Deepanshu Sharma

DM me #DevSecOps folks !! ( #immediate ) Strong fundamentals with at least 2.5 -4 years implementation exp : Gradle, GIT, MS Azure (ARM templates, AKS, GitHub Adv Security( #ghas), Application gateway, #Azure AD)Frameworks: #Kubernetes, Gradle, #GitHubActions , CICD set up azure deployment, terraform script understanding .

3

Kubernetes Autoscaling: HPA, VPA, CA, and Using Them Effectively #WorkSmartWithK8s #kubernetes #autoscaling #hpa #vpa #ca #bestpractices https://lnkd.in/eZfhtUzm

11

𝗦𝘁𝗼𝗽 𝘁𝗵𝗲 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 𝗖𝗼𝗻𝗳𝘂𝘀𝗶𝗼𝗻! 𝗔 𝗤𝘂𝗶𝗰𝗸 𝗚𝘂𝗶𝗱𝗲 𝘁𝗼 𝗗𝗶𝗳𝗳𝗲𝗿𝗲𝗻𝘁 𝗖𝗜/𝗖𝗗 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 𝗧𝘆𝗽𝗲𝘀 🚀 ✔️ Infra Pipeline ✔️ App Pipeline ✔️ Mobile SDK Pipeline ✔️ Data Pipeline ✔️ IOT Pipeline ✔️ Firmware Pipeline 👉 𝗜𝗻𝗳𝗿𝗮 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 This pipeline is used for automated creation and provisioning of infrastructure resources with IAC (Infrastructure-as-Code) tools like Terraform, Pulumi, Cloudformation etc. the resources are created in Cloud providers. It helps to maintain uniformity of resources created across different environments like dev, qa, uat, prod etc. 👉 𝗔𝗽𝗽 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 Used to build, test and package the application code for Frontend and Backend apps. The programming languages include Java, Python, ReactJs, NodeJs, AngularJs etc. The packaged code is known as artifacts (jar, war, docker images etc) and they are stored and maintained in Artifact registries like Jfrog, Nexus, ECR etc. 👉 𝗠𝗼𝗯𝗶𝗹𝗲 𝗦𝗗𝗞 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 This is similar to app pipeline but the programming language and the testing tools differs. The programming languages include IOS, Android, Flutter etc. The packaged code is known as artifacts (.ipa, .apk files) and they are stored in package managers like CocoaPods for iOS and Gradle for Android. The mobile app is built on top of the functionalities provided by the SDKs. 👉 𝗗𝗮𝘁𝗮 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 The pipeline is used for receiving the data from various sources, transforming it and storing them into the databases and datalakes. This is created by connecting different services together like for example AWS Glue, Lambda, Step functions, Message brokers etc to achieve the consistency and reliability of data quality. The pipelines jobs use programming languages like Python, Java and R. 👉 𝗜𝗢𝗧 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 This is used to handle the collection and processing of data, events and messages generated by IOT (Internet of Things) devices. The IOT devices uses varying communication protocols and send high volumes of data in different formats. Similar to data pipelines, the IOT pipeline consists of combination of many services like for example IOT topics, gateways, lambda functions, message brokers etc. Usually the IOT pipelines acts as first interface to process the data in real-time or batches from IOT devices and send it to the data pipelines for transformation and storing to databases & datalake. 👉 𝗙𝗶𝗿𝗺𝘄𝗮𝗿𝗲 𝗣𝗶𝗽𝗲𝗹𝗶𝗻𝗲 This is used to build, test and package the firmware code for the embedded and IOT devices. The programming languages include C++, Embedded C, Rust etc. Most of the time the code has to be compiled on the specific Linux processor architecture (like ARM, x86) as per the requirement of the hardware device on which it would be flashed to run. Follow me for simplified learning and best practices on DevSecOps 🙂 Save for future reference. 🔁 Consider a Repost if this is useful. #devops #cicd #pipelines #terraform #awscommunity

16

2 Comments

Why do you manage CPU and Memory differently on a Kubernetes cluster? In K8s, it is good practice to set requests & limits for your applications. These values define the minimum and maximum amount of compute and memory resource that an application will need. So when our application needs memory, it's really simple, we can define some basic values to determine the request and the limit. But it's different for CPU, why? Let's assume there are 4 available cores. Application A has a request of 0.5 and a limit of 2. Application B has the same. This means that 25% of resources are reserved and each application can use up to 50% of the available resources. This seems sensible on the surface. They each get half at most. This means if both applications need to increase utilisation of the CPU, they should not interfere with each other. Now, as often happens, there is a sudden spike in traffic and Application A is suddenly in high demand. It scales to 2 cores, while Application B is sitting at 0.5 cores, but there is a problem. 2 Cores is not enough for Application A! It tries to scale further, but the limits have been set, while 1.5 cores sit idle on the server. It is for this reason that we don't set CPU limits on Kubernetes pods, because CPU is what is known as a compressible resource. In other words, it can be taken back after it has been allocated. Why is this different for memory? Simple. Memory is non-compressible, meaning after memory has been allocated to a process, it can not be reclaimed without first killing that process. This makes sense. If I am using 4GB memory, it means I have written to 4GB worth of space. If I stop writing to all 4GB at any given time, it is no indication that I no longer need it. Unlike CPU, that can be fluidly shared between processes. That's why CPU and Memory are managed differently in a Kubernetes cluster. It's also why the K8s monitoring feature in Coralogix doesn't just show the limits, requests and usage, but it also shows the **total available resource on all servers**, so that operators can ensure a shared pool of CPU resource is available between all applications. You can find out more at https://lnkd.in/eknwwjXc

17

Yahoo’s Kubernetes Journey from on-Prem to Multi-Cloud   🎯 Key Innovations:     - Migrated from on-prem infrastructure to a multi-cloud ecosystem with GCP and AWS.     - Developed ‘Omega’ interface simplifying app deployment to Kubernetes clusters.     - Introduced ‘Cadence’ tool for seamless cluster creation across multiple clouds.    💡 Notable Features:     - **Omega YAML**: A simplified deployment template ensuring quick app setups.     - **Cadence Tool**: A Golang-based tool for managing EKS and GKE workloads without deep cloud provider knowledge.     - **Node Manager Operator**: Automates Kubernetes version upgrades while maintaining workloads.    🛠️ Perfect for:     - Cloud Engineers     - Kubernetes Administrators     - DevOps Teams     - Platform Engineers     - Organizations transitioning to multi-cloud environments    ⚡️ Impact:     - Enhanced deployment efficiency by streamlining application setups.     - Achieved significant cost reduction for operations in cloud environments.     - Facilitated rapid responses to infrastructure changes and reduced upgrade times from months to weeks.    🔍 Preview of the Talk:     Nandhakumar Venkatachalam and Payal Patel from Yahoo discuss the company's transition from a traditional on-prem environment to a cutting-edge multi-cloud strategy using Kubernetes. They delve into the challenges faced, such as networking and security, while showcasing breakthrough tools like Omega and Cadence. The talk emphasizes how these innovations enhance user experiences and streamline operations. Participants will gain valuable insights into effective cloud strategies and operational efficiencies in Kubernetes environments. Watch the full session: https://lnkd.in/gcZS--rY

1

This blog by JFrog is must read for DevOps Engineers. ✔️Quality binaries come from smart systems that know how to manage and distribute them.👇

13

Introduction to KubeVirt, Virtual Machines on Kubernetes #WorkSmartWithK8s #kubernetes #kubevirt #containers #virtualmachines #differences #similiarites https://lnkd.in/eMsCxGjB

28

Simplify your Kubernetes networking challenges with Sumudu Liyanage at #KCDHyderabad! From IP address pools to pod communication, learn to optimize your clusters for better performance and security in this in-person session. #Kubernetes #DevOps #CloudComputing

102

How many down nodes can your application afford to run with?? ✅ Application are designed for resilience with replicas or active active or active passive set ups For example, let's take a multi-az cluster having the following size: 1. Say total 18 nodes distributed across 3 data center 2. Each DC has 6 nodes it's very well designed for resilience and HA Question is , how many nodes can go down and still your cluster is not saturated?? There is always a limit based on traffic at that point in time when the cluster will get overwhelmed with excessive traffic.. Even more worrisome is the cascade effect even a single node failure produces!! Two of the solution which I find (keeping aside auto scaling): ✅ Rate limiting at Load balancer but here how will you keep this dynamic as per the count of healthy node ✅ Using circuit breakers on each and every node but then how could you balance traffic even while just 1 node is down.. pls share your opinion and suggestions on this? particular how do you measure contingent capacity and bring it in action..

6

Enforce your policies as code on Kubernetes using GateKeeper(OPA) #WorkSmartWithK8s #kubernetes #opagatekeeper #policyascode #policies https://lnkd.in/enFnPy2a

6

How Kubernetes Evicts Pods: Understanding Resource Limits, Requests, and Priority #WorkSmartWithK8s #kubernetes #pods #podeviction #resourceelimits #resourcerequests https://lnkd.in/eEahnX2b

31

🚀 Understanding the Binpacking Problem in Kubernetes In Kubernetes, the binpacking problem refers to efficiently scheduling and allocating resources (like CPU and memory) to Pods, ensuring they are packed into nodes in a way that maximizes resource utilization while meeting constraints like resource limits, availability, and performance. Binpacking Problem in Kubernetes Context: The main objective is to place the pods on nodes such that resources (CPU, memory, etc.) are efficiently utilized, and the system remains stable. Here's how it works: 🔹 Pods as Items: Each pod requires a certain amount of resources. 🔹 Nodes as Bins: Nodes have a fixed amount of resources. 🔹 Objective: Maximize resource usage without overcommitting any node. Key Concepts in Binpacking within Kubernetes: 🔸 Scheduler: Kubernetes uses a scheduler to decide where a pod should run, taking into account resource availability, constraints (node selectors, taints, tolerations), and affinity rules. 🔸 Resource Requests & Limits: Pods define the minimum and maximum resources needed. The scheduler uses these to prevent over- or under-utilization of resources. 🔸 Node and Pod Resource Allocation: The scheduler ensures that pods are placed in nodes without exceeding the available resources on each node. Efficient Binpacking Algorithms: Best Fit: Place pods on the node with the least available resources while meeting the pod’s needs. Worst Fit: Place pods on the node with the most resources left to leave space for future pods. First Fit: Place pods on the first available node. Example: Resource Requests & Limits: Pod 1: Requests 1 CPU, 2GB memory. Pod 2: Requests 2 CPU, 1GB memory. Pod 3: Requests 1 CPU, 1GB memory. Nodes: Node A: 4 CPUs, 8GB memory. Node B: 2 CPUs, 4GB memory. Scheduling Decision: Pod 1 and Pod 2 could be placed on Node A. Pod 3 could be placed on Node B. This ensures efficient use of resources without overcommitting any node. Challenges in Binpacking: Overcommitment: Pods may consume more resources than requested, starving other pods. Resource Fragmentation: Small amounts of available capacity may hinder efficient pod placement. Scalability: As clusters grow, the complexity of the binpacking problem increases. Optimizing Binpacking in Kubernetes: To optimize binpacking: Set proper resource requests and limits for all pods. Use Vertical Pod Autoscaling (VPA) and Horizontal Pod Autoscaling (HPA) to scale based on resource usage. Implement taints and tolerations to avoid overpacking specific nodes. Efficient binpacking ensures that your Kubernetes cluster is performing optimally, maximizing resource utilization and minimizing waste. #Binpacking #KubernetesScheduling #ResourceOptimization #CloudComputing #ContainerOrchestration #Kubernetes #DevOps #CloudInfrastructure #DistributedSystems #Microservices

6

Understanding CPU & Memory with the Kubernetes Vertical Pod Autoscaler #WorkSmartWithK8s #kubernetes #autoscaling #vpa #cpu #memory https://lnkd.in/gidUvXEx

3

✨ Engineering, specially deploying and managing applications at scale is a very big challenging task. 🤔 Managed Kubernetes - EKS, AKS and GKE only solves half of the problem of managing the control plane of Kubernetes. It's still the work of your Devops / SRE / Backend engineers to do all these - - Upgrade K8s and node versions frequently - Build, manage, update Helm Charts as needed - Maintain and monitor your application status, debug any node/infra/network failures - Build and maintain a full logging, monitoring stack Devops is costly, either hire experienced devops engineers, or make your backend engineers learn all this (developer time cost) 💰💰

61

2 Comments

15 straight-cut points to save AWS bills: 1) 𝐑𝐢𝐠𝐡𝐭-𝐬𝐢𝐳𝐞 𝐢𝐧𝐬𝐭𝐚𝐧𝐜𝐞𝐬: Match instance types to your workload needs to avoid over-provisioning and reduce compute costs. 2) 𝐔𝐬𝐞 𝐒𝐩𝐨𝐭 𝐈𝐧𝐬𝐭𝐚𝐧𝐜𝐞𝐬: Leverage AWS Spot Instances for non-critical workloads to save up to 90% on compute costs. 3) 𝐒𝐜𝐡𝐞𝐝𝐮𝐥𝐞 𝐢𝐧𝐬𝐭𝐚𝐧𝐜𝐞 𝐬𝐡𝐮𝐭𝐝𝐨𝐰𝐧𝐬: Use automation tools to turn off dev/test environments outside business hours. (if it's a non-critical 24*7 Resource) 4) 𝐔𝐬𝐞 𝐬𝐞𝐫𝐯𝐞𝐫𝐥𝐞𝐬𝐬 𝐬𝐞𝐫𝐯𝐢𝐜𝐞𝐬: Opt for AWS Lambda or Fargate to run code without provisioning servers, paying only for actual compute time used. 5) 𝐋𝐞𝐯𝐞𝐫𝐚𝐠𝐞 𝐒𝐚𝐯𝐢𝐧𝐠𝐬 𝐏𝐥𝐚𝐧𝐬 𝐨𝐫 𝐑𝐞𝐬𝐞𝐫𝐯𝐞𝐝 𝐈𝐧𝐬𝐭𝐚𝐧𝐜𝐞𝐬: Commit to 1- or 3-year usage plans for predictable workloads to reduce costs. 6) 𝐔𝐬𝐞 𝐒3 𝐥𝐢𝐟𝐞𝐜𝐲𝐜𝐥𝐞 𝐩𝐨𝐥𝐢𝐜𝐢𝐞𝐬: Move infrequently accessed data to cheaper storage classes like S3 Glacier. 7) 𝐔𝐬𝐞 𝐏𝐫𝐢𝐯𝐚𝐭𝐞 𝐈𝐏𝐬 𝐟𝐨𝐫 𝐢𝐧𝐭𝐞𝐫-𝐬𝐞𝐫𝐯𝐢𝐜𝐞 𝐜𝐨𝐦𝐦𝐮𝐧𝐢𝐜𝐚𝐭𝐢𝐨𝐧: Ensure services communicate over private IPs within the same VPC to avoid public data transfer costs. 8) 𝐎𝐩𝐭𝐢𝐦𝐢𝐳𝐞 𝐝𝐚𝐭𝐚 𝐭𝐫𝐚𝐧𝐬𝐟𝐞𝐫: Use AWS PrivateLink or Direct Connect for internal data transfers to avoid public data transfer charges. 9) 𝐔𝐬𝐞 𝐜𝐚𝐜𝐡𝐢𝐧𝐠: Implement caching with CloudFront or ElastiCache to reduce backend compute and data transfer costs. 10) 𝐌𝐨𝐧𝐢𝐭𝐨𝐫 𝐮𝐧𝐮𝐬𝐞𝐝 𝐫𝐞𝐬𝐨𝐮𝐫𝐜𝐞𝐬: Use cost management tools to find and terminate idle resources like unused EBS volumes or Elastic IPs. 11) 𝐈𝐦𝐩𝐥𝐞𝐦𝐞𝐧𝐭 𝐚𝐮𝐭𝐨-𝐬𝐜𝐚𝐥𝐢𝐧𝐠: Automatically scale your instances to handle traffic spikes and downscale during off-peak times. 12) 𝐂𝐨𝐧𝐬𝐨𝐥𝐢𝐝𝐚𝐭𝐞 𝐂𝐥𝐨𝐮𝐝𝐖𝐚𝐭𝐜𝐡 𝐥𝐨𝐠𝐬: Use CloudWatch log groups efficiently to avoid unnecessary data retention costs. 13) 𝐋𝐞𝐯𝐞𝐫𝐚𝐠𝐞 𝐀𝐑𝐌-𝐛𝐚𝐬𝐞𝐝 𝐢𝐧𝐬𝐭𝐚𝐧𝐜𝐞𝐬: Use AWS Graviton-based instances for workloads that support ARM architecture for better cost-performance. 14) 𝐌𝐢𝐧𝐢𝐦𝐢𝐳𝐞 𝐜𝐫𝐨𝐬𝐬-𝐫𝐞𝐠𝐢𝐨𝐧 𝐝𝐚𝐭𝐚 𝐭𝐫𝐚𝐧𝐬𝐟𝐞𝐫𝐬: Keep services in the same AWS region to avoid inter-region data transfer fees. (to achieve high availability, you can keep minimum resources running in other AZs or run in standby mode) 15) 𝐂𝐨𝐦𝐩𝐫𝐞𝐬𝐬 𝐚𝐧𝐝 𝐨𝐩𝐭𝐢𝐦𝐢𝐳𝐞 𝐝𝐚𝐭𝐚: Reduce data transfer sizes by compressing files before sending between services. Did I miss any important points? please add in the comment section below! Cheers, Sandip Das

308

37 Comments

Solving CPU throttling issue in Golang applications before hitting the CPU limit in Kubernetes. #WorkSmartWithK8s #kubernetes #cputhrottling #cpulimit #golang #goroutines https://lnkd.in/gm34nFVN

10