Age | Commit message (Collapse) | Author | Files | Lines |
|
LAVA 2020.09 dropped ZMQ and use now a token to authenticate worker
against master.
lava-docker already handle this by getting token via lavacli.
But we need to support to set token in boards.yaml, this patch permits
this.
Furthermore, the token was given to worker via --token, but this is bad
since token can be found by anyone running ps.
A better secure way is to use --token-file.
|
|
The version of cu/ser2net we want are already in buster now, so no need
to play with testing/sid.
Furthermore, sid give a new ser2net with incompatible config file.
|
|
Since 2020.09, workers does not use ZMQ anymore but use a token.
This patch permit our setup to detecte when we use LAVA 2020.09+ and retrieve it from master.
|
|
When I have added screen support, it was a hack for made M3ulcb serial console works.
But for a long time ser2net works on m3ulcb.
So lava-docker discouraged usage of screen and now it is time to remove
it.
|
|
There is no need to listen on ipv6 inside docker.
Furtheremore, this break lava-docker on non-ipv6 machine.
|
|
Signed-off-by: Corentin LABBE <clabbe@baylibre.com>
|
|
|
|
If there are no entrypoint to copy, the COPY docker action will fail.
For fixing this issue, simply copy all files instead of just "*sh".
Since the entrypoint.d directory is not empty (got an empty file) it
will always works.
The LAVA script run only *sh files so we are still safe.
|
|
|
|
This patch permit to adds custom entrypoints.
|
|
|
|
|
|
|
|
|
|
This patch adds support for the device-type aliases.
|
|
This patch upgrade LAVA to 2019.07 via their official docker images.
Along with the change of the baseimage from our lava-xx-base to official
2019.07, some minor changes are needed:
- Activate the en_US.UTF-8 locale needed for postgresql
- chown to lavaserver all copied device-types
- Fix the start scripts for using the official entrypoints
|
|
When /var/lib/lava/dispatcher/tmp is a volume (like when using NFS), it
masks all PXE stuff installed during the build.
For preventing this, this patchs made this data installed at runtime.
|
|
|
|
This patchs adds a new flag "lava-coordinator" for slave which enables
the generation of a lava-coordinator.conf file.
|
|
When checkouting tests, git is necessary.
|
|
biosdisk is not provided in default install and is not used, remove it.
|
|
Since 2019.03 is ready, use it
|
|
The package grub-efi-amd64-bin is only availlable on amd64.
Installing it on arm64 lead to a build failure.
This patch permits to install it even on arm64
|
|
|
|
|
|
|
|
This patch adds support for booting from PXE
|
|
By default, all process use the DEBUG loglevel, this could be too much
for a production server.
This patch permits to tune the loglevel.
|
|
The first action need to be an apt-get update since we dont know how old
is the base image.
|
|
This patch made lava-docker use a known to work dockerhub tag.
Keep the latest tag for internal work.
|
|
Since we use the lavacli package, there are no reason to chdir in
/root/lavacli.
|
|
A recent enough lavacli is present on stretch-backports.
It is better to use it.
This will fix also some commands not working with recent LAVA server.
|
|
Previoulsy it was possible to backport LAVA patch only on master, this patch permits it on
slave.
|
|
extra_actions is broken since commit f3d53d64922d ("lava-slave/Dockerfile: copy all scripts in one actions")
It is not anymore copied in /root but in /usr/local/bin/
|
|
This patch permit to add a group on boards
|
|
By default LAVA give owning of boards to admin, this patch permit to
give board to a specific user.
|
|
|
|
This patch permits to add tags to device
Signed-off-by: Corentin Labbe <clabbe@baylibre.com>
|
|
The ZMQ key file names are expected to have specific names in the
containers (e.g. $LAVA_MASTER.key). However, when using existing key
files, they are simply copied into the containers, and if they don't
match the exact requirements, encryption will silently fail.
Fix this by allowing arbitrary filenames for existing keys, but ensure
they are copied into the container with the expected filenames.
Related, when using auto-generated keys, the generated master key is
simply "master.key" in the slave. Fix this by ensuring that
"master.key" is copied to $LAVA_MASTER.key when the slave container
starts.
Signed-off-by: Kevin Hilman <khilman@baylibre.com>
|
|
Current ser2net present in debian is too old, we need at least the 3.2
release for having the multiple connection support.
This patch download, compile and install ser2net 3.5
|
|
When trying to detect if master is up, we have set a timeout of 60s.
This is good for most of the time, but when the master is restoring a
database, 60s is not enougth.
This patch increase timeout to 300s.
|
|
When re-adding devices to a worker, setup.sh re-add them with UNKNOWN
status.
This patch prevent this for RETIRED/MAINTENANCE status and keep it.
|
|
Instead of calling devices/device-types list for each device, call them
once.
|
|
|
|
The process of building LAVA is hard to maintain.
But since we need to be able to go back in time
(or just stay longer with a specific version)
we need a way to keep a LAVA version.
For achieving this, we build a minimal image with just LAVA
and tag it with the version of LAVA inside.
|
|
This patch add support for using ZMQ auth.
Basicly adding "zmq_auth: True" to a master is sufficient to enable it.
Since "ZMQ certificates" are using a custom format (vs X509 classic), we need to use the custom generator.
For helping with that a temporary docker is generated which handle generating thoses files.
|
|
This patch add support for using screen instead of conmux-console.
screen can be used by adding "use_screen: True" to uart.
screen is necessary for some board with internal uart which have a micro
cut when switching power (like DRA7-EVM and M3ULCB)
|
|
This patch add support for ser2net.
ser2net can be used by adding "use_ser2net: True" to uart.
|
|
This patch upgrade lava-docker to LAVA 2018.4
|
|
This patch implement multiple slave support.
Instead of having a maximum of one master and one slave in one docker
image, it is now possible to have multiple slave accross several docker
host.
For helping this change, a new boards.yaml format is introduced (See
README.md for details)
Note that tokens.yaml is also squashed in boards.yaml
|