On Thu, 11 Oct 2018 at 13:50, john zhang email@example.com wrote:
For example, to test a quality of hardware peripheral on the board, If multiple boards are involved in the test from same job at the same time, the error rate can be quickly obtained.
You will need to manage that in the client doing the submissions using XML-RPC calls to determine how many devices of the device-type(s) you want are Idle at the time.
However, it is actually more efficient to just submit as many test jobs as you need and let the queue sort it out.
At 2018-10-11 20:34:26, "Milosz Wasilewski" firstname.lastname@example.org wrote:
On Thu, 11 Oct 2018 at 13:17, john zhang email@example.com wrote:
The scheduler assigns the test job to the first available device of the requested device-type.
Does that mean the scheduler select only one device to run the test job, other devices with same device-type are not run this job even though in idle status? Is there a method that all devices with same device-type can run the same job as long as available? That is to say, a job is executed "simultaneously" by multiple boards.
do you mean sharding or exact copy of the same job? LAVA doesn't support sharding because it's not aware of the test suite that is supposed to be executed. If you want exact copy, just submit as many identical jobs as you have devices. You can get the number of available devices through API.
Out of curiosity, why would you want to execute the same test job on multiple devices simultaneously?