Skip to content
Install

Run on a small server

Orchard runs on 2 GiB of RAM. 8 GiB is recommended. On a small server, the installer gives Orchard its smallest footprint, so that most of the memory stays free for your apps. Below 3.5 GiB, it also starts with the MCP endpoint and image scanning turned off, because each one holds memory all the time.

The installer reads the RAM of the server when it runs.

RAM Profile
Under 2 GiB The installer warns that Orchard leaves little or no room for apps, and continues.
2 to 3.5 GiB The smallest profile. Refer to The smallest profile.
3.5 to 8 GiB Metrics, logs, and the scanner use lower requests and limits, 7 days of logs are kept, and one scan runs at a time.
8 GiB and more The defaults.

Under 3.5 GiB of RAM, a new install:

  • Gives the control plane small memory requests. A request is the memory the scheduler keeps free for a pod, so small requests leave room to schedule your apps. The limits stay high, so a pod that needs more memory for a moment can use it.

    Pod Memory request Memory limit
    server 192 MiB 1 GiB
    frontend 64 MiB 512 MiB
    Postgres 256 MiB 1 GiB
    Redis 32 MiB 256 MiB
  • Limits the Node.js heap of the server to 512 MiB and of the dashboard to 192 MiB, with NODE_OPTIONS.

  • Requests 512 MiB for each build, instead of 2 GiB, with BUILDER_MEMORY_REQUEST. A build can still use more memory up to its limit. A large build can run out of memory on a small server. To build somewhere else, refer to Build outside the server.

  • Turns off the MCP endpoint and image scanning, unless you pass --mcp true, --mcp-domain, or --image-scanning true. To turn them on later, run sudo orchardctl mcp enable --apply, or sudo orchardctl config set trivy.enabled true --apply.

The installer applies this profile to a new install only. A re-run of the installer keeps the values that are already in /etc/orchard/values.yaml. To use a value from the profile on an existing install, set it with sudo orchardctl config set or sudo orchardctl env set.

These parts of Orchard are optional. Each one uses memory all the time.

Part Memory request To turn it off at install To turn it off later
Metrics About 210 MiB --metrics false sudo orchardctl config set metrics.enabled false --apply
Logs About 160 MiB --logs false sudo orchardctl config set logs.enabled false --apply
Image scanning 256 MiB --image-scanning false sudo orchardctl config set trivy.enabled false --apply
The MCP endpoint 128 MiB --mcp false sudo orchardctl mcp disable --apply

Without metrics, the dashboard does not show charts of CPU, memory, and requests. Without logs, the dashboard shows only the logs of running pods, not their history.

Orchard sends traces and profiles to Datadog only when you turn it on. The installer and the chart leave it off. When it is off, the server does not load the Datadog tracer, so the tracer uses no memory. The tracer, the profiler, and App and API Protection each add to the memory of every server pod, so leave Datadog off on a small server unless you use it.

To turn it on, run a Datadog agent in the cluster and set datadog.enabled to true. For the values, refer to the chart reference.

A build uses the most memory of anything that Orchard does. On a server with 2 GiB of RAM, a large build can fail with an out-of-memory error. To keep builds off the server, build the image in CI, push it to a registry, and deploy the image. For more information, refer to Deploy an image.

An installer host is single-node, so it cannot take a second node for builds. If you outgrow one machine, run the Helm chart on a multi-node cluster and reserve a node for builds. For the reasons, refer to Add and remove nodes.

A small server usually has a small disk too. To keep the build cache and old images from filling it, refer to Free disk space.

  • Install - Install Orchard on a Linux machine with the one-command installer.
  • Scale the control plane - Resize the metrics stack and the control-plane pods with chart values.