who like to learn more about how Kubernetes works under the hood, and also give an idea about how they can look for issues in the upstream Kubernetes repository and how they can work on them (setting up the local development environment using kind etc).

    Relevant PRs where I optimized static pod behavior

    The format of the talk is inspired by Madhav Jivrajani's talk Adventures in Optimizing the API Server. This talk is also part of a series of talks I've given about understanding the inner workings of Kubernetes. Some of the previous talks in this series are:

    SPEAKERS
  10. 1:00 PM IST

    Networking

    in-person
Organizers