Worked example of migrating a known device

This guide makes the following assumptions:

  1. You have access to a working LAVA instance with pipeline support.

  2. You have at least one working device of a known device type

  3. You have at least one working JSON job submission for that device.

  4. You are migrating to a deployment type and boot type which is already supported by the pipeline code.

  5. You have read access to the current configuration of that device, including PDU port numbers and serial port access.

Things will go more easily if you also have:

  1. admin access to the django configuration of the LAVA instance

  2. root command line access to the dispatcher currently using the device.

  3. a browser tab open at the Online YAML parser

Note

some parts of the refactoring are still in development, so not all of the support may be available. However, as YAML supports comments, this data is not lost.

The objective is to migrate the configuration of the existing device such that exactly the same commands are sent to the device as in the current dispatcher jobs, with the benefit of pipeline validation, results and metadata.

Note

The specific example follows a real conversion but the hardware concerned has since changed to a different deployment method. The examples here worked when the example was written but will not work on current deployments of the hardware concerned.

Writing a device configuration in YAML

The current dispatcher configuration is in two parts and these will typically be respected in the migration.

The device type configuration will become a device type template.

The device configuration will become a device dictionary.

However, initially, we need a single YAML file which contains the data that the pipeline will send to the dispatcher - a combination of the device type and device information. You can see examples of such content when validating pipeline jobs. (This is example and has been edited slightly to take out some of the noise.):

$ sudo lava-dispatch --target devices/bbb-01.yaml bbb-uboot-ramdisk.yaml --validate --output-dir=/tmp/test/

Don’t worry about running this example yourself at this stage. The files themselves may be useful for reference. The device YAML file comes from the lava-dispatcher unit tests:

https://git.lavasoftware.org/lava/lava/blob/master/lava_scheduler_app/tests/devices/bbb-01.yaml

The job submission YAML used in the example comes from the lava-team refactoring repository of functional tests:

https://git.linaro.org/lava-team/refactoring.git/tree/bbb-uboot-ramdisk.yaml

device: !!python/object/new:lava_dispatcher.device.NewDevice
 dictitems:
   actions:
     boot:
       prompts:
         - 'linaro-test'
         - 'root@debian:~#'
       connections: {serial: null, ssh: null}
       methods:
         u-boot:
           parameters: {boot_message: Booting Linux, bootloader_prompt: U-Boot}
           ramdisk:
             commands: [setenv autoload no, setenv initrd_high '0xffffffff', setenv
                 fdt_high '0xffffffff', 'setenv kernel_addr_r ''{KERNEL_ADDR}''',
               'setenv initrd_addr_r ''{RAMDISK_ADDR}''', 'setenv fdt_addr_r ''{DTB_ADDR}''',
               'setenv loadkernel ''tftp ${kernel_addr_r} {KERNEL}''', 'setenv loadinitrd
                 ''tftp ${initrd_addr_r} {RAMDISK}; setenv initrd_size ${filesize}''',
               'setenv loadfdt ''tftp ${fdt_addr_r} {DTB}''', 'setenv bootargs ''console=ttyO0,115200n8
                 root=/dev/ram0 ip=dhcp''', 'setenv bootcmd ''dhcp; setenv serverip
                 {SERVER_IP}; run loadkernel; run loadinitrd; run loadfdt; {BOOTX}''',
               boot]
     deploy:
       methods:
         tftp: null
         usb: null
   commands: {connect: telnet localhost 6000}
   hostname: bbb-01
   parameters:
     bootm: {dtb: '0x815f0000', kernel: '0x80200000', ramdisk: '0x81600000'}
     bootz: {dtb: '0x81f00000', kernel: '0x81000000', ramdisk: '0x82000000'}
   power_state: 'off'
   timeouts:
     apply-overlay-image: {seconds: 120}
     lava-test-shell: {seconds: 30}
     power_off: {seconds: 5}
     umount-retry: {seconds: 45}
 state: {target: bbb-01}

This snippet includes the connection command (telnet localhost 6000) from the device configuration and the ramdisk uboot parameters from the device type configuration - note that as this is the validation output, no job files have been downloaded, so the substitution placeholders remain, {DTB}, {SERVER_IP}, {KERNEL} etc. - this is correct and will help with the next steps. What isn’t so helpful at the moment is the layout of this YAML dump.

Migrating a mustang

Existing configuration:

device_type = mustang
hostname = staging-mustang01
hard_reset_command = /usr/bin/pduclient --daemon services --hostname pdu15 --command reboot --port 05
power_off_cmd = /usr/bin/pduclient --daemon services --hostname pdu15 --command off --port 05
connection_command = telnet serial4 7012
reset_port_command = flock /var/lock/serial1.lock /usr/local/lab-scripts/reset-serial5000 serial4 12
image_boot_msg_timeout = 240

Start with a new file:

# hostname is irrelevant in the refactoring, the dispatcher uses what it is given.
commands:
  connect: telnet serial4 7012
  hard_reset: /usr/bin/pduclient --daemon services --hostname pdu15 --command reboot --port 05
  power_off: /usr/bin/pduclient --daemon services --hostname pdu15 --command off --port 05
  power_on: /usr/bin/pduclient --daemon services --hostname pdu15 --command on --port 05
  # power_on is new in the refactoring.
  # reset_port_command not yet ported:
  # reset_port: flock /var/lock/serial1.lock /usr/local/lab-scripts/reset-serial5000 serial4 12
  # timeouts are handled later in the file.

So far, so good. Now add the device type configuration blocks. This is the existing configuration:

client_type = bootloader

bootloader_prompt = Mustang
uimage_only = True
boot_cmd_timeout = 60
text_offset = 0x80000

u_load_addrs =
   0x4002000000
   0x4004000000
   0x4003000000

z_load_addrs =
   0x4002000000
   0x4004000000
   0x4003000000

boot_cmds_nfs =
   setenv autoload no,
   setenv kernel_addr_r "'{KERNEL_ADDR}'",
   setenv initrd_addr_r "'{RAMDISK_ADDR}'",
   setenv fdt_addr_r "'{DTB_ADDR}'",
   setenv loadkernel "'tftp ${kernel_addr_r} {KERNEL}'",
   setenv loadinitrd "'tftp ${initrd_addr_r} {RAMDISK}'",
   setenv loadfdt "'tftp ${fdt_addr_r} {DTB}'",
   setenv nfsargs "'setenv bootargs root=/dev/nfs rw nfsroot={SERVER_IP}:{NFSROOTFS},tcp,hard,intr panic=1 console=ttyS0,115200 earlyprintk=uart8250-32bit,0x1c020000 debug ip=dhcp'",
   setenv bootcmd "'dhcp; setenv serverip {SERVER_IP}; run loadkernel; run loadinitrd; run loadfdt; run nfsargs; {BOOTX}'",
   boot

boot_cmds_ramdisk =
   setenv autoload no,
   setenv kernel_addr_r "'{KERNEL_ADDR}'",
   setenv initrd_addr_r "'{RAMDISK_ADDR}'",
   setenv fdt_addr_r "'{DTB_ADDR}'",
   setenv loadkernel "'tftp ${kernel_addr_r} {KERNEL}'",
   setenv loadinitrd "'tftp ${initrd_addr_r} {RAMDISK}'",
   setenv loadfdt "'tftp ${fdt_addr_r} {DTB}'",
   setenv bootargs "'root=/dev/ram0 rw panic=1 console=ttyS0,115200 earlyprintk=uart8250-32bit,0x1c020000 debug ip=dhcp'",
   setenv bootcmd "'dhcp; setenv serverip {SERVER_IP}; run loadkernel; run loadinitrd; run loadfdt; {BOOTX}'",
   boot

boot_cmds =
   boot

boot_options =
   boot_cmds

[boot_cmds]
default = boot_cmds

Extend the existing YAML file, to add:

  1. parameters

  2. actions

  3. deploy and boot methods

  4. method parameters

  5. method commands

Parameters

Note how the existing config just lists the addresses without identifying which is the kernel load addr. Although these blocks are the same in this example, the addresses can differ between z_load and u_load.:

u_load_addrs =
   0x4002000000
   0x4004000000
   0x4003000000
z_load_addrs =
   0x4002000000
   0x4004000000
   0x4003000000

Use a working job log file to identify which is where:

<LAVA_DISPATCHER>2015-06-19 08:32:29 AM DEBUG: boot_cmds(after preprocessing):
['setenv autoload no', u"setenv kernel_addr_r '0x4002000000'",
u"setenv initrd_addr_r '0x4004000000'",
u"setenv fdt_addr_r '0x4003000000'",
u"setenv loadkernel 'tftp ${kernel_addr_r} tmplv_wQe/uImage_1.11'",
"setenv loadinitrd 'tftp ${initrd_addr_r} {RAMDISK}'",
u"setenv loadfdt 'tftp ${fdt_addr_r} tmplv_wQe/mustang.dtb_1.11'",
u"setenv nfsargs 'setenv bootargs root=/dev/nfs rw
nfsroot=10.3.2.1:/var/lib/lava/dispatcher/tmp/tmplv_wQe/tmprhrAXO,tcp,hard,intr
panic=1 console=ttyS0,115200 earlyprintk=uart8250-32bit,0x1c020000 debug ip=dhcp'",
u"setenv bootcmd 'dhcp; setenv serverip 10.3.2.1; run loadkernel;
run loadinitrd; run loadfdt; run nfsargs; bootm ${kernel_addr_r} - ${fdt_addr_r}'", 'boot']

Note here that the action job uses bootm, so it is bootm parameters we need to specify.

parameters:
  bootm:
    kernel: '0x4002000000'
    ramdisk: '0x4004000000'
    dtb: '0x4003000000'

Only add bootz support if you know that the U-Boot bootz command is present in the U-Boot version on the board and that it works with zImage kernels. The eventual templates will exist on the server and can be used to declare the detailed device support so that test writers know in advance what kind of images the device can use.

Actions

For this example, the deployment method is relatively simple - you can see from the working job that it is using tftp to deploy.

actions:
  deploy:
    methods:
    - tftp

Always check your YAML syntax. The YAML parser can provide links to small snippets of YAML, like the one above

The boot support is where things become more detailed.

boot:
 prompts:
   - 'linaro-test'
   - 'root@debian:~#'
 methods:
   u-boot:
     parameters:
       bootloader_prompt: Mustang
       boot_message: Starting kernel

The bootloader prompt (at this stage) comes from the device type configuration. The boot message will later be supportable as image-specific. For now, you need whatever values work with the current state of the device. The boot_message is a string emitted during the boot which denotes a successful attempt to boot. There is no need to quote the string unless it contains an illegal character in YAML like a colon.

Next are the commands for the deployment method itself:

nfs:
  commands:
  - setenv autoload no
  - setenv kernel_addr_r '{KERNEL_ADDR}'
  - setenv initrd_addr_r '{RAMDISK_ADDR}'
  - setenv fdt_addr_r '{DTB_ADDR}'
  - setenv loadkernel 'tftp ${kernel_addr_r} {KERNEL}'
  - setenv loadinitrd 'tftp ${initrd_addr_r} {RAMDISK}'
  - setenv loadfdt 'tftp ${fdt_addr_r} {DTB}'
  - "setenv nfsargs 'setenv bootargs root=/dev/nfs rw nfsroot={SERVER_IP}:{NFSROOTFS},tcp,hard,intr panic=1 console=ttyS0,115200 earlyprintk=uart8250-32bit,0x1c020000 debug ip=dhcp'"
  - setenv bootcmd 'dhcp; setenv serverip {SERVER_IP}; run loadkernel; run loadinitrd; run loadfdt; run nfsargs; {BOOTX}'
  - boot

These are retained with only formatting changes - after all, these are what the device needs to be able to boot.

  1. Remove trailing commas (remnants of the old config)

  2. Remove one level of quote marks unless the command embeds a colon (e.g. NFS), in which case the whole line is quoted.

  3. Make each line part of a list by prefixing with a hyphen and a space.

Note

Trailing commas are known to cause problems on devices - check the config carefully and be particularly watchful for failures where the device reports cannot find device 'net0,' when working V1 jobs would report using device 'net0'. Commas are required in V1 but YAML processing for V2 will include trailing commas as part of the string, not part of the formatting.

Timeouts

A process of trial and error will illuminate which timeouts are appropriate to set at this level.

timeouts:
  power_off:
    seconds: 5

Complete device YAML

Untested at this point, but this is the start of the integration.

# hostname is irrelevant in the refactoring, the dispatcher uses what it is given.
commands:
  connect: telnet serial4 7012
  hard_reset: /usr/bin/pduclient --daemon services --hostname pdu15 --command reboot --port 05
  power_off: /usr/bin/pduclient --daemon services --hostname pdu15 --command off --port 05
  power_on: /usr/bin/pduclient --daemon services --hostname pdu15 --command on --port 05
  # power_on is new in the refactoring.
  # reset_port_command not yet ported:
  # reset_port: flock /var/lock/serial1.lock /usr/local/lab-scripts/reset-serial5000 serial4 12
  # timeouts are handled later in the file.
parameters:
  bootm:
    kernel: '0x4002000000'
    ramdisk: '0x4004000000'
    dtb: '0x4003000000'
actions:
  deploy:
    methods:
    - tftp
  boot:
    prompts:
      - 'linaro-test'
      - 'root@debian:~#'
    methods:
      u-boot:
        parameters:
          bootloader_prompt: Mustang
          boot_message: Starting kernel
        nfs:
          commands:
          - setenv autoload no
          - setenv kernel_addr_r '{KERNEL_ADDR}'
          - setenv initrd_addr_r '{RAMDISK_ADDR}'
          - setenv fdt_addr_r '{DTB_ADDR}'
          - setenv loadkernel 'tftp ${kernel_addr_r} {KERNEL}'
          - setenv loadinitrd 'tftp ${initrd_addr_r} {RAMDISK}'
          - setenv loadfdt 'tftp ${fdt_addr_r} {DTB}'
          - "setenv nfsargs 'setenv bootargs root=/dev/nfs rw nfsroot={SERVER_IP}:{NFSROOTFS},tcp,hard,intr panic=1 console=ttyS0,115200 earlyprintk=uart8250-32bit,0x1c020000 debug ip=dhcp'"
          - setenv bootcmd 'dhcp; setenv serverip {SERVER_IP}; run loadkernel; run loadinitrd; run loadfdt; run nfsargs; {BOOTX}'
          - boot

timeouts:
  power_off:
    seconds: 5

Writing a job submission in YAML

Warning

Do not be tempted into writing a script to convert the JSON to YAML. You need to understand what the job is doing and why. e.g. the original job gives no clue that U-Boot is involved nor that the required U-Boot parameters for this job are bootm and not bootz. Any such attempts would re-introduce assumptions that the refactoring is deliberately removing. Just because a file has a particular name or suffix does not mean that the job can make any safe assumptions about the content of that file.

Migrating a job for the mustang

Existing JSON:

{
   "actions": [
       {
           "command": "deploy_linaro_kernel",
           "metadata": {
               "distribution": "debian"
           },
           "parameters": {
               "dtb": "http://images-internal/mustang/mustang.dtb_1.11",
               "kernel": "http://images-internal/mustang/uImage_1.11",
               "login_prompt": "login:",
               "nfsrootfs": "https://people.linaro.org/~neil.williams/arm64/debian-jessie-arm64-rootfs.tar.gz",
               "target_type": "ubuntu",
               "username": "root"
           }
       },
       {
           "command": "boot_linaro_image"
       },
       {
           "command": "lava_test_shell",
           "parameters": {
               "testdef_repos": [
                   {
                       "git-repo": "https://git.linaro.org/people/neil.williams/temp-functional-tests.git",
                       "testdef": "singlenode/singlenode03.yaml"
                   }
               ],
               "timeout": 900
           }
       },
       {
           "command": "submit_results",
           "parameters": {
               "server": "https://staging.validation.linaro.org/RPC2",
               "stream": "/anonymous/lava-functional-tests/"
           }
       }
   ],
   "device_type": "mustang",
   "job_name": "mustang-singlenode-jessie",
   "timeout": 900
}

Identifying the elements of the job

Forget the deploy_linaro_kernel, this is a deployment of a kernel, a DTB and an NFS root filesystem.

Start with the top level structures:

device_type: mustang
job_name: mustang-singlenode-jessie
timeouts:
  job:
    minutes: 15

device_type isn’t strictly necessary at this point but it will become necessary once this job is able to be submitted via the server rather than directly to the dispatcher.

Now identify the actions - a single deploy, a single boot and a single test.

Deploy
- deploy:
      to: tftp
      kernel:
        url: http://images-internal/mustang/uImage_1.11
        type: uimage
      nfsrootfs:
        url: https://people.linaro.org/~neil.williams/arm64/debian-jessie-arm64-rootfs.tar.gz
        compression: gz
      dtb:
        url: http://images-internal/mustang/mustang.dtb_1.11
Boot

Note that boot has the details of the autologin which will occur at the end of the boot action.

- boot:
    prompts:
      - 'root@linaro-nano:'
    method: u-boot
    commands: nfs
    auto_login:
      login_prompt: "login:"
      username: root
Test

Note how the test action can have a name and the test definition can also have a name, separate from the content of the YAML file.

- test:
    timeout:
      minutes: 5
    definitions:
      - repository: https://git.linaro.org/people/neil.williams/temp-functional-tests.git
        from: git
        path: singlenode/singlenode03.yaml
        name: singlenode-advanced

Complete YAML submission

# mustang devices now use grub-efi, not UBoot
# This file exists only as an example of the initial UBoot support.

device_type: mustang
job_name: mustang singlenode with Debian jessie
priority: medium
visibility: public

metadata:
  # please change these fields when modifying this job for your own tests.
  docs-source: pipeline-admin-example
  docs-filename: mustang-admin-example-job.yaml

timeouts:
  job:
    minutes: 15
  action:
    minutes: 2
  connection:
    minutes: 2
actions:
- deploy:
      to: tftp
      kernel:
        url: http://images-internal/mustang/uImage_1.11
        type: uimage
      nfsrootfs:
        url: https://people.linaro.org/~neil.williams/arm64/debian-jessie-arm64-rootfs.tar.gz
        compression: gz
      dtb:
        url: http://images-internal/mustang/mustang.dtb_1.11
- boot:
    prompts:
      - 'root@linaro-nano:'
    method: u-boot
    commands: nfs
    auto_login:
      login_prompt: "login:"
      username: root
- test:
    timeout:
      minutes: 5
    definitions:
      - repository: https://git.linaro.org/people/neil.williams/temp-functional-tests.git
        from: git
        path: singlenode/singlenode03.yaml
        name: singlenode-advanced

Writing a device type template

The purpose of a template is to move as much common data out of each individual template and into the base template for sharing of code. Where parameters differ (e.g. the console port), these are supplied as variables. The device dictionary then only needs to supply information which is specific to that one device - usually including the serial connection command and the power commands.

The first point of reference with a new template is the lava-server base.jinja2 template and existing examples (e.g. beaglebone-black) - templates live on the server, are populated with data from the database and the resulting YAML is sent to the dispatcher.

Starting a new device type template

For example, a new mustang template starts as:

{% extends 'base.jinja2' %}
{% block body %}

{% endblock %}

The content is a jinja2 template based directly on the working device jinja2 template above. Where there are values, these are provided with defaults matching the currently working values. Where there are common blocks of code in base.jinja2, these are pulled in using Jinja2 templates. The commands block itself is left to the device dictionary (and picked up by base.jinja2).

ramdisk and nfs are particularly common deployment methods, so the majority of the commands are already available in base.jinja2. These commands use {{ console_device }} and {{ baud_rate }}, which need to be defined with defaults:

{% set console_device = console_device | default('ttyS0') %}
{% set baud_rate = baud_rate | default(115200) %}

 parameters:
   bootm:
    kernel: '{{ bootm_kernel_addr|default('0x4002000000') }}'
    ramdisk: '{{ bootm_ramdisk_addr|default('0x4004000000') }}'
    dtb: '{{ bootm_dtb_addr|default('0x4003000000') }}'

The actions are determined by the available support for this device, initially, templates can simply support the initial working configuration, more support can be added later.

 actions:
   deploy:
     methods:
       tftp

 boot:
   prompts:
     - 'linaro-test'
     - 'root@debian:~#'
   methods:
     u-boot:
       parameters:
         bootloader_prompt: {{ bootloader_prompt|default('Mustang') }}
         boot_message: {{ boot_message|default('Starting kernel') }}
       nfs:
         commands:
{{ base_uboot_commands }}
{{ base_uboot_addr_commands }}
{{ base_tftp_commands }}
         # Always quote the entire string if the command includes a colon to support correct YAML.
         - "setenv nfsargs 'setenv bootargs console={{ console_device }},{{ baud_rate }}n8 root=/dev/nfs rw {{ base_nfsroot_args }} panic=1 earlyprintk=uart8250-32bit,0x1c020000 debug ip=dhcp'"
{{ base_uboot_bootcmd }}

Completed mustang template

{% extends 'base.jinja2' %}
{% block body %}

device_type: mustang
{% set console_device = console_device | default('ttyS0') %}
{% set baud_rate = baud_rate | default(115200) %}

 parameters:
   bootm:
    kernel: '{{ bootm_kernel_addr|default('0x4002000000') }}'
    ramdisk: '{{ bootm_ramdisk_addr|default('0x4004000000') }}'
    dtb: '{{ bootm_dtb_addr|default('0x4003000000') }}'

 actions:
   deploy:
     methods:
     - tftp

   boot:
     prompts:
       - 'linaro-test'
       - 'root@debian:~#'
     methods:
       u-boot:
         parameters:
           bootloader_prompt: {{ bootloader_prompt|default('Mustang') }}
           boot_message: {{ boot_message|default('Starting kernel') }}
         nfs:
           commands:
           - setenv autoload no
{{ base_uboot_addr_commands }}
{{ base_tftp_commands }}
           # Always quote the entire string if the command includes a colon to support correct YAML.
           - "setenv nfsargs 'setenv bootargs console={{ console_device }},{{ baud_rate }}n8 root=/dev/nfs rw {{ base_nfsroot_args }} panic=1 earlyprintk=uart8250-32bit,0x1c020000 debug ip=dhcp'"
{{ base_uboot_bootcmd }}

{% endblock %}

Creating a device dictionary for the device

Examples of exported device dictionaries exist in the lava-server codebase for unit test support. The dictionary extends the new template and provides the device-specific values.

{% extends 'mustang.jinja2' %}
{% set connection_list = [uart0] %}
{% set connection_commands = {uart0: telnet serial4 7012} %}
{% set connection_tags = {uart0: [primary, 'telnet']} %}
{% set hard_reset_command = "/usr/bin/pduclient --daemon services --hostname pdu15 --command reboot --port 05" %}
{% set power_off_command = "/usr/bin/pduclient --daemon services --hostname pdu15 --command off --port 05" %}
{% set power_on_command = "/usr/bin/pduclient --daemon services --hostname pdu15 --command on --port 05" %}

Testing the template and dictionary

lava-tool has support for comparing the templates with working YAML files and this can be done using files already deployed or local changes prior to submission. To test the local files, create a new directory, add the YAML file used when calling lava-dispatch directly and add two sub-directories:

mkdir ./device-types
mkdir ./devices

Copy base.jinja2 into the device-types directory, along with your new local template. Copy the device dictionary file to devices. If your locally working jinja2 file is called working.jinja2, the comparison would be:

$ lava-tool compare-device-conf --wdiff --dispatcher-config-dir . devices/mustang01.yaml working.jinja2
$ lava-tool compare-device-conf --dispatcher-config-dir . devices/mustang01.yaml working.jinja2

Iterate through the changes, testing any changes to the working.jinja2 at each stage, until you have no differences between the generated YAML and the working jinja2.

Pay particular attention to whitespace and indentation which have a direct impact on the structure of the object represented by the file. wdiff output is very useful for identifying content changes and it is often necessary to change the order of fields within a single command to get an appropriate match, even if that order has no actual effect. By ensuring that the content does match, it allows the comparison to show other changes like indents. Be prepared to change both the working.jinja2 and the template so that the indenting is the same in each even after commands have been substituted.

Note

The snippets here are just examples. In particular, formatting these examples for the documentation has changed some of the indents, so take particular care to compare and fix the indents of your files and ensure that your working YAML file continues to work as well as to match the output of the template.

Adapting the base commands to the device type

base.jinja2 for most devices uses the command base_uboot_commands which expands to:

- setenv autoload no
- setenv initrd_high '0xffffffff'
- setenv fdt_high '0xffffffff'

This command works well on 32-bit systems, on the mustang, it causes:

- {target: ERROR: Failed to allocate 0xa38c bytes below 0xffffffff.}
- {target: Failed using fdt_high value for Device TreeFDT creation failed! hanging...### ERROR ### Please RESET the board ###}

So the mustang template simply omits base_uboot_commands, using:

- setenv autoload no

Completing the migration

The device dictionary and the template need to be introduced into the lava-server configuration and database entries created for the device type and device. Helpers may be implemented for this in due course but the process involves:

  1. Add a device type to lava_scheduler_app in the admin interface

  2. Populate fields (you can omit health check for now - pipeline health checks are not yet ready).

  3. Add a device of the specified type to lava_scheduler_app in the admin interface. Set the device as a pipeline device by checking the “Pipeline Device” box.

  4. Add the template to the lava-server configuration:

    $ sudo cp device-types/mustang.jinja2 /etc/lava-server/dispatcher-config/device-types/
    
  5. Import the device dictionary to provide the device-specific configuration:

    $ sudo lava-server manage device-dictionary --hostname mustang1 --import mustang1.yaml
    
  6. Review the generated YAML:

    $ sudo lava-server manage device-dictionary --hostname mustang1 --review
    
  7. Submit a test job against localhost and ensure it runs to completion:

    $ lavacli jobs submit mustang-nfs.yaml
    
  8. Offer the new template as a code review against lava-server.