CODEWITHSUNDEEP
scalabilitygoogle-compute-engine
Deleplace
Deleplace

Reputation: 7792

Compute Engine Autoscaler : choice of the machine to terminate

The Autoscale (Beta) documentation states :

[The Autoscaler] randomly selects the virtual machines to terminate.

This is a problem because a machine may be running a long process, and get abruptly terminated... even if other machines are idle! Do you know a workaround to prevent tasks from being aborted ?

It would make sense IMO to use a heuristic to determine the machine to be terminated, based on CPU or other metrics. The shutdown script can be used to perform some cleanup actions and gives a 90s relief, but doesn't solve the abortion problem.

My use case is :

Tasks may run for a while (up to 15mn).

Upvotes: 2

Views: 91

Answers (1)

vaqars
vaqars

Reputation: 136

You are the correct. The Auto-Scaler does terminate instances randomly. From my understanding, the Google Auto-Scaler team is aware of the customer demand on this and are working to get this feature implemented.

Upvotes: 1

Related Questions