The goal of lava-docker is to simplify the install and maintenance of a LAVA lab in order to participate in distributed test efforts such as kernelCI.org.
With lava-docker, you describe the devices under test (DUT) in a simple YAML file, and then a custom script will generate the necessary LAVA configuration files automatically.
Similarly, LAVA users and authentication tokens are described in (another) YAML file, and the LAVA configurations are automatically generated.
This enables the setup of a LAVA lab with minimal knowledge of the underlying LAVA configuration steps necessary.
lava-docker has currently been tested primarily on Debian stable (buster). The following packages are necessary on the host machine:
- docker
- docker-compose
- pyyaml
If you plan to use docker/fastboot tests, you will need probably also to install lava-dispatcher-host.
Example to use lava-docker with only one QEMU device:
- Checkout the lava-docker repository
- Generate configuration files for LAVA, udev, serial ports, etc. from boards.yaml via
./lavalab-gen.py
- Go to output/local directory
- Build docker images via
docker-compose build
- Start all images via
docker-compose up -d
-
Once launched, you can access the LAVA web interface via http://localhost:10080/. With the default users, you can login with admin:admin.
-
By default, a LAVA healthcheck job will be run on the qemu device. You will see it in the "All Jobs" list: http://localhost:10080/scheduler/alljobs
-
You can also see full job output by clicking the blue eye icon ("View job details") (or via http://localhost:10080/scheduler/job/1 since it is the first job ran)
-
For more details, see https://validation.linaro.org/static/docs/v2/first-job.html
-
Note that the port used above can change depending on the setting of webinterface_port in boards.yaml
To add a board you need to find its device-type, standard naming is to use the same as the official kernel DT name. (But a very few DUT differ from that)
You could check in https://github.com/Linaro/lava/tree/master/etc/dispatcher-config/device-types if you find yours.
Example: For a beagleboneblack, the device-type is beaglebone-black (Even if official DT name is am335x-boneblack) So you need to add in the boards section:
- name: beagleboneblack-01
type: beaglebone-black
Next step is to gather information on UART wired on DUT.
If you have a FTDI, simply get its serial (visible in lsusb -v or for major distribution in dmesg)
For other UART type (or for old FTDI without serial number) you need to get the devpath attribute via:
udevadm info -a -n /dev/ttyUSBx |grep ATTRS|grep devpath | head -n1
Example with a FTDI UART:
[ 6.616707] usb 4-1.4.2: New USB device strings: Mfr=1, Product=2, SerialNumber=3
[ 6.704305] usb 4-1.4.2: SerialNumber: AK04TU1X
The serial is AK04TU1X
So you have now:
- name: beagleboneblack-01
type: beaglebone-black
uart:
idvendor: 0x0403
idproduct: 0x6001
serial: AK04TU1X
Example with a FTDI without serial:
[2428401.256860] ftdi_sio 1-1.4:1.0: FTDI USB Serial Device converter detected
[2428401.256916] usb 1-1.4: Detected FT232BM
[2428401.257752] usb 1-1.4: FTDI USB Serial Device converter now attached to ttyUSB1
udevadm info -a -n /dev/ttyUSB1 |grep devpath | head -n1
ATTRS{devpath}=="1.5"
So you have now:
- name: beagleboneblack-01
type: beaglebone-black
uart:
idvendor: 0x0403
idproduct: 0x6001
devpath: "1.5"
Final step is to manage the powering of the board.
Many PDU switchs could be handled by a command line tool which control the PDU.
You need to fill boards.yaml with the command line to be ran.
Example with an ACME board: If the beagleboneblack is wired to port 3 and the ACME board have IP 192.168.66.2:
pdu_generic:
hard_reset_command: /usr/local/bin/acme-cli -s 192.168.66.2 reset 3
power_off_command: /usr/local/bin/acme-cli -s 192.168.66.2 power_off 3
power_on_command: /usr/local/bin/acme-cli -s 192.168.66.2 power_on 3
beagleboneblack, with FTDI (serial 1234567), connected to port 5 of an ACME
- name: beagleboneblack-01
type: beaglebone-black
pdu_generic:
hard_reset_command: /usr/local/bin/acme-cli -s 192.168.66.2 reset 5
power_off_command: /usr/local/bin/acme-cli -s 192.168.66.2 power_off 5
power_on_command: /usr/local/bin/acme-cli -s 192.168.66.2 power_on 5
uart:
idvendor: 0x0403
idproduct: 0x6001
serial: 1234567
The basic setup is composed of a host which runs the following docker images and DUT to be tested.
- lava-master: run lava-server along with the web interface
- lava-slave: run lava-dispatcher, the component which sends jobs to DUTs
The host and DUTs must share a common LAN.
The host IP on this LAN must be set as dispatcher_ip in boards.yaml.
Since most DUTs are booted using TFTP, they need DHCP for gaining network connectivity.
So, on the LAN shared with DUTs, a running DHCPD is necessary. (See DHCPD below)
Lava-docker support multi-host architecture, Master and slaves could be on different host.
Lava-docker support multiples slaves, but with a maximum of one slave per host. This is due to that slave need TFTP port accessible from outside.
You need to have a PDU for powering your DUT. Managing PDUs is done via pdu_generic
By default, the following ports are used by lava-docker and are proxyfied on the host:
- 69/UDP proxyfied to the slave for TFTP
- 80 proxyfied to the slave for TODO (transfer overlay)
- 5500 proxyfied to the slave for Notification
- 5555 proxyfied to the master (LAVA logger)
- 5556 proxyfied to the master (LAVA master)
- 10080 proxyfied to the master (Web interface)
- 55950-56000 proxyfied to the slave for NBD
A DHCPD service is necessary for giving network access to DUT.
The DHCPD server could be anywhere with the condition that it is accessible of DUTs. (Could be on host, in a docker in the host, or is the ISP box on the same LAN.
Router: 192.168.1.1 which handle DHCP for 192.168.1.10-192.168.1.254
Lab: 192.168.1.2
So the dispatcher_ip is set to 192.168.1.2
Lab: 192.168.1.2 which handle DHCP for 192.168.1.10-192.168.1.254
So the dispatcher_ip is set to 192.168.1.2
A dedicated LAN is used for DUTs. (192.168.66.0/24) The host have two NIC:
- eth0: (192.168.1.0/24) on home LAN. (The address could be static or via DHCP)
- eth1: (192.168.66.0/24) with address set to 192.168.66.1
On the host, a DHCPD give address in range of 192.168.66.3-192.168.66.200
So the dispatcher_ip is set to 192.168.66.1
A sample isc-dhcpd-server DHCPD config file is available in the dhcpd directory.
Simply set interface=interfacename where interfacename is your shared LAN interface
You can use the lavalab-gen.sh helper script which will do all the above actions for you.
This file describe how the DUTs are connected and powered.
masters:
- name: lava-master name of the master
host: name name of the host running lava-master (default to "local")
version: "202x.xx" LAVA version for master
webadmin_https: Does the LAVA webadmin is accessed via https
webinterface_port: Port number to use for the LAVA web interface (default to "10080")
lava-coordinator: Does the master should ran a lava-coordinator and export its port
persistent_db: true/false (default false) Is the postgresql DB is persistent over reboot
pg_lava_password: The postgresql LAVA server password to set
http_fqdn: The FQDN used to access the LAVA web interface. This is necessary if you use https otherwise you will issue CSRF errors.
healthcheck_url: Hack healthchecks hosting URL. See hosting healthchecks below
build_args:
- line1 A list of line to set docker build-time variables
- line2
allowed_hosts: A list of FQDN used to access the LAVA master
- "fqdn1"
- "fqdn2"
loglevel:
lava-logs: DEBUG/INFO/WARN/ERROR (optional) select the loglevel of lava-logs (default to DEBUG)
lava-slave: DEBUG/INFO/WARN/ERROR (optional) select the loglevel of lava-slave (default to DEBUG)
lava-master: DEBUG/INFO/WARN/ERROR (optional) select the loglevel of lava-master (default to DEBUG)
lava-server-gunicorn: DEBUG/INFO/WARN/ERROR (optional) select the loglevel of lava-server-gunicorn (default to DEBUG)
users:
- name: LAVA username
token: The token of this user (optional)
password: Password the this user (generated if not provided)
email: email of the user (optional)
superuser: yes/no (default no)
staff: yes/no (default no)
groups:
- name: Name of the group this user should join
groups:
- name: LAVA group name
submitter: True/False Can this group can submit jobs
tokens:
- username: The LAVA user owning the token below. (This user should be created via users:)
token: The token for this callback
description: The description of this token. This string could be used with LAVA-CI.
smtp: WARNING: Usage of an SMTP server makes it mandatory for each user to have an email address
email_host: The host to use for sending email
email_host_user: Username to use for the SMTP server
email_host_password: Password to use for the SMTP server
email_port: Port to use for the SMTP server (default: 25)
email_use_tls: Whether to use a TLS (secure) connection when talking to the SMTP server
email_use_ssl: Whether to use an implicit TLS (secure) connection when talking to the SMTP server
email_backend: The backend to use for sending emails (default: 'django.core.mail.backends.smtp.EmailBackend')
slaveenv: A list of environment to pass to slave
- name: slavename The name of slave (mandatory)
env:
- line1 A list of line to set as environment
- line2
slaves:
- name: lab-slave-XX The name of the slave (where XX is a number)
host: name name of the host running lava-slave-XX (default to "local")
version: "202x.xx" LAVA version for worker
dispatcher_ip: the IP where the slave could be contacted. In lava-docker it is the host IP since docker proxify TFTP from host to the slave.
remote_master: the name of the master to connect to
remote_address: the FQDN or IP address of the master (if different from remote_master)
remote_rpc_port: the port used by the LAVA RPC2 (default 80)
remote_user: the user used for connecting to the master
remote_user_token: The remote_user's token. This option is necessary only if no master node exists in boards.yaml. Otherwise lavalab-gen.py will get from it.
remote_proto: http(default) or https
lava_worker_token: token to authenticate worker to master/scheduler (LAVA 2020.09+)
default_slave: Does this slave is the default slave where to add boards (default: lab-slave-0)
bind_dev: Bind /dev from host to slave. This is needed when using some HID PDU
use_tftp: Does LAVA need a TFTP server (default True)
use_nbd: Does LAVA need a NBD server (default True)
use_overlay_server: Does LAVA need an overlay server (default True)
use_nfs: Does the LAVA dispatcher will run NFS jobs
use_tap: Does TAP netdevices could be used
use_docker: Permit to use docker commands in slave
arch: The arch of the worker (if not x86_64), only accept arm64
host_healthcheck: If true, enable the optional healthcheck container. See hosting healthchecks below
lava-coordinator: Does the slave should ran a lava-coordinator
expose_ser2net: Do ser2net ports need to be available on host
custom_volumes:
- "name:path" Add a custom volume
expose_ports: Expose port p1 on the host to p2 on the worker slave.
- p1:p2
extra_actions: An optional list of action to do at end of the docker build
- "apt-get install package"
build_args:
- line1 A list of line to set docker build-time variables
- line2
env:
- line1 A list of line to set as environment (See /etc/lava-server/env.yaml for examples)
- line2
tags: (optional) List of tag to set on all devices attached to this slave
- tag1
- tag2
devices: A list of devices which need UDEV rules
- name: The name of the device
vendorid: The VID of the UART (Formated as 0xXXXX)
productid: the PID of the UART (Formated as 0xXXXX)
serial: The serial number of the device if the device got one
devpath: The UDEV devpath to this device if more than one is present
boards:
- name: devicename Each board must be named by their device-type as "device-type-XX" (where XX is a number)
type: the LAVA device-type of this device
slave: (optional) Name of the slave managing this device. Default to first slave found or default_slave if set.
kvm: (For qemu only) Does the qemu could use KVM (default: no)
uboot_ipaddr: (optional) a static IP to set in uboot
uboot_macaddr: (Optional) the MAC address to set in uboot
custom_option: (optional) All following strings will be directly append to devicefile included in {% opt %}
- "set x=1"
raw_custom_option: (optional) All following strings will be directly append to devicefile
- "{% set x=1 %}"
tags: (optional) List of tag to set on this device
- tag1
- tag2
aliases: (optional) List of aliases to set on the DEVICE TYPE.
- alias1
- alias2
user: (optional) Name of user owning the board (LAVA default is admin) user is exclusive with group
group: (optional) Name of group owning the board (no LAVA default) group is exclusive with user
# One of uart or connection_command must be choosen
uart:
idvendor: The VID of the UART (Formated as 0xXXXX)
idproduct: the PID of the UART (Formated as 0xXXXX)
serial: The serial number in case of FTDI uart
baud: (optional) Change the baud rate of the this uart (default is 115200)
devpath: the UDEV devpath to this uart for UART without serial number
interfacenum: (optional) The interfacenumber of the serial. (Used with two serial in one device)
use_conmux: True/False (Use conmux-console instead of ser2net)
use_ser2net: True/False (Deprecated, ser2net is the default uart handler)
ser2net_keepopen: True/False (optional) Use the recent ser2net keepopen
ser2net_options: (optional) A list of ser2net options to add
- option1
- option2
connection_command: A command to be ran for getting a serial console
pdu_generic:
hard_reset_command: commandline to reset the board
power_off_command: commandline to power off the board
power_on_command: commandline to power on the board
Notes on UART:
- Only one of devpath/serial is necessary.
- For finding the right devpath, you could use
udevadm info -a -n /dev/ttyUSBx |grep devpath | head -n1
- VID and PID could be found in lsusb. If a leading zero is present, the value must be given between double-quotes (and leading zero must be kept) Example:
Bus 001 Device 054: ID 0403:6001 Future Technology Devices International, Ltd FT232 Serial (UART) IC
This device must use "0403" for idvendor and 6001 for idproduct.
- Some boards reset serial on power on. This can cause ser2net/telnet to disconnect resulting in the LAVA Worker being unable to program the board. This may be mitigated by passing LOCAL as an option to ser2net in the boards.yaml. Example:
ser2net_options:
- LOCAL
Note on connection_command: connection_command is for people which want to use other custom way than ser2net to handle the console.
Examples: see boards.yaml.example or boards.yaml.minimal
lavalab-gen.py
this script will generate all necessary files in the following locations:
output/host/lava-master/tokens/ This is where the callback tokens will be generated
output/host/lava-master/users/ This is where the users will be generated
output/host/lab-slave-XX/conmux/ All files needed by conmux
output/host/lab-slave-XX/devices/ All LAVA devices files
output/host/udev/99-lavaworker-udev.rules udev rules for host
output/host/docker-compose.yml Generated from docker-compose.template
All thoses files (except for udev-rules) will be handled by docker.
You can still hack after all generated files.
Note that the udev-rules are generated for the host, they must be placed in /etc/udev/rules.d/ They are used for giving a proper /dev/xxx name to tty devices. (where xxx is the board name) (lavalab-gen.sh will do it for you)
To build all docker images, execute the following from the directory you cloned the repo:
docker-compose build
For running all images, simply run:
docker-compose up -d
A squid docker is provided for caching all LAVA downloads (image, dtb, rootfs, etc...)
For the moment, it is unsupported and unbuilded.
For using an external squid server see "How to made LAVA slave use a proxy" below
All upstream LAVA patches could be backported by placing them in lava-master/lava-patch/
For backupping a running docker, the "backup.sh" script could be used. It will store boards.yaml + postgresql database backup + joboutputs.
For restoring a backup, postgresql database backup + joboutputs must be copied in master backup directory before build.
Example: ./backup.sh This produce a backup-20180704_1206 directory For restoring this backup, simply cp backup-20180704_1206/* output/local/master/backup/
For upgrading between two LAVA version, the only method is:
- backup data by running ./backup.sh on the host running the master (See Backups / restore)
- checkout the new lava-docker and update your boards.yaml
- Move the old output directory away
- run lavalab-gen.sh
- copy your backup data in output/yourhost/master/backup directory
- build via docker-compose build
- Stop the old docker via docker-compose down
- Run the new version via docker-compose up -d
- Check everything is ok via docker-compose logs -f
Note that this container provides defaults which are unsecure. If you plan on deploying this in a production environment please consider the following items:
- Changing the default admin password (in tokens.taml)
- Using HTTPS
- Re-enable CSRF cookie (disabled in lava-master/Dockerfile)
Since LAVA upstream provides only amd64 and arm64 debian packages, lava-docker support only thoses architectures. For building an arm64 lava-docker, some little trick are necessary:
- replace "baylibre/lava-xxxx-base" by "baylibre/lava-xxxx-base-arm64" for lava-master and lava-slave dockerfiles
For building lava-xxx-base images
- replace "bitnami/minideb" by "arm64v8/debian" on lava-master-base/lava-slave-base dockerfiles.
You need to set use_nfs: True on slave that will ran NFS jobs. A working NFS server must be working on the host. Furthermore, you must create a /var/lib/lava/dispatcher/tmp directory on the host and export it like: /var/lib/lava/dispatcher/tmp 192.168.66.0/24(no_root_squash,rw,no_subtree_check)
You can add custom or backported LAVA patchs in lava-master/lava-patch Doing the same for lava-slave will be done later.
There are two way to add custom devices types.
- Copy a device type file directly in lava-master/device-types/ If you have a brand new device-type, it is the simpliest way.
- Copy a patch adding/modifying a device-type in lava-master/device-types-patch/ If you are modifying an already present (upstream) device-type, it is the best way.
Add env to a slave like: slave: env:
- "http_proxy: http://dns:port"
Or on master via
slaveenv:
- name: lab
env:
- "http_proxy: http://squid_IP_address:3128"
- "https_proxy: http://squid_IP_address:3128"
- name: lab
env:
All boards which uses PXE, could be used with LAVA via grub. But you need to add a configuration in your DHCP server for that board. This configuration need tell to the PXE to get GRUB for the dispatcher TFTP. Example for an upsquare and a dispatcher available at 192.168.66.1:
host upsquare {
hardware ethernet 00:07:32:54:41:bb;
filename "/boot/grub/x86_64-efi/core.efi";
next-server 192.168.66.1;
}
Healthchecks jobs needs externals resources (rootfs, images, etc...). By default, lava-docker healthchecks uses ones hosted on our github, but this imply usage of external networks and some bandwidth. For hosting locally healthchecks files, you can set healthcheck_host on a slave for hosting them. Note that doing that bring some constraints:
- Since healthchecks jobs are hosted by the master, The healthcheck hostname must be the same accross all slaves.
- You need to set the base URL on the master via healthcheck_url
- If you have qemu devices, Since they are inside the docker which provides an internal DNS , you probably must use the container("healthcheck") name as hostname.
- In case of a simple setup, you can use the slave IP as healthcheck_url
- In more complex setup (slave sprayed on different site with different network subnets) you need to set a DNS server for having the same DNS available on all sites.
For setting a DNS server, the easiest way is to use dnsmasq and add in /etc/hosts "healthcheck ipaddressoftheslave"
Example: One master and slave on DC A, and one slave on DC B. Both slave need to have healthcheck_host to true and master will have healthcheck_url set to http://healthcheck:8080 You have to add a DNS server on both slave with an healthcheck entry.
The preferred way to submit bugs are via the github issue tracker You can also contact us on #lava-docker on the Libera.chat IRC network