From performance perspective what is the best architecture for an application writing data over OMF to PI - Send container call on each restart or check for the container existence which is time consuming when the number of containers are big?

Process for the OMF application is - Create Type message which is creating an AF template placeholder, then create containers which acts as the destination for our application's time-series data. The PI Web API creates PI Points in response to container messages and at last form the time-series data values and write them to the container created.

Now in my scenario I have a source from which I am reading the data which can update new tags also, so if I am doing a container creation call for all the existing container it is time consuming and also not sure if any -ve impact, also checking the tags existence takes lot of time, Any suggestions here to handle the scenario will be appreciating.