i found the the p value and FDR are lower using exactTest than using glmQLFTest in edgeR when comparing two groups , I want to know why the results is very different between exactTest and glmQLFTest?

Because they are different tests that make different assumptions about the underlying data. The exact test is similar to an exact binomial test, but allowing for over-dispersion. The glmQLFTest uses a quasi-likelihood F-test that incorporates uncertainty in the count estimation. The help pages for both functions have citations that you can read if you care to know more.