MongoDB World 2019: Mastering MongoDB in Kubernetes
1 like430 views
The document discusses the integration and benefits of using MongoDB in Kubernetes environments, particularly for Amadeus, highlighting the need for efficient management of databases at scale. It emphasizes the capabilities of Kubernetes operators to streamline operations, automate tasks, and enhance responsiveness in handling large volumes of data for travel-related services. Additionally, it details the architecture and functionalities of the MongoDB Enterprise Operator for Kubernetes, showcasing its impact on improving database deployment and management processes.
MongoDB World 2019: Mastering MongoDB in Kubernetes
- 1. ©AmadeusITGroupanditsaffiliatesandsubsidiaries Mastering MongoDB in Kubernetes - MongoDB Enterprise Operator for Kubernetes Paul Hubert, Arkadiusz Borucki Amadeus
- 2. 2 ©AmadeusITGroupanditsaffiliatesandsubsidiaries Back in the day at a travel agency… FRA IST SLC DEN SIN PHL YYZ YUL PHX TLV PEK HND SYD ICN DEL CLT BOM DUB SEA BRU ORD CTU LAX MNL MSP CGK VKO CDG BCN EWR RCO CAN FCO MCO KUL SAN DFW GRU HKG LGW AMS
- 3. Now, you have the personalized travel agent in your hand … ©AmadeusITGroupanditsaffiliatesandsubsidiaries Personalized Tailor-made experiences unique to the traveler’s preferences A smooth, door-to- door experience Connected 3
- 4. 4 ©AmadeusITGroupanditsaffiliatesandsubsidiaries Amadeus: Evolving demands(latency & volume) - Shopping experience! 15.2M availability request per sec (peak) 580 availability request per sec (peak) 40% Growth / Year year2011 2019 AvailabilityRequests 600K+ availability requests for one booking! (peak) 25K avail. requests for one booking Shopping services Response Time impact Business impact Amadeus Fareportal + 1000 ms -20% in revenue Amadeus Ctrip + 700 ms -30% in revenue Source 1. http://www.marketingprofs.com/charts/2014/25719/the-website-experience-consumers-value-most-infographic
- 5. 5 ©AmadeusITGroupanditsaffiliatesandsubsidiaries Amadeus: Need for Hybrid IT Reliable Stateful Scale up Long cycles Price for performance System of Record Agile Stateless Scale out Short cycles Customer experienceSystem of Engagem ent 630+ Million Total bookings processed in 2017 300+ Trillion Availability requests processed in 2017 6%p.a. 40%p.a. Challenges in harmonizing the two system types
- 7. Agenda 7 35 minutes _Amadeus MongoDB infrastructure _What are Kubernetes operators _Introducing the MongoDB Enterprise Operator _Demo: Deploying a Sharded Cluster _Software Defined Infrastructure for Kubernetes
- 8. 8 ©AmadeusITGroupanditsaffiliatesandsubsidiaries MongoDB deployment @ huge scale! 8 Ops Managers 4 Multi Data Centers Cluster Size 130TB Micro sharding 68 MongoD processes Cloud 6500 GCE, OVH, OpenStack 72 Nodes per cluster
- 9. MongoDB deployment @ huge scale! 9 _ Instant searching - flight recommendations data 3 huge MongoDB clusters 75TB, 80TB, 130 TB with 2 ms (avg) response time _ MongoDB for real-time analytics – 124 shards (micro sharding) agregation framework _ Data agregation from multiple sources into a central cluster - A Single View application _ Apache Spark on MongoDB – fraud detection - run machine learning on MongoDB _ MongoDB Amadeus Payments - payment flows _ Dashboards, scheduling tools, change viewers, applications error viewers, and more
- 10. 10 ©AmadeusITGroupanditsaffiliatesandsubsidiaries Kubernetes - how to operate smarter _ Operational overhead, effort, human resources, time _ Why Kubernetes ? Why Operator ? _ As an Operations / DevOps engineer - Fewer manual tasks, automation provided by vendor _ As a Developer - Your company can provide MongoDB as a Service Easy access to production-ready MongoDB with a single click or command _ Build consistent and identical environments! _ Speed!
- 12. ©AmadeusITGroupanditsaffiliatesandsubsidiaries Why Kubernetes Operators are a game changer 12 _ You can extend the Kubernetes API with CustomResourceDefinitions _ Enables developers to add new/repleace old functionalities _ Custom Controllers allow you to write an application to fully manage another like statefull MongoDB databases farm _ Help you focus on a desired configuration state via Custom Resource, not the details of manual deployment/configuration _ You can work on service logic rather than spending weeks on environment automation _ Operators will change/simplify the way how we manage databases / repleace already created automation Operator represents human operational knowledge in sortware to reliably manage an application Kubernetes 1.7 add Custom Controllers feature An Operator is set of app-specific custom controllers. Controllers have direct access to Kubernetes API
- 13. Operator – teach Kubernetes about MongoDB! 13 Declarative definition of what MongoDB services you want An Operator is an abstraction for deploying non-trivial applications on Kubernetes Easy to scale up / scale down _ Deploying and scaling stateless apps like nginx is easy _ Deploying stateful applications - like databases - is more complicated. We’ll need more objects to be created and managed together, and there’s more to do to update versions, scale or organize backups _ Operators are a way to add application-specific awareness to Kubernetes, so you can automate these complex tasks while taking advantage of Kubernetes Orchestration.
- 14. 14 MongoDB Enterprise Operator for Kubernetes Ops Manager features on K8s cluster! K8s API and Ops Manager integration Controller, extends the Kubernetes API Operator for Kubernetes supports: _ pod’s backup, automation, monitoring _ provisioning StatefulSets for replica set _ creates deployment in Ops Manager project _ automates cluster provisioning, scalability, auto recovery, upgrades _ automates common tasks, to achieve the desired state _ security - tls _ many more features are coming!
- 20. 20 How is Operator build ? Application specific controllers Custom Resource _ Custom Resource allow us to define your own resources in the Kubernetes API using CustomResourceDefinition _ Additinal resource is available via Kubernets API _ Perform CRUD operations upon them _ Interact with threm via kubectl --- apiVersion: apiextensions.k8s.io/v1beta1 kind: CustomResourceDefinition metadata: name: mongodb.mongodb.com ............. oneOf: - properties: # Standalone schema - properties: # ReplicaSet schema - properties: # ShardedCluster schema components
- 21. Custom resources 21 --- apiVersion: mongodb.com/v1 kind: MongoDB metadata: name: my-sharded-cluster spec: type: ShardedCluster shardCount: 2 mongodsPerShardCount: 3 mongosCount: 2 configServerCount: 3 version: 4.0.6 --- apiVersion: mongodb.com/v1 kind: MongoDB metadata: name: my-replica-set spec: members: 3 version: 4.0.6 service: my-service
- 22. ©AmadeusITGroupanditsaffiliatesandsubsidiaries Controllers 22 _ Observe the actual state _ Analyse difference from desired state _ Take action _ Drive the actual state towards the desired stare _ Repeat! Observe Analyse Act my-replica set has 2 running pods Differs from desired state of 3 replicas (pods) Action: - Start a pod - Add pod to the cluster - Waiting for replication
- 25. 25 Software Defined for Kubernetes - Procurement scenario High level building block perspective Availability Zone 3 Availability Zone 2 Blue print 1 mongo cluster 1 shard 2 secondaries Availability Zone 1 Infrastructure Pod Compute tenant Storage tenant VolumeContainerized App K8s Pod 10.10.10.1
- 26. 26 Software Defined for Kubernetes – User Expectations Public Clouds are setting the standards aaS • Need to manage pre-provisioned pool of resources • All is software driven (SDS, SDN, …) Contract • Tenant • Performance (# of io and throughput per GB) Compliancy • PCI, PII, … Resiliency • AZ, Region, hybrid-Cloud, Multi-Cloud
- 27. 27 Software Defined for Kubernetes – Infrastructure Perspective Necessary enablers Network • Ethernet is default transport • Less over-commitment at the spine and super spine level • Bandwidth increase for East-West traffic • Virtualized networks / security zones • Moving away from VLAN towards BGP Storage • New low latency protocols like NVMeoF • S3 increases it’s footprint thanks to its REST API and IAM support
- 28. 28 Software Defined for Kubernetes – Stack evolution The PaaS is driving the standard towards simplification Server Storage (appliance) Network Nova Cinder Neutron K8sPaaS IaaS (OpenStack) HW lightweight Linux Storage (SDS) Network (SDN) HW PaaS K8s Docker container CSI CNI Classic Stack Stack Evolution (k8s on bare metal) Coordination and pool management
- 29. ©AmadeusITGroupanditsaffiliatesandsubsidiaries Let’s shape the future of travel together NoSQL Private Cloud Compliance & MonitoringNetwork Storage -V -T