What It Does
A trading application is not finished when it is deployed. Strategies change, data sources change, brokers change their interfaces, and the application itself is extended as the trader’s edge develops. Every one of those changes is a chance for the running system to drift away from what it is supposed to do.
The i26g Control Plane is being developed to validate, release, deploy, operate, and retire versions of the i26g Application, and to manage the VMs, networks, and volumes it runs on, within a boundary the infrastructure owner defines. Its desired state is declared in Git, and it keeps what is running aligned with that state continuously, not only at the moment something is deployed.
Like the application, it is in early development. The capabilities below describe the intended design.
Two Paths: Changes and Protection
The control plane works in a loop — observe, orient, decide, act — over the application and its environment: deployed versions, data freshness, broker connectivity, resource use, and the boundary’s contract. What it does with what it finds follows one of two paths.
Controlled changes. Releases, configuration changes, and infrastructure changes go through a declared, reviewable path: validated in an isolated environment, promoted in a controlled order, and checked again once they run.
Immediate protection. Some situations cannot wait for a review: stale data, a lost broker connection, or a breached health check. For those, the control plane acts without waiting for manual review, under rules approved in advance and declared in Git — pausing an affected strategy or blocking further deployments — and records what it did and why.
The Control Plane and the Application’s Risk Controls
The two divide the protection of live trading between them:
- The application’s risk component enforces the trader’s trading limits — position sizes, exposure, and losses — before and after each order.
- The control plane supervises operational health — whether data is fresh, the broker is reachable, and the environment meets its contract — and can pause strategies or block deployments when it is not.
Neither is the only safeguard. They have distinct responsibilities and may respond to some of the same failures.
From Boundary to Running Application
A trading operation cannot take its foundation on trust. Too little memory, a shared network, missing backups, or a clock that drifts can all break it quietly. So the i26g Control Plane does not assume its preconditions; it is given a boundary and checks it. Handing the Trading Platform a Boundary describes the thinking behind it.
- The boundary is set. Whoever runs the infrastructure creates a resource pool, a network zone and IP range, storage of a fixed size, and credentials scoped to exactly that boundary.
- The control plane is bootstrapped. One VM is created in the pool and booted with the i26g Control Plane.
- It takes over. It checks the boundary against i26g’s infrastructure contract — the resources, isolation, and access it requires — and refuses to proceed if anything is missing. After validation, it creates and manages the application’s VMs, networks, and volumes within the supplied boundary.
- The boundary stays with its owner. The boundary itself and the control plane’s bootstrap VM remain under the infrastructure owner’s management, so a broken control plane can always be repaired, rebuilt, or cut off from outside. The control plane monitors its own VM and the boundary’s resources, and when it needs more or different resources, it asks for them — for example as a proposed change to the infrastructure repository — and the owner decides.
When Checks Fail
The contract is checked continuously, not only at startup. When a running environment fails a check, the control plane responds within its scope:
- It alerts the owner, with what failed and what it affects.
- It protects trading, pausing affected strategies and blocking further deployments under the pre-approved rules.
- It repairs what it manages, such as recreating a failed application VM from its declared state. Recreating the VM does not restore the operation on its own: persisted state is restored, and orders and positions are reconciled with the broker, before the affected strategies may trade again.
- It asks for what it cannot fix. Problems with the boundary, the underlying infrastructure, or its own VM are reported to the owner, with a proposed change where one would help.
When the Control Plane Fails
The control plane cannot supervise itself through its own VM. So the application’s protective behaviour does not depend solely on the control plane remaining available:
- The application keeps its own safeguards. Its risk component goes on enforcing trading limits without the control plane.
- Losing contact is a protective trigger. If the application stops hearing from the control plane for longer than a declared interval, it blocks new exposure under the declared policy and accounts for outstanding orders. It resumes only after contact is restored and the required health and reconciliation checks pass.
- The owner is told from outside. The control plane sends a regular heartbeat to a monitor outside the boundary, which alerts the owner when the heartbeat stops. Repairing or rebuilding the control plane is then the owner’s job, as for any part of the boundary.
How It Fits
The control plane is one part of the i26g Platform, which explains the layering, the boundary, and the adapter model. Inside the boundary it works with virtual machines rather than a Kubernetes cluster, for the reasons in Why i26g Runs on VMs, Not Kubernetes, and reaches the infrastructure through the i26g Infrastructure Adapters, described in detail in The Interface Between i26g and Your Infrastructure.
It is the life cycle thinking I apply as a solution architect, made concrete for one specific platform — specific enough to be built, rather than general enough to describe any solution.
Interested in how a self-hosted trading platform can be run with this kind of discipline? Follow the build via RSS, or get in touch directly.