1453649 - Missing required properties in /v1/state/<worker-type>

Status: RESOLVED FIXED

(Taskcluster :: Services, defect)

Description

[Hassan Ali (:hassan)]

Attachment: Screen Shot 2018-04-12 at 9.37.29 AM.png

There may be a bug with the /v1/state/<worker-type GET endpoint. It is not returning the `type` and `ami` property inside the `instances` property: 

https://github.com/taskcluster/aws-provisioner/blob/cb1aed6de14b8617400d9f03810f98b2a3963c6c/schemas/state-definition.json#L28-L37

As a result, because the `type` is not returned, the worker-type status tab shows 0's in the UI for the running and pending capacity columns (see screenshot).

Comment 2

[Hassan Ali (:hassan)]

John, any idea why this is happening?

Comment 3

[John Ford [:jhford] CET/CEST Berlin Time]

(In reply to Hassan Ali (:hassan) from comment #2)
> John, any idea why this is happening?

I'm not sure why.  The EC2-Manager endpoints are returning this information, but the provisioner is not.  I'm not entirely sure why, but the problem is in the /state methods of the provisioner.

Comment 5

[Chris Cooper [:coop] (he/him)]

Can we get this prioritized, please? During outages (like today), it's hard to pinpoint bad regions/AZs without this data.

Comment 7

[Hassan Ali (:hassan)]

Attachment: resources tab

The instance type and AMI columns in the EC2 Resources tab are also broken in production.

Comment 8

[John Ford [:jhford] CET/CEST Berlin Time]

Attachment: patch

This was broken when case started to be preserved in the ec2-manager database.

Comment 9

[John Ford [:jhford] CET/CEST Berlin Time]

Attachment: GitHub Pull Request

This also adds much more detail to the state schema so that we can catch something like this in the future.