Keeping integrations apart
Weik.io 2026.2Run integrations that must not see each other's data on agents with different allowed projects. Every build at the hub runs isolated with Landlock. Video
Watch: What runs where, 3:42
The video shows where the hub and the agents run and the two outbound connections an agent makes on port 443.
This part of the video couldn't be loaded.
On this page
Because an integration runs Java code that its authors wrote, Weik.io keeps integrations apart with the operating system and the hub’s credentials. While integrations run, the agent is the boundary. It reaches only the projects and environments an administrator allowed it. At the hub, every build runs isolated and can use only the files and connections it needs.
Agents open no inbound port. Each agent connects out to the hub on port 443, and the hub publishes port 443 only, with TLS 1.2 and 1.3. Secrets are sealed with AES-256-GCM under a master key kept on the host. The audit stream can't be deleted or shortened with the backend's own credentials. An agent's credential reaches only the projects and environments an admin allowed for it, such as finance in production.
Separate integrations by agent#
All integrations on one agent run as the agent’s user and could read each other’s files. Run integrations that must not see each other’s data, such as two suppliers’ code, on agents with different allowed projects:
$ weikio agent enrol supplier-a-01 --host linux-compose --capabilities environment=production --projects supplier-a --out ./supplier-a-01
$ weikio agent enrol supplier-b-01 --host linux-compose --capabilities environment=production --projects supplier-b --out ./supplier-b-01The separation covers the entity store of an agent’s projects, their integration sources and built files, and everything else the agent reaches from its host. Only an administrator can change which projects and environments an agent serves.
One agent can serve several environments, but to keep two environments apart, run an agent for each of them. The hub keeps the two agents’ connections separate.
What an agent’s credential reaches#
- The agent’s own requests to the hub and the hub’s messages to it, such as what it should run, each deployment’s configuration and secrets, its new credential after a re-issue and calls into its deployments
- The event channels of its allowed projects and environments
- Read access to the integration sources and built files of those projects
- Write access to the entity store of those projects and environments, and read access to its changes
Nothing on an agent’s host can write a project’s integration sources or the files its builds produced.
Because each deployment’s JVM gets a telemetry credential of its own when it starts, its telemetry can’t show up as another integration’s or under another project. The credential stops working when its agent is revoked. After the deployment is deleted, the credential keeps working for as long as the installation keeps telemetry. Telemetry that the agent held back still lands under the deployment.
When an administrator changes an agent’s environments or projects, Weik.io issues a new credential for the agent to fetch and withdraws the one it replaced. As soon as an administrator revokes an agent, its connection to the hub goes down. The agent can’t connect again after that.
Builds at the hub#
Weik.io builds every version of an integration once, from its authors’ code. On Linux 6.7 or later with Landlock enabled, such as Ubuntu 24.04, Weik.io limits what each build step can do:
- Apart from the system folders and the folders it may write, it can open only the build’s JDK and the Camel and Maven files the build uses.
- It can write only to its build folder and its project’s Maven folder and can’t change the Maven settings.
- It can make TCP connections only to the ports of the Maven repositories and the Maven mirror that Weik.io is set up with.
- It can’t open the master key or any other secret file of the hub’s backend service, such as the installation’s keys.
- It can’t make a symbolic link.
Each step also runs with fixed environment variables and with Java’s attach mechanism off. The backend doesn’t follow symbolic links when it collects build output.
A build step can see the repository credentials in the Maven settings. Give Weik.io a repository account that can only read.
To see which of the limits the host provides, read the build isolation line of doctor:
$ docker compose run --rm infra weikio-backend doctorBelow Linux 6.7 the kernel can’t limit a step’s TCP connections. Below Linux 5.13, or with Landlock off at boot, it can’t limit the step’s file access either. On such a hub server, doctor warns about the kernel. Until a platform administrator allows builds, Weik.io builds nothing there and reports the problem Builds are stopped for the server. An apply that would build a version fails.
To build on such a server anyway, allow builds without the network limit:
$ weikio settings set --allow-builds-without-network-limit=trueIn the admin UI, open Settings › Platform › Installation and choose Edit installation settings. Set Builds without a network limit to Allowed. A build step on that server then runs with the limits its kernel has. Because its TCP connections aren’t limited, the step can reach the hub’s own service ports. To get every limit back, move the hub to Linux 6.7 or later with Landlock enabled, such as Ubuntu 24.04.
On every kernel, some of a build step’s access isn’t limited:
- UDP traffic
- Unix sockets
- Signals
The TCP limit allows the repositories’ ports on any host.
Build folders per project#
Each project has a separate Maven folder and build cache. No build uses libraries or compiled classes that another project’s build wrote. A project’s first build on a Camel version downloads what Weik.io’s shared Maven folder lacks into the project’s folder.
When you delete a project, Weik.io deletes its Maven folder and build cache once its running builds end. A project created later under the same name starts clean.