CNCG Chandigarh Meetup - Sustainability edition

1. 9:00 AM - 10:00 AM IST

   Welcome and Registration

   in-person

   Enjoy Networking with professionals.

2. 10:00 AM - 10:30 AM IST

   Your AI Agent Is a Distributed System: Operating Multi-Agent Workflows on Kubernetes

   in-person

   Agentic AI is surprisingly easy to demo. Operating it reliably is a very different problem.

   In one production platform, what started as an AI workflow evolved into a distributed system involving multiple dependent agents, FastAPI services, Elasticsearch-backed data services, execution workers, model endpoints, and workloads running on Kubernetes.

   At that point, the difficult questions were no longer just about prompts or models:

   Where should state and persistence live? What happens when one step in a multi-agent workflow fails? How do you separate agent logic from execution and data layers? How do you observe a request that crosses several services and agent calls? What should Kubernetes manage, and what should remain an application concern? How do you debug a workflow that is technically "running" but not behaving correctly?

   This session walks through the architecture and engineering decisions behind operating a multi-agent system on Kubernetes, including service boundaries, failure handling, deployment patterns, and observability using Prometheus, Grafana, and application-level tracing.

   Rather than focusing on a particular AI framework, the session treats agents for what they eventually become in production: distributed workloads that need the same reliability, visibility, and operational discipline as any other cloud-native system.

   The architecture and examples will be anonymized and will not contain client-sensitive information.

3. 10:30 AM - 11:00 AM IST

   eBPF at Scale: From Observing One Process to 50 Kubernetes Nodes

   in-person

   eBPF is often introduced through small demos: attach a program, observe a syscall, inspect a packet, and suddenly the Linux kernel feels much less mysterious.

   But what happens when that same idea has to operate across a real Kubernetes environment with dozens of nodes, hundreds of workloads, and a continuously growing stream of telemetry?

   In this session, we will start from the fundamentals of eBPF and build the mental model from the ground up. We will look at where eBPF programs run, how they attach to kernel and userspace events, how maps and ring buffers move data back to userspace, and how these primitives make zero-code observability possible.

   Then we will move from theory to a live engineering demo.

   We will start with an uninstrumented application and use OpenTelemetry eBPF Instrumentation (OBI) to automatically generate telemetry without adding an OpenTelemetry SDK to the application.

   From there, we will scale the same architecture to a Kubernetes cluster with approximately 50 Linux worker nodes and explore the engineering problems that appear at scale: per-node eBPF agents, Kubernetes metadata discovery, API server pressure, telemetry volume, OpenTelemetry Collector scaling, load balancing, and stateful processing such as tail sampling.

   The goal of the session is not just to show that eBPF works, but to understand what changes when eBPF-based observability moves from a laptop demo to a distributed cloud-native system.

   Expect Linux internals, Kubernetes, OpenTelemetry, eBPF, live debugging, and a real scaling story.

   SPEAKERS
4. 11:00 AM - 11:30 AM IST

   Your Observability Bill Is Bigger Than Your Cluster Bill

   in-person

   Plenty of teams find out what telemetry costs when someone in finance asks why monitoring is more expensive than the cluster it watches. The usual fix is to sample harder and hope nothing important gets dropped.

   This session looks at where telemetry volume actually comes from, and what to cut without going blind. We go through the three biggest sources of waste: unbounded label cardinality in Prometheus metrics, debug logs shipped to hot storage and kept forever, and traces sampled at the wrong layer. For each one: how to find out if you have it, what to change, and what it typically saves.

   Then the part that fits Sustainability Month. Every series and span you retain is storage, CPU and network you pay for twice, in money and in energy. Dropping data nobody has ever queried is one of the few efficiency wins with no trade-off attached, and it is usually sitting in plain sight.

   We close with OpenTelemetry Collector patterns that reduce volume at the edge instead of at the vendor: tail sampling, filtering and aggregation before export, and how to keep that configuration portable so the next cost conversation does not mean re-instrumenting everything.

   You leave with a checklist for auditing your own telemetry spend and the specific collector changes that move the needle first.

5. 11:30 AM - 12:00 PM IST

   Tea Break

   in-person

   Enjoy Networking over Cup of Coffee.

6. 12:00 PM - 12:30 PM IST

   From Kubernetes Resources to Energy: Finding Waste in Your Cluster

   in-person

   Kubernetes workloads are often over-provisioned, leading to underutilised resources and unnecessary infrastructure consumption. This talk explores how we can identify this waste and connect Kubernetes resource utilisation with energy consumption. Using Kubernetes metrics, Prometheus, Grafana, and Kepler, we'll measure workloads, identify inefficiencies, optimise resource allocation and scaling, and compare the results before and after optimisation.

   SPEAKERS
7. 12:30 PM - 12:45 PM IST

   TBD

   in-person

   By Piyush!

8. 12:45 PM - 1:00 PM IST

   A practical approach to sustainable cloud-native infrastructure

   in-person

   This session explores how cloud-native infrastructure can become more sustainable by making compute resources more measurable, optimizable, and efficiently utilized. We will look at the hidden resource and energy costs of modern AI and Kubernetes workloads and introduce a practical cloud-native sustainability approach through three stages: Measure, Optimize, and Utilize. The session will cover Kepler for energy-related workload visibility, KubeGreen for reducing unnecessary workloads during defined low-demand periods, and vCluster + HAMi for better infrastructure and accelerator utilization through sharing and isolation. The session will also connect these technologies to real-world use cases and discuss how sustainability improvements can be measured through resource utilization and energy-related metrics.

   SPEAKERS
9. 1:00 PM - 1:15 PM IST

   Declarative Validation: How Kubernetes is streamlining API reviewing and onboarding new Contributors

   in-person

No sessions match the selected labels.