The standard text is Applied Logistic Regression 3rd ed (2013) by David W. Hosmer, Jr., Stanley Lameshow and Rodney X. Sturdivant. On page 168, they state
We feel quite strongly that [goodness of fit tests] should not be used to build [logistic] models the likelihood ratio tests for significance of coefficients are much more powerful and appropriate.
Their key message in evaluating a variable is "does the model perform better with or without its inclusion?" Their recommended methods are log likelihood, score and Wald.
For a highly technical subject, the text is very understandable. Two R oriented texts, Introductory Statistics with R by Peter Dalgaard and *Practical Statistics for Data Scientists" by Peter Bruce and Andrew Bruce have the best treatment of logistic regression that I've seen for R, but neither goes into the level of detail as Hosmer et al.