[image-generation] support userdata overriding worker version to run
Categories
(Taskcluster :: Workers, enhancement)
Tracking
(Not tracked)
People
(Reporter: miles, Unassigned)
References
(Blocks 1 open bug)
Details
This bug covers ensuring that we can control the versions of workers that worker-runner starts via config in userdata.
This is important for allowing fast roll-forward changes when time does not allow for a full image generation cycle. This can also let us roll new images with older versions of workers for compatibility testing, rather than simply using old images baked with a specific worker version.
A cloud-init / cloudbase-init script that runs on instance start and downloads the correct version of the worker as specified in userdata could satisfy this requirement. Such a script would be provisioned into the base image and run by instances based on that image on startup.
For Linux, we can add a file to cloud-init's expected scripts directory into monopacker's files directory. These files get untarred on the base image filesystem.
|
Reporter | |
Comment 1•5 years ago
|
||
taskcluster-worker-runner queries userdata for taskcluster credentials and other relevant information on startup.
We could overload userdata to specify versions of workers in addition, but that would be an essentially unspecified contract between worker-pool configuration and worker images.
This would have been nice to have but isn't a blocker. Marking it WONTFIX.
Description
•