DocSpring’s eight-year Convox retrospective shows how Heroku-style deployments coexist with substantial AWS responsibilities.
In its account, DocSpring describes moving from Heroku to AWS in 2018, when the product was still called FormAPI. AWS credits prompted the move; Convox supplied a deployment platform without requiring the team to build one first. The retrospective combines operational experience with practical guidance for teams considering a similar setup—not an announcement of a new platform.
The central trade-off is control versus responsibility. DocSpring kept a relatively simple application deployment workflow while gradually taking ownership of Terraform configuration, security controls and infrastructure changes. Its experience illustrates that simplifying deployments and eliminating infrastructure operations are different goals.
What Convox handles
Convox calls an installation of its platform a rack. According to DocSpring, the Apache-2.0-licensed rack runs in the user’s cloud account and can operate independently of the separate Convox Console, which provides a web interface and team workflows through free and paid plans.
DocSpring now runs US and EU production racks, plus staging, and helps maintain racks for some Enterprise customers. Its infrastructure has moved from Amazon ECS to Kubernetes. The account says Convox v2 used CloudFormation and ECS; the Kubernetes-based, Terraform-installed platform was announced in February 2020. On AWS, v3 uses Amazon EKS.
For a Rails application, a convox.yml manifest can describe web processes, a background worker and scheduled commands. DocSpring’s example specifies two web processes, a Sidekiq worker and a daily task at 03:00 UTC, with the database and Redis provisioned separately. Environment variables without defaults must be supplied through application configuration rather than stored as values in the manifest.
The everyday commands remain short:
convox deploy
convox logs
convox run web bundle exec rails console
As DocSpring explains, deployment builds containers, creates a release combining the build and application environment, and promotes it through a rolling deployment. New processes undergo health checks as old ones are replaced; startup or health-check failures can cause a rollback. That separation matters: application developers can use a consistent deployment interface even when the infrastructure beneath it has been customized.
Where ownership expands
DocSpring began managing rack Terraform directly to customize CloudWatch configuration and certificate DNS validation, and to manage database and storage infrastructure alongside the rack. Its examples use Convox 3.26.0, while its own deployment uses a customized 3.25.14 base. Those are distinct configurations, not evidence that the example exactly reproduces production.
The company also built Rack Gateway, an independent proxy for the Convox API accessible through its private Tailscale network. DocSpring says this adds security-key approval for agent commands and deployments, with a read-only database user as the default for command execution. These are DocSpring-specific additions supporting its SOC 2 program, not prerequisites for installing Convox or proof that the platform alone satisfies compliance requirements.
The operational boundary becomes clearest in maintenance. DocSpring remains responsible for upgrades, monitoring, backups, restore testing and its Terraform customizations. Its move from ingress-nginx to Contour required a custom zero-downtime transition, including waiting for persistent Cloudflare connections to stop reaching nginx before removing old routes.
It also added Envoy access-log collection and redaction and fixed two certificate-handling bugs.
For prospective users, this distinguishes routine deployment convenience from infrastructure lifecycle work. A platform can preserve familiar application commands while still requiring substantial engineering when networking, security or upstream components change. Maintaining a fork also creates work beyond contributing a fix upstream.
Understanding the AWS baseline
DocSpring estimates about $306 per month before usage charges and databases for an illustrative US East (N. Virginia) rack spanning three availability zones. The configuration includes an EKS control plane, three NAT gateways, a Network Load Balancer, six public IPv4 addresses, three t3.medium workers and their boot disks. About $210 is attributed to the control plane and fixed networking charges.
This is a configuration estimate, not DocSpring’s disclosed production bill or an all-in application price. Databases, backups, image storage, logging, traffic, load-balancer capacity and possible burst CPU charges are additional. Workers host both Convox services and applications, so their full capacity is not available to workloads. A separate staging rack adds its own baseline costs.
The cost structure makes rack utilization an important evaluation question. Multiple small applications can share a rack while capacity permits, spreading fixed costs; evaluating one application in isolation may therefore give a different economic picture from a shared deployment. DocSpring explicitly notes that a hosted platform can be a better starting point when control of the cloud account is unnecessary.
A bounded evaluation
DocSpring recommends beginning with Convox’s Apache httpd example in a sandbox AWS account, without a database or real application data. Its suggested setup uses the AWS CLI, Terraform and the Convox CLI, with a dedicated rack configured for private nodes, high availability and Contour.
The account estimates rack provisioning at 5–20 minutes, excluding additional setup, deployment and cleanup time. For the illustrated configuration, it puts four hours at roughly $2 in fixed charges before usage and recommends budgeting $10 for a short evaluation. Cleanup includes deleting the sample application, uninstalling the dedicated rack, waiting for deletion to finish and checking for retained storage and logs.
That experiment answers whether the daily workflow suits a team; it does not establish the full cost of operating production infrastructure. DocSpring’s longer experience supplies the complementary lesson: adopting an inspectable deployment platform can provide room to grow, but the ability to customize it comes with responsibility for the changes.
