EVE aims to develop an open, agnostic and standardized architecture unifying the approach to developing and orchestrating cloud-native applications across the enterprise on-premises edge. It offers users new levels of control through hardware-assisted virtualization of on-prem edge devices. Once installed, EVE has direct access to and control of underlying resources and provides standard APIs that allow more efficient use of resources and can effectively partition hardware to increase workload consolidation and application multi-tenancy.
EVE supports both ARM and Intel architectures and requires hardware-assisted virtualization. While EVE can run on a board as small as a $20 Orange Pi, the sweet spot for its deployment are IoT Gateways and Industrial PCs.
To get its job done, EVE leverages a lot of great open source projects: Xen Project, Linuxkit and Alpine Linux just to name a few. All of that functionality is being orchestrated by the Go microservices available under pkg/pillar. Why pillar? Well, because pillar is the kind of a monolith we need to break out into true, individual microservices under pkg/.
You will need QEMU 3.x+, Docker, Make and go 1.12+ installed in your system.
Note, that since Linuxkit and manifest-tool are evolving pretty rapidly, we're vendoring those under build-tools/src. This means you don't have to have them locally installed, but it also means your first build time will be much longer.
If you're on MacOS the following steps should get you all the dependencies:
https://golang.org/dl/https://store.docker.com/editions/community/docker-ce-desktop-macMake sure that Docker is up and running on your system. On MacOS just start a docker Application, on Linux make sure docker service is running. Regardless of how you start Docker you can make sure that it is ready for you by running the following command and making sure that it returns both a version of the client AND a version of the server:
docker versionInstall On OSX (using Brew)
$ brew install make$ sudo apt-get install makeOn OSX using Brew
$ brew install qemu$ sudo apt install qemu-utils # for make live
$ sudo apt install qemu-system-x86 # for make runEVE requires being built in a Git repository (the tools keep looking up git commit IDs). The easiest way is to clone EVE repository from GitHub:
git clone https://github.com/lf-edge/eve.git
cd eveBuild both the build-tools as well as the live image in the source directory:
make build-tools
make liveThis will download the relevant docker images from docker hub and create a bootable
image dist/<ARCH>/live.img.
Since almost all developer workflow is driven by the Makefile, it may be useful
to familiarize yourself with various Makefile targets that Project EVE offers.
A short summary of those is available by simply running make without any arguments
make and a more detailed explanation is available as part of our documentation
NOTE: Since the initial build fetches a LOT of bits over the network it may occasionally time out and fail. Typically re-running
makefixes the issue. If it doesn't you can attempt a local build of all the required EVE packages first by runningmake pkgsNOTE: use make parameter "-j" edit number of threads to build faster. set number of threads suggestions CPU*2. on OSX show number of CPU :
sysctl hw.ncpu; on Ubuntu Linux show number of CPU :nproc;
Building of the various images may require downloading packages from the Internet. If you have direct Internet access, everything will "just work". On the other hand, if you need to run behind a proxy, you may run into issues downloading. These manifest in two key areas:
- docker: docker needs to download images from the image registries. Configuring your local installation of docker is beyond the scope of this document, please see here.
- packages: the package updates inside the images running in docker may need to use http/s proxies.
To configure your build process to use proxies, you can set the following environment variables. They will be picked up automatically when running
any make commands and used within the building containers. If they are not set, no proxy is set:
HTTP_PROXYHTTPS_PROXYALL_PROXYNO_PROXY
Finally run the resulting image in QEMU with some default assumptions:
make runNOTE: The default QEMU configuration needs 4GB of memory available. If you get an error message about being unable to allocate memory, try freeing up some RAM. If you can't free up 4GB, you can reduce the memory allocation to qemu from 4096 (4GB) to 2048 (2GB). Running QEMU with less than 2GB of memory is not recommended. To run with a different amount of memory, provide the desired amount in KB as:
make run QEMU_MEMORY=2048NOTE:
make runlaunches qemu with the-biosoption. On some systems, this does not work, and you need to run it with the-pflashargument (or its equivalent properly configured-driveinstead). To enable pflash, run:
make run PFLASH=trueOnce the image boots you can interact with it either by using the console
(right there in the terminal window from which make run was executed).
Your shell in the console is connected to the 'host' os. Everything
interesting is actually happening in the pillar container. Use
eve enter command to enter it (or if you're comfortable with ctr CLI
from containerd - use that instead).
Once in a container you can run the usual xl commands to start VMs and interact with Xen.
To exit out of the QEMU environment, press Ctrl-A + C to reach the QEMU console, then q to quit.
$ exit # leave eve
$ poweroff -f # leave qemuAs described in BUILD.md and REGISTRATION.md, a booting EVE looks in its config partition to determine:
- the URL to a Controller
- hostnames to add to the
/etc/hostsfile - certificates to trust
When run in an emulator using make run, you can override the built-in /config partition by passing it the path of a directory to mount as that partition:
make run CONF_PART=/path/to/partitionNote that the directory must exist to be mounted; if not, it will be ignored. The most common use case is a config directory output on the run of adam.
While running everything on your laptop with QEMU could be fun, nothing beats real hardware. The most cost-effective option, not surprisingly, is ARM. We recommend two popular board HiKey and Raspberry Pi 4. The biggest difference between the two is that on Raspberry Pi (since it doesn't have any built-in flash storage) you won't be able to utlize EVE's installer and you'll have to build a live image. With HiKey you can use a standard EVE's installer. The steps to do both are outlined below:
Raspberry Pi 4 is a tiny, but capable enough ARM board that allows EVE to run with KVM hypervisor (and soon Xen) hypervisors. While EVE would run in the lowest memory configuration (1GB) if you plan to use it for actual EVE development we strongly recommend buying a 4GB RAM option.
Since a full Raspberry Pi 4 support is only available in upstream Linux kernels starting from 5.6.0, you'll have to use that bleeding edge kernel for your build. Another peculiar aspect of this board is that it doesn't use a standard bootloader (e.g. u-boot or UEFI) so we need to trick it into using our own u-boot as UEFI environment. Thankfully, our Makefile logic tries to automate as much of it as possible. Thus, putting it all together, here are the steps to run EVE on Raspberry Pi 4:
- Make sure you have a clean build directory (since this is a non-standard build)
rm -rf dist/arm64 - Build a live image
make ZARCH=arm64 HV=rpi-kvm live-raw - Flash the
dist/arm64/live.rawlive EVE image onto your SD card by following these instructions
Once your Raspberry Pi 4 is happily running an EVE image you can start using EVE controller for further updates (so that you don't ever have to take an SD card out of your board). Build your rootfs by running make ZARCH=arm64 rootfs-kvm-rpi and give resulting dist/arm64/installer/rootfs.img to the controller.
You may notice that the Makefile logic doesn't produce regular devicetree blob for Raspberry Pi 4 and only enables it to run in KVM hypervisor mode. Both of these may change in the future as we evolve support for this board, but for now you just need to keep this in mind and alway build with a special option HV=rpi-kvm.
One final note about Raspberry Pi 4 GPU support is that since we are running EVE in 64bit (aarch64) mode we are still waiting for the proper VC4 DRM BCM2711 drivers to be upstreamed. Currently it is expected that Kernel 5.7 may actually ship the fully functional driver.
Unlike Raspberry Pi boards, HiKey boards come with a built-in flash, so we will be using EVE's installer to install a copy of EVE onto that storage. You can follow these steps to prepare your installation media:
- Start by cloning EVE git repository
git clone https://github.com/lf-edge/eve.git - Build an installer image
cd eve ; make ZARCH=arm64 installer - Flash the
dist/arm64/installer.rawonto the USB stick by following these instructions
Since by default HiKey is using WiFi for all its networking, you will also have to provide SSID and password for your WiFi network. On Mac OS X you can simply re-insert SD card and edit wpa_supplicant.conf that will appear on volume called EVE.
At this point you have everything you need to permanently install onto HiKey's internal flash. This, of course, will mean that if you have anything else installed there (like a Debian or Android OS) it will be replaced so make sure to make a backup if you nee to.
Additionally, our installer will try to configure an entry point to the
initial boot sequence via GRUB. Since the only reliable way to do so is
by replacing a file called fastboot.efi in the system boot partition you
need to make sure that you have fastboot.efi present there (since if isn't
there installer will refuse to proceed). The easiest way to check for
all that is to invoke an EFI shell on HiKey. Here's how: put the SD card
into the KiKey, connect HiKey to your serial port, start screen, poweron
HiKey and immediately start pressing <ESC> key to trigger EFI shell:
screen /dev/tty.usbserial-* 115200
[1] fastboot
[2] boot from eMMC
[3] boot from SD card
[4] Shell
[5] Boot Manager
Start: 4
.....
Press ESC in 4 seconds to skip startup.nsh or any other key to continue.
Shell> ls fs2:\EFI\BOOT\fastboot.efi
Shell> setsize 1 fs2:\EFI\BOOT\fastboot.efiNOTE: you only need to execute the last (setsize) command if, for whatever reason, the previous command doesn't show fastboot.efi present on your system. Once you've either verified that there's an existing fastboot.efi (or created a dummy one via the setsize command) you can proceed with the rest of the installation from the same EFI shell by executing:
Shell> fs0:\EFI\BOOT\BOOTX64.EFIYou will see an installation sequence scroll on screen and the output that indicates a successful install will look like this:
[ 85.717414] mmcblk0: p1 p2 p3 p4 p5 p6 p7 p8 p11
[ 87.420407] mmcblk0: p1 p2 p3 p4 p5 p6 p7 p8 p11 p12
[ 118.754353] mmcblk0: p1 p2 p3 p4 p5 p6 p7 p8 p11 p12 p13
[ 119.801805] mmcblk0: p1 p2 p3 p4 p5 p6 p7 p8 p11 p12 p13 p14
[ 120.992048] mmcblk0: p1 p2 p3 p4 p5 p6 p7 p8 p11 p12 p13 p14 p19
[ 127.191119] reboot: Power down
(XEN) Hardware Dom0 halted: halting machineAt this point you should remove your SD card from HiKey's slot and reboot the board. If everything went as planned you will boot right into the running system. One thing that you will notice is that a successful installation sequence made a backup copy of your existing fastboot.efi under the fastboot.efi.XXX name. This allows you to restore your HiKey to a pristine state without going through a full fledged re-flashing sequence.
Alternatively, if you're not quite ready to commit to replace your current OS on the HiKey, you can try running from the SD card. For that you will have to put a live system on the SD card, not the installer. Here's how you can do that on Mac OS X:
vi conf/wpa_supplicant.conf
# put your WIFI passwords in and/or add your own networks
make ZARCH=arm64 MEDIA_SIZE=8192 live
sudo dd if=dist/arm64/live.raw of=/dev/rdiskXXX bs=1mThen you can boot into a live system from triggering UEFI shell like shown above and executing exactly the same boot command:
Shell> fs0:\EFI\BOOT\BOOTX64.EFIThe following steps have been tested on Intel UP Squared Board (AAEON UP-APL01) and the bootable USB Disk containing the installer image has been made on Ubuntu 16.04:
- Start by cloning EVE git repository
git clone https://github.com/lf-edge/eve.git - Build an installer image
cd eve ; make ZARCH=amd64 installer - Flash the
dist/amd64/installer.rawonto the USB stick by following these instructions - Now plug the USB Disk on your UP Squared Board and the installer should now replace the existing OS on the UP Squared board with EVE
You will see an installation sequence scroll on screen and the output that indicates a successful install will look like this:
[10.69716164] mmcblk0:
[11.915943] mmcblk0: p1
[13.606346] mmcblk0: p1 p2
[29.656563] mmcblk0: p1 p2 p3
[30.876806] mmcblk0: p1 p2 p3 p4
[32.156930] mmcblk0: p1 p2 p3 p4 p9
NOTICE: Device will now power off. Remove the USB stick and power it back on to complete the installation.
[43.185325] ACPI: Preparing to enter system sleep state S5
[43.187349] reboot: Power downAt this point you should remove your USB Disk from the UP Squared Board slot and reboot the board. If everything went as planned you will boot right into the running system.
EVE is an very low-level engine that requires producing USB sticks and SD cards that are formatted in a very particular way in order to make EVE install and/or run on a given Edge Node. This, in turn, requires EVE hackers to be comfortable with following instructions which, with a simple typo, can completely destroy the system you're running them on (by overwriting your own disk instead of SD card or a USB stick).
PROCEED AT YOUR OWN RISK
If you want to write any binary artifact foo.bin produced by an EVE build onto an SD card (or any other installation medium) try the following:
Find the device that you will be writing to using
fdisk -ldiskutil listNow format the USB Disk and run the following commands
umount /dev/sdXXX
sudo dd if=dist/XXX/foo.bin of=/dev/sdXXX
eject /dev/sdXXXdiskutil unmountDisk /dev/sdXXX
sudo dd if=dist/XXX/foo.bin of=/dev/sdXXX
diskutil eject /dev/sdXXXAlternatively the image can be written with tools like Balena's Etcher
You may be wondering why do we have a container-based architecture for a Xen-centric environment. First of all, OCI containers are a key type of a workload for our platform. Which means having OCI environment to run them is a key requirement. We run them via:
- Set up the filesystem root using containerd
- Launch the domU using Xen via
xl
In addition to that, while we plan to build a fully disagregated system (with even device drivers running in their separate domains) right now we are just getting started and having containers as a first step towards full disagreagation seems like a very convenient stepping stone.
Let us know what you think by filing GitHub issues, and feel free to send us pull requests if something doesn't quite work.