Whenever you use partial differentiation, the coordinate system must be specified at the outset. Changes of variables must be handled by the multivariable chain rule.
Start by writing everything down as functions. If you have $f(\sigma_1,\sigma_2) = \sigma_1+\sigma_2$ then it's certainly the case that $\partial_{\sigma_1}f = 1.$ How you express this in terms of $x_1$ and $x_2$ is to write $$g(x_1,x_2)= f(\sigma_1(x_1,x_2),\sigma_2(x_1,x_2))$$ so that, say, $$ \frac{\partial}{\partial x_1}g(x_1,x_2) = \frac{\partial \sigma_1}{\partial{x_1}}\frac{\partial f}{\partial{\sigma_1}}+\frac{\partial \sigma_2}{\partial{x_1}}\frac{\partial f}{\partial{\sigma_2}}$$
where $\partial_{\sigma_1}f$ and $\partial_{\sigma_2}f$ are evaluated at $(\sigma_1(x_1,x_2),\sigma_2(x_1,x_2)).$
So in the case where $f(\sigma_1,\sigma_2) = \sigma_1+\sigma_2,$ the partial derivatives are both one and you should have $$ \frac{\partial}{\partial x_1}g(x_1,x_2) =\frac{\partial \sigma_1}{\partial{x_1}}+\frac{\partial \sigma_2}{\partial{x_1}}.$$
In your example, you say $z = x+y,$ which in my / your original notation corresponds to $f(\sigma_1,\sigma_2) = \sigma_1 +\sigma_2$. And then you say $y= 2x,$ which means we're taking $\sigma_1(x,y) = x$ and $\sigma_2(x,y) = 2x.$ (note this is not a change of variables, but the above all still holds.) So the formula gives $\partial_x g(x,y) = 3,$ not 1.
(Note this could be come to the answer more easily by substituting $x$ and $y$ in at the outset and noting $g(x,y) = 3x,$ but the chain rule works in more general cases)