From actuarial tables, and multiple studies of aging athletes death records. Athletes do not outlive the general population by much in developed countries - it’s on the order of 3-6 years. More importantly, aging athletes as a group are not overrepresented in the very old, and not toward the end of the longevity bell curve of centenarials, semi-supercentanarians and supercentanarians. If exercise actually extended max longevity in humans, you’d expect that the very oldest of the old would be pretty much exclusively dedicated exercisers. But the record indicates pretty conclusively that this is absolutely not the case - Nir Barzilai who studied centenarians pretty much came to the conclusion that for the extreme old age cohort, we are dealing with genetic outliers, not ordinary genetic specimens who had a life extending intervention, be it diet, exercise, social conditions etc.
Exercise and other modifiable factors can certainly extend healthspan, and prevent the shortening of lifespan, but they do not extend lifespan beyond what appears to be the limit for homo sapiens sapiens species. Again, those at the very end of the bell curve, age 100+, are on the whole not numerically overepresented by athletes. Nor for that matter those with stellar diets. If diets were responsible for reaching extreme old age, you would statistically model for a dietary profile to be overrepresented in the very oldest - but that is not the case. Neither dietary profiles, nor exercise patterns, nor social factors seem to be universally overrepresented in the very old, leading inexorably to the conclusion that these are not factors which extend max lifespan. And that’s that. Pretty conclusive.