| When I was a dev lead at a previous company, I was tasked with leading two initiatives. I had designed everything out, received approval, and then we merged with another company and decided to “move to the cloud”. I didn’t know the first thing about AWS and neither did anyone else - including the infrastructure guys. So we hired some “consultants”. I did a PowerPoint slide of my architecture with the consultants for thier guidance for best practices. Basically they just set up some VMs, security groups, a VPC etc and they treated AWS like an overpriced colo. I spent the next few months designing and architecting everything like I would have done on prem: - 7 servers for Hashicorp’s Consul (1 in Dev,QA, UAT, and a cluster in production). They were used for configuration, service discovery, internal load balancing (with Fabio), and health checks. - 7 servers for Nomad (same as above) used for orchestration and to schedule jobs. Think Kubernetes but with the flexibility of using raw executables and not just Docker containers. - 7 servers for Mongo (same as above) - 12 always running app servers. - 2 build servers orchestrated with Visual Studio Online and local build agents. Of course all of these machines had Microsoft agents (?) on them for deployments, Consul agents, and the app servers had Nomad agents. This would have all been perfectly well architected for an on prem environment. But anyone who knows anything about AWS (and I didn’t then) would know that’s a dumb, overpriced, hard to maintain design and we weren’t taking advantsge of AWS services. If I were doing that now. I would: - get rid of all of the Consul servers and use AWS’s Parameter Store for configuration. Use internal AWS application load balancers and route 53 for services along with autoscaling and health checks. - instead of Nomad, I would have used CloudWatch Events and done a combination of lambda, step functions and Docker. That would have cut out 7 more servers. - I would have used AWS CodeBuild that basically let’s you use either prebuilt Docker containers for builds or create your own. Cutting down on 2 more servers. - Today, I would use AWS’s hosted ElasticSearch solution instead of Mongo. But given the needs of the project, I would have used Mlab’s (?) managed offerings. Of course I would use CloudFormation to manage all of this including the configuration key/values instead of my own bespoked app to source control configuration changes. |