RFC-0191: Fuchsia-on-GCE as a supported system configuration

RFC-0191: Fuchsia-on-GCE as a supported system configuration
  • Governance

Add Google Compute Engine (x64 and arm64) as a supported system configuration for Fuchsia.

Gerrit change
Date submitted (year-month-day)2022-08-29
Date reviewed (year-month-day)2022-09-28


This document proposes the addition of GCE (Google Compute Engine) as a supported system configuration by the Fuchsia project. This simply formalizes our existing usage of GCE as a platform for testing Fuchsia, and codifies our commitment to keep GCE working at ToT.

We should note, however, that while we do utilize this at scale, codifying its usage as a supported Fuchsia platform will require that we maintain support for ACPI and UEFI in the Zircon kernel. This is a relatively well-exercised path today for x64, but less so for ARM64.


Fuchsia-on-GCE is already utilized by infrastructure to improve the scalability, diversity, and performance of CI/CQ testing. Concretely, we already test the following configurations:

  • bringup.x64
  • bringup.x64 on AMD
  • bringup.arm64
  • core.x64

Now that we are performing a significant amount of work on GCE, we want to formalize this as supported hardware by the Fuchsia project by adding it to the supported hardware YAML file in /docs/reference/hardware. RFC-0130: Supported Hardware states that "The process for adding new entries to this YAML file, and hence to add new hardware in the Supported category, is to create a new RFC."


Facilitator: rlb@

The person appointed by FEC to shepherd this RFC through the RFC process.


  • cpu@google.com
  • curtisgalloway@google.com
  • tkilbourn@google.com
  • travisg@google.com


  • maniscalco@google.com
  • venkateshs@google.com
  • nmulcahey@google.com
  • andresoportus@google.com
  • simonshields@google.com


This RFC was socialized with all of Zircon and Engprod, and was then shared with all of Fuchsia.


Adding GCE to the list of supported system configurations simply requires editing the /docs/reference/hardware/_supported.yaml file. We will add two entries to the file, one for x86 and one for ARM.

x86 entry

  • name: 'GCE x86_64'
  • description: 'Google Compute Engine emulated x86 board'
  • architecture: 'x86'
  • manufacturer_link: 'https://cloud.google.com/compute'
  • board_driver_location: '/src/devices/board/drivers/x86'

ARM entry

  • name: 'GCE ARM64'
  • description: 'Google Compute Engine emulated ARM64 board'
  • architecture: 'ARM'
  • manufacturer_link: 'https://cloud.google.com/compute'
  • board_driver_location: 'src/devices/board/drivers/acpi-arm64'


Fuchsia on GCE is faster than running Fuchsia in an emulator on a Linux host. This is especially true on x86, where we use nested virtualization, so running directly on GCE removes 2 layers of virtualization. It is also only slightly (~10%) slower than running on a physical NUC.


Infrastructure would like to reduce its dependence on inelastic hardware fleets in presubmit testing, as these often increase pending times and have a disproportionate impact on CI/CQ latency. Having a supported, representative, and virtualized platform that we can scale easily and quickly will reduce this dependency. This will, in turn, improve the developer experience by decreasing the latency to submission.

We have also invested in tooling to improve the ergonomics of Fuchsia-on-GCE for developers at their desk. Specifically, the build system already produces the UEFI disk image needed to create a compute instance, and the fx gce commands provide a simple way to create, connect to, and delete GCE VMs.

Backwards Compatibility

Supporting GCE does not require any breaking changes within the platform.


As mentioned earlier, infrastructure will run a variety of Fuchsia configurations on GCE in CI/CQ to ensure that it is stable. We will also run in the GCE staging environment to make sure that we get early notice of any cloud changes that affect/break Fuchsia.

Drawbacks, alternatives, and unknowns

The main drawback of supporting GCE is dealing with unexpected changes from Google Compute that break Fuchsia. To avoid this situation, infra will run all of the configurations we plan to run on GCE (bringup.x64, bringup.arm64, core.x64, etc.) in the GCE staging environment. This environment receives changes about 4 weeks before they enter production, meaning that we will be able to detect breakages a month ahead of time. We can then use this time to either:

  1. Work with cloud to find a solution, or
  2. Modify Fuchsia to work with whatever new changes cloud released

It's worth noting that in the ~18 months we have been running on GCE, cloud has launched only one feature that broke Fuchsia, so the rate of breakages should be low.