EM should handle different experiments on the same deployment

[chandrams]

Different experiments on the same deployment should be handled by EM. Since it is the same deployment, EM should run the exps sequentially?

[bharathappali]

In the earlier design of the EM there is a waiting queue which stores the experiment trials whose target deployment is currently in a existing experiment, once the experiment trial is complete the trial of the other experiment is started. A similar design implementation can handle this scenerio.

[bharathappali]

EM should run the exps sequentially?

This might depend on the type of deployment in the trial, if the type is a new deployment then EM run sequentially, if it's a rolling update and there is a trial running currently then the EM should wait for launching the new trial, till the old trial completes.

[msvinaykumar]

Bharath Can you confirm if earlier design using task polling technique to hold experiment in queue from running until another experiment of same deployment gets completed ?

[bharathappali]

Earlier implementation used to set a notification after completion if there is an existing trial. Ref:implementation here and used to launch the waiting experiments upon completion of the existing trial Ref:implementation here . The check happens in RunExperiment.java Ref:implementation here

It's not strictly a polling mechanism but the check happens after the completion of existing to see if same deployment trails are in waiting queue.

[msvinaykumar]

Ok thanks Bharath .. Thats good, pooling mechanism is not required in current design too and extra Wating queue is also not required. Will explain in detail in next design call.