Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix AppState when Engine connection is terminated #6722

Open
wants to merge 8 commits into
base: master
Choose a base branch
from

Conversation

joaopamaral
Copy link

@joaopamaral joaopamaral commented Oct 7, 2024

🔍 Description

Issue References 🔗

This issue was noticed a few times when the batch state was set to ERROR, but the appState kept the non-terminal state forever (e.g. RUNNING), even if the application was finished (in this case Yarn Application).

{
"id": "********",
"user": "****",
"batchType": "SPARK",
"name": "*********",
"appStartTime": 0,
"appId": "********",
"appUrl": "********",
"appState": "RUNNING",
"appDiagnostic": "",
"kyuubiInstance": "*********",
"state": "ERROR",
"createTime": 1725343207318,
"endTime": 1725343300986,
"batchInfo": {}
}

It seems that this happens when there is some intermittent failure during the monitoring step and the batch ends with ERROR, leaving the application metadata without an update. This can lead to some misinterpretation that the application is still running. We need to set this to UNKNOWN state to avoid errors.

Describe Your Solution 🔧

This is a simple fix that only checks if the batch state is ERROR and the appState is not in a terminal state and changes the appState to UNKNOWN, in these cases (during the batch metadata update).

Types of changes 🔖

  • Bugfix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)
  • Breaking change (fix or feature that would cause existing functionality to change)

Test Plan 🧪

Behavior Without This Pull Request ⚰️

If there is some error between the Kyuubi and the Application request (e.g. YARN client), the batch is finished with ERROR state and the application keeps the last know state (e.g. RUNNING).

Behavior With This Pull Request 🎉

If there is some error between the Kyuubi and the Application request (e.g. YARN client), the batch is finished with ERROR state and the application has a non-terminal state, it is forced to UNKNOWN state.

Related Unit Tests

I've tried to implement a unit test to replicate this behavior but I didn't make it. We need to force an exception in the Engine Request (e.g. YarnClient.getApplication) but we need to wait for the application to be in the RUNNING state before raising this exception, or maybe block the connection between kyuubi and the engine.


Checklist 📝

Be nice. Be informative.

@joaopamaral joaopamaral marked this pull request as ready for review October 8, 2024 17:31
@pan3793
Copy link
Member

pan3793 commented Oct 18, 2024

@joaopamaral please fix the code style. (simply run dev/reformat if you are using Linux or macOS)

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants