Skip to content

fedora-iot/greenboot

Repository files navigation

greenboot

Generic Health Check Framework for systemd on rpm-ostree based systems.

Table of contents

Installation

Greenboot is comprised of two packages:

  • greenboot itself, with all core functionalities: check provided scripts, reboot if these checks don't pass, rollback to previous deployment if rebooting hasn't solved the problem, etc.
  • greenboot-default-health-checks, a series of optional and curated health checks provided by Greenboot maintainers.

In order to get a full Greenboot installation on Fedora Silverblue, Fedora IoT or Fedora CoreOS:

rpm-ostree install greenboot greenboot-default-health-checks

systemctl reboot

Usage

Health checks with bash scripts

Place shell scripts representing health checks that MUST NOT FAIL in the /etc/greenboot/check/required.d directory. If any script in this folder exits with an error code, the boot will be declared as failed. Error message will appear in both MOTD and in journalctl -u greenboot-healthcheck.service. Place shell scripts representing health checks that MAY FAIL in the /etc/greenboot/check/wanted.d directory. Scripts in this folder can exit with an error code and the boot will not be declared as failed. Error message will appear in both MOTD and in journalctl -u greenboot-healthcheck.service -b. Place shell scripts you want to run after a boot has been declared successful (green) in /etc/greenboot/green.d. Place shell scripts you want to run after a boot has been declared failed (red) in /etc/greenboot/red.d.

Unless greenboot is enabled by default in your distribution, enable it by running systemctl enable greenboot-task-runner greenboot-healthcheck greenboot-status greenboot-loading-message greenboot-grub2-set-counter greenboot-grub2-set-success greenboot-rpm-ostree-grub2-check-fallback redboot-auto-reboot redboot-task-runner. It will automatically start during the next boot process and run its checks.

When you ssh into the machine after that, a boot status message will be shown:

Boot Status is GREEN - Health Check SUCCESS
Boot Status is RED - Health Check FAILURE!

Directory structure:

/etc
└── greenboot
    ├── check
    │   ├── required.d
    │   └── wanted.d
    ├── green.d
    └── red.d

Health checks included with subpackage greenboot-default-health-checks

These health checks are available in /usr/lib/greenboot/check, a read-only directory in rpm-ostree systems. If you find a bug in any of them or you have an improvement, please create a PR with such fix/feature and we'll review it and potentially include it.

  • Check if repositories URLs are still DNS solvable: This script is under /usr/lib/greenboot/check/required.d/01_repository_dns_check.sh and makes sure that DNS queries to repository URLs are still available.
  • Check if update platforms are still reachable: This script is under /usr/lib/greenboot/check/wanted.d/01_update_platform_check.sh and tries to connect and get a 2XX or 3XX HTTP code from the update platforms defined in /etc/ostree/remotes.d.
  • Check if current boot has been triggered by hardware watchdog: This script is under /usr/lib/greenboot/check/required.d/02_watchdog.sh and checks whether the current boot has been watchdog-triggered or not. If it is, but the reboot has occurred after a certain grace period (default of 24 hours, configurable via GREENBOOT_WATCHDOG_GRACE_PERIOD=number_of_hours in /etc/greenboot/greenboot.conf), Greenboot won't mark the current boot as red and won't rollback to the previous deployment. If has occurred within the grace period, at the moment the current boot will be marked as red, but Greenboot won't rollback to the previous deployment. It is enabled by default but it can be disabled by modifying GREENBOOT_WATCHDOG_CHECK_ENABLED in /etc/greenboot/greenboot.conf to false.

Health Checks with systemd services

Overall boot success is measured against boot-complete.target. Ordering of units can be achieved using standard systemd vocabulary.

Required Checks

Create a oneshot health check service unit that MUST NOT FAIL, e.g. /etc/systemd/system/required-check.service. Make sure it calls redboot.target when it fails (OnFailure=redboot.target). Run systemctl enable required-check to enable it.

[Unit]
Description=Custom Required Health Check
Before=boot-complete.target
OnFailure=redboot.target
OnFailureJobMode=fail

[Service]
Type=oneshot
ExecStart=/usr/libexec/mytestsuite/required-check

[Install]
RequiredBy=boot-complete.target
WantedBy=multi-user.target

Wanted Checks

Create a oneshot health check service unit that MAY FAIL, e.g. /etc/systemd/system/wanted-check.service. Run systemctl enable wanted-check to enable it.

[Unit]
Description=Custom Wanted Health Check
Before=boot-complete.target

[Service]
Type=oneshot
ExecStart=/usr/libexec/mytestsuite/wanted-check

[Install]
WantedBy=boot-complete.target
WantedBy=multi-user.target

Configuration

At the moment, it is possible to customize the following parameters via environment variables. These environment variables can be described as well in the config file /etc/greenboot/greenboot.conf:

  • GREENBOOT_MAX_BOOT_ATTEMPTS: Maximum number of boot attempts before declaring the deployment as problematic and rolling back to the previous one.
  • GREENBOOT_WATCHDOG_CHECK_ENABLED: Enables/disables Check if current boot has been triggered by hardware watchdog health check. More info on Health checks included with subpackage greenboot-default-health-checks section.
  • GREENBOOT_WATCHDOG_GRACE_PERIOD: Number of hours after an upgrade that we consider the new deployment as culprit of reboot.

How does it work

  • greenboot-rpm-ostree-grub2-check-fallback.service runs before greenboot-healthcheck.service and checks whether the GRUB2 environment variable boot_counter is -1.
    • If it is -1, this would mean that the system is in a fallback deployment and would execute rpm-ostree rollback to go back to the previous, working deployment.
    • If boot_counter is not -1, nothing is done in this step.
  • greenboot-healthcheck.service runs before systemd's boot-complete.target. It launches /usr/libexec/greenboot/greenboot check, which runs the required.d and wanted.d scripts.
    • If any script in the required.d folder fails, redboot.target is called.
      • It triggers redboot-task-runner.service, which launches /usr/libexec/greenboot/greenboot red. This will run the scripts in red.d folder.
      • After the above:
        • greenboot-status.service is run, creating the MOTD specifying which scripts have failed.
        • redboot-auto-reboot.service is run. It performs a series of checks to determine if there's a requirement for manual intervention. If there's not, it reboots the system.
    • If all scripts in required.d folder succeeded:
      • boot-complete.target is reached.
      • greenboot-grub2-set-success.service is run. It unsets boot_counter GRUB env var and sets boot_success GRUB env var to 1.
      • greenboot-task-runner.service launches /usr/libexec/greenboot/greenboot green, which runs the scripts in green.d folder, scripts that are meant to be run after a successful update.
      • greenboot-status.service is run, creating the MOTD with a success message.

Development

Please refer to development/README.md file.