Exactly, i deal with a few of data processing programs, wish my more experienced colleagues told me about these algorithms before, but doesn't this algorithm only compute an approximation and not the exact value ?

To about the extent that a 64-bit float can ever be said to contain an exact value, this algorithm will compute the exact value. There's no sampling.

If readers would like to read an implementation, one is here: https://github.com/sharpobject/runningstats

It includes skewness and kurtosis.