Rolling or continuous surveying or sampling

I have been looking everywhere for an answer to this problem, so far with no luck. Hopefully the statistics gurus on this board will know the answer, here's hoping :)

So I have a large population of sample data, and more coming in all the time in a continuous stream. I would like to use both sets of data as the sample that I'm doing analysis against, but I would like to definitely include recently-added data so that short-term changes are reflected in the results of analysis.

In the end what I really want to do is get a valid sample from the population that will accurately reflect recent changes.

There are two ways I can think of to do this:

1) Take a random sample from the entire dataset (including recent additions) each time I want to do an analysis. I tend to think of this as continuous sampling because it is taking a random data sample over and over again.

or 2) Take a random sample from the entire dataset and another sample from recent data to ensure that I get data from that pool. I think of this as a rolling sample because while it includes data from the entire population it also includes a rolling sample subset that is continuously updated with fresh data as it comes in (perhaps updating the rolling sample randomly from the new data).

My only concern is that I want the sample to remain statistically valid, and I can’t find any examples of anyone manipulating the sample like that. You’d think this would be covered somewhere but I can’t find it. I posed the question to a “statistics expert” on LivePerson and he said he couldn’t help me.

Anyone care to venture an opinion?