KubeDay Colombia 2024

As organizations increasingly utilize container images for their production workloads, securing them becomes more critical. What happens if the vulnerability scanners used to scan these images contain vulnerabilities themselves? This talk will explore the hidden risks associated with container vulnerability scanning tools and how to mitigate these potential threats. Container vulnerability scanning tools, like any other software, can have vulnerabilities. If not regularly updated and patched, they can become targets for attackers who exploit these weaknesses to compromise critical systems. Also, these tools often require elevated privileges, leading to misuse or unauthorized access, posing significant security risks. After this session, you'll learn: - False positives and false negatives - Identifying dependency issues - Strategies for isolation and segmentation - Secure configurations - Managing permissions and access - Applying defense-in-depth strategies

We will embark on a journey to unravel the world of Kubernetes resource management, shedding light on foundational primitives and uncovering best practices and real-world scenarios over a 90-minute lab session. We'll demonstrate how requests play a pivotal role in Kubernetes scheduling. We'll delve into the pod Quality of Service (QoS) and its importance in a real-world scenario of node pressure. We will determine pod usage over key cAdvisor metrics, pinpointing a typical situation of a pod being throttled and impacting performance. We will deploy Resource Quotas and Limit Ranges while explaining the lessons learned in multi-tenancy on large Kubernetes deployments. By the end of this session, the audience leaves the room with a clear understanding of how to avoid performance degradation, avoid outages, provide better multi-tenancy, and ultimately reduce cloud costs.

In this session, you will learn how Kubernetes, designed years before AI became the mainstream force that is today, brings a host of key features that make it a stable platform to support the ML model development lifecycle. From data ingestion to serving, K8s offers autoscaling, GPU device management, idempotent operations, Pod templating, and more; supporting the iterative and distributed nature of AI-powered app development. Using the Flyte open source orchestrator, you'll see how all these features can be leveraged today from an ML pipeline developed in Python.