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

Reach-Frequency Variance in LLv2 measurements is being calculated as a negative value causing an error in report retrieval #1858

Closed
harrymeadows opened this issue Oct 15, 2024 · 5 comments
Assignees
Labels
bug Something isn't working

Comments

@harrymeadows
Copy link

harrymeadows commented Oct 15, 2024

Describe the bug

  • Multi-EDP reports submitted in Origin Staging environment (Halo v0.5.8).

  • All direct and LLv2 requisitions are successfully fulfilled by EDPs.

  • All measurements appear to be successfully computed by the MPC consortium (according to Kingdom spanner)

  • However, the report cannot be retrieved and remains 'in progress' in the Origin UI.

  • Following error is observed in reporting server logs:

  • Unable to get metric. Unable to compute variance of reach-frequency metric. The reach variance cannot be negative.

Steps to reproduce

  1. Submit multi-EDP report (single media type) for a date range of your choice (report in question was 1 month)
  2. Await completion

Component(s) affected
Error logs identified in reporting server

Version
v0.5.8

Environment
Origin Staging env (Prod mirror)

Additional context
In Origin Production and Staging environments, the differential privacy params configuration files are following this Confluence page https://isbaorigin.atlassian.net/wiki/spaces/INT/pages/549290009/Privacy+Params#Origin-Production-Privacy-Params

The key difference is that:
In Production (v2.4) we have 2 configuration files. One for direct requisitions and one for LLv2 requisitions.
In Staging (v2.5) we only 1 configuration file, for both direct and LLv2 requisitions

@harrymeadows harrymeadows added the bug Something isn't working label Oct 15, 2024
@kungfucraig kungfucraig self-assigned this Oct 15, 2024
@kungfucraig
Copy link
Member

@harrymeadows can you provide the report definition here?

Meanwhile, @tristanvuong2021 is ensuring a single Metric variance computation failure doesn't fail the whole report. That will allow us to determine exactly which Metric is causing the problem.

@ple13 is looking into why we might be getting a negative variance. We think it may have something to do with a unique reach calculation, but seeing the report definition would allow us to refute this if there are no unique reach computations.

Will update here as we know more, and daily on slack.

@SanjayVas
Copy link
Member

SanjayVas commented Oct 15, 2024

@harrymeadows I took the liberty of updating the initial filing to clarify this is in Origin environments.

However, the report cannot be retrieved and remains 'in progress' in the Origin UI.

For future reference, please keep filings focused on what is happening in Halo software.

@harrymeadows
Copy link
Author

report stg.json
@kungfucraig - report definition shared by Sergejs on slack is attached

@kungfucraig
Copy link
Member

This was determined to be an EDP issue. However, error messages in the variance calculator have been improved.

@SanjayVas
Copy link
Member

Error message improvement is in #1860

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
bug Something isn't working
Projects
None yet
Development

No branches or pull requests

3 participants