Kruskal-wallis Test is a non-parametric method. It does not assume a normal distribution of the residuals. This is not like the analogous one-way analysis of variance(AVOVA). If you meet the assumptions, you can use Kruskal-Waillis
There are two common non-parametric tests that are analogous to two-way anova: Friedman Test and Quade Test. However, each is only for un-replicated block designs. So, probably not what you want in this case.
Another approach is to use a robust method. See the following link for an an example of two-way anova. But it is not clear to me now far from parametric assumptions this kind of robust estimation allows.
http://rcompanion.org/rcompanion/d_08a.html
You might look at the Scheirer-Ray-Hare extension of the K-W test mentioned at the link below. I am not familiar with it.
Finally, you might want to use ordinal regression, especially if your "job satisfaction" is really a Likert response. Ordinal regression as a generalized K-W test is mentioned at the link below. I have used the ordinal package in R for this, and it is both powerful and relatively easy to use. If use R, this would be my recommendation.