This really depends on your sample size. There is a lot of literature describing the sensitivity of Shapiro-Wilk and other Normality tests. My personal opinion based on the literature and my own simulations is that if your data set is large enough, these tests will always conclude that the data isn't normal "enough" (You can read more about what these tests are really assessing to understand why). So depending on your sample size and goals, i'd pay different amount of attention to the Shapiro-Wilk test.

Dear Fjodor. Thanks for your comment. I am applying this test for 30 years Annual rainfall data. I have different rainfall stations and elevation differs for each station. My Goal is to check normality, homogeneity and trends of these data sets.

If the data is measured annually (i.e. 30 years means 30 observations), and the data is substantially non-normal, i'd probably do any non-parametric (for example, permutation tests) when comparing means of different groups/stations. However that doesn't mean you can't check for trends over time with OLS or regular correlations; those don't explicitly require normality. I'd just be cautious with what you say about significance when p-values are not "very" small. If i were you, i'd make a bunch of plots for the data distributions and residuals (if you run a regression over vs time) and make sure the data distribution or residuals don't look too far off normality or white noise, respectively. If the data distributions look very far from normal, i'd use permutation test for p-values in all cases.

The bigger issue with the time series data might emerge if you are interested in associations between several time series instead of just trends over time; i.e. OLS can't handle the error correlation of time series well and you will be likely to find spurious associations. But that's only if you are looking for association between multiple time-dependent variables.

Use non-parametric methods which doesn't assume normality of the data. Or if you are intended to use parametric tests, you have to check for Normality.