There is no solution in the admin settings, nor according to their enterprise support. It is plain old tough luck. The vast majority of our traffic is from de

There is no solution in the admin settings, nor according to their enterprise support. It is plain old tough luck. The vast majority of our traffic is from devices that retry later if anything goes wrong. It is analytics information and hence there is no human affected by it. Despite having identical settings and code across a number of appengine apps, we see variance in the number of instances they decide to run (eg they will start 4 at once even though only 1 is needed for the load). At one point they were starting hundreds of instances due to some of our aggressive inter-app traffic (thundering herd problem). Things would have worked fine if we could limit the number of instances, but tough luck there. It required considerable engineering and tuning to make the problem go away that would have trivially been solved (in the short/medium term) by limiting instance numbers.