Azure
Nodes reported as idle while tasks are in running state during app package extraction
Problem Description
The monitoring state of Batch nodes is confusing while a big app package is being extracted. In this case the node is reported as idle which gives the impression that something is broken with the scheduler.
Steps to Reproduce
- Create a pool with a single node
- Upload an app package of several GB in size
- Create a job with a single task that has a dummy command line and a dependency on the app package
Expected Results
As soon as the task transitions to "running" the node should transition away from "idle" as well.
Actual Result
The node stays "idle" as long as the app package is being extracted, while the task transitions to "running".
Currently, this is by design, but I agree that the task/node state consistency is confusing. We're looking to improve node states in the future and this will be considered.
The same happens during container pull, the task is started but the node is idle. There is also no obvious way to get the docker pull logs (e.g. to monitor time taken for the pull).
