It does seem contested. The 2015 meta-meta-analysis takes an approach that I haven't seen before, which is to quantify the lack of nonsignificant results in the 2010 meta-analysis, showing that if it was a meta-analysis with little publication bias, there would have been far more nonsignificant results. They basically suggest to re-do the reproduction of the original 1998 study:
We hope that the findings we present here will motivate researchers to re-examine the replicability and the magnitude of the depletion effect. Because our findings suggest that very large experiments will produce estimates of the depletion effect that are approximately zero, a useful next step would be a coordinated series of large, pre-registered direct replications of the original experiments (e.g., Baumeister et al., 1998).
I wouldn't be surprised if it does exist but only/mostly under some narrow set of circumstances, e.g. low glucose availability impacting decision making.