The state machine describes and controls the state and health of Workers, Devices and Test Jobs.
For each worker, two variables describe the current status:
state:health:state is an internal variable, set by lava-master when receiving (or
not) pings from each worker.
Caution
When a worker is in Offline, none of the attached devices will be used to schedule new jobs.
health can be used by admins to control the health of all attached
devices. For instance, when set to Maintenance, all attached devices will be
automatically put into maintenance mode so that no jobs will be scheduled on
those devices.
For each device, two variables describe the current status:
state:health:state is an internal variable, set by lava-server-gunicorn and lava-scheduler
when scheduling, starting, canceling and ending test jobs.
health can be used by admins to indicate if a device should be used by the
scheduler or not. Moreover, when ending an health-check, the device health will
be set according to the test job health.
For each test job, two variables are describing the current status:
state:Note
Only multinode test jobs use Scheduling. When all
sub-jobs are in Scheduling, lava-master will transition all test
jobs to Scheduled.
health:The scheduler is called by lava-master approximately every 20 seconds.
The scheduler starts by scheduling health-checks. The remaining devices are
then considered for test jobs.
To ensure that health-checks are always scheduled when needed, they will be considered first by the scheduler before regular test jobs.
The scheduler will only consider devices where:
Note
A device whose health is Bad, Maintenance or Retired is
never considered by the scheduler when it is looking for devices to run test
jobs
The scheduler will only consider devices where:
state is Idlehealth is Good or Unknownstate is Online