On 18 January 2018 at 09:24, Chetan Sharma <chetan.sharma@timesys.com> wrote:
Hi Neil/Senthil 
    Thank you for sharing complete process to install Lava-coordinator on Existing System. 

    I have found one limitation while Executing a MultiNode Job on different devices. If a device is not allocated to any subjob of MultiNode job, then all other job stays in Submitted state allocates device. 


The MultiNode job *must* wait until all devices defined in the MultiNode group are available for scheduling. It is a requirement of MultiNode scheduling.
 

    Can we execute subjobs of Multinode Job which has allocated Devices after specific time or immediately after allocation of device. 

No. There is no guarantee that this would happen within the timeout set for the test job. All subjobs must start at the same time and that only happens when all devices are available for scheduling.
 

    I was reading about concept of lava-start https://validation.linaro.org/static/docs/v2/actions-protocols.html#lava-start-api-call . But not able to make use of it with UseCase defined as above. 


Delayed start is only for situations where all devices are available for scheduling at the start of the test job but one or more role needs to wait until another role has not only deployed and booted but also set up a running service.
 
 

On Wed, Jan 17, 2018 at 1:28 PM, Neil Williams <neil.williams@linaro.org> wrote:
On 17 January 2018 at 06:56, Senthil Kumaran S <senthil.kumaran@linaro.org> wrote:


On Wednesday 17 January 2018 12:01 PM, Chetan Sharma wrote:
>                 Can anyone share a step by step process to install
> Lava-coordinator in existing Setup to execute multinode job.

More information about your existing setup will be helpful here, like
the base operating system, lava-server and lava-dispatcher versions, etc.

>             Existing Setup :-
>               1. Lava-Server (Machine 1)

As of lava-server and lava-dispatcher version 2018.1, lava-coordinator
should be installed only on the lava-server machine. It is not required
on the workers.

An installation of lava-server should automatically pull
lava-coordinator as a dependent package. If for some reason you want to
install it separately, then,

$ sudo apt install lava-coordinator

After installation of lava-coordinator on the same machine which runs
lava-server, check /etc/lava-coordinator/lava-coordinator.conf which
should look like the following:

<snip>
{
    "port": 3079,
    "blocksize": 4096,
    "poll_delay": 3,
    "coordinator_hostname": "localhost"
}
</snip>


The additional step is to copy that configuration file to each worker, changing the coordinator_hostname to the hostname of the master, so that each worker can find the coordinator. lava-coordinator should NOT be installed on the worker itself, just the configuration file.



 



--



--
Thanks & Regards
 Chetan Sharma



--