What happens to requests in Instance Request Pending Queue when a backend instance gets terminated due to “Exceeded soft private memory limit”?

[Ankur C]

down votefavorite

I have B* instances running on App Engine(Python env) to serve user facing requests. Sometimes I see B* instances getting terminated due to Exceeded soft private memory limit I understand that increasing the instance class will solve the issue but I have few queries regarding requests that are present in the Instance Pending Queue ! Assume we have 2 instances of B* instance class and we call it let say => I-1, I-2 What will happen to those requests that are there in I-1 Instance Request Pending Queue after the I-1 instance gets terminated due to some reason? Will those requests gets evicted from the instance queue as this instance got terminated ? Will the requests in Instance Pending Queue are Dequeued from the Instance Pending Queue for I-1 and will be put in I-2 Request Queue by the Request scheduler as soon as Request Scheduler finds that I-1 is shutting down due to some reason. Any help regarding understanding these things will be highly appreciated !

[Yannick (Cloud Platform Support)]

Hello Ankur,

Normally, those requests should be rerouted to other available instances or be converted to loading requests (if applicable).

It's possible that if the startup time of a particular app is too long and the scaling parameters are tightly constrained, pending requests will timeout.  Generally, though, these requests should seamlessly route to available instances.