In Open Source Software (OSS) communities, coordination happens based on artifact-mediated interactions among geographically dispersed individuals. Prior literature on OSS and software evolution argues that that the coordination structures that emerge from dyadic coordination relationships may emerge from endogenous non-random relational mechanisms. The objective of this research is to understand whether and how relational endogenous mechanisms predict the formation of coordination relationships in evolving networks of software developers. We implemented an Exponential Random Graph Model (p*) and studied the significance of antecedent factors of coordination patterns among software developers. Our empirical analysis encompasses five years of development activity between 2012 and 2016 involving 619 developers in Nova, one of the oldest packages in OpenStack. In contrast to other network formation process, we find evidence of non-linear rather than linear preferential attachment among software developers. Knowledge similarity of two developers also has a significantly positive effect on the formation of coordination relationships. Both mechanisms unfold differently over time: While non-linear preferential attachment explains the formation of coordination relationships across different evolutionary periods, does knowledge similarity only sustain its effect in the early stages of OSS coordination. Our findings contribute to the scholarship on OSS and online communities more broadly.