1159004 - [docker-worker] [raptor] Add ramdisk support to taskcluster

Status: RESOLVED WONTFIX

(Taskcluster :: Workers, defect)

Description

[Robert Wood [:rwood]]

Ideally for raptor performance tests the emulator would be run on a ramdisk. The taskcluster worker would need to be able to create a ramdisk on the aws instance host, before the docker container is started; and then volume mount the ramdisk into the docker container at container start up.

Comment 1

[Robert Wood [:rwood]]

Hi Greg, just a ping, any word on ramdisk support for TC? Is that something that is still being planned? Thanks!

Comment 2

[Greg Arndt [:garndt]]

It's not something purposely ignored, just keeps getting bumped down on the list of things to do.  

We mostly likely could add something to the task definition to specify that a ramdisk is needed, where it should be mounted within the container, and the size of it.

The harder part is creating something in the worker to make sure that tasks are not claimed if there is not enough RAM available for tasks it could claim (would probably be another device manager within the worker).

Do you know what the sizes of the ramdisks would typically be?

Comment 3

[Robert Wood [:rwood]]

That sounds perfect re: task definition. As for ramdisk size, 1GB would be great. The tests that I did awhile back used a 1GB ramdisk and the emulator worked well on that, thanks.

Comment 4

[Robert Wood [:rwood]]

Actually please hold off on this, I notice the emulator-l size on disk is 1.2 GB so the ramdisk would have to be at least 1.5 GB. Let me do some tests on emulator-l first when it is more stable, and see if running on a ramdisk still improves speed and variance or not. It might not be worth adding ramdisk support if variance isn't improved significantly on emulator-l, thanks.

Comment 5

[Robert Wood [:rwood]]

This is no longer required (at least not for Raptor, as it is being removed from gaia-ci).