Recent "Strange" Work Distribution Patterns from Livepeer Inc. Broadcasters

There have been a number of reports of “strange” work distribution patterns from Livepeer Inc. broadcasters recently where Os with smaller amounts of stake appear to be receiving significantly more work than Os with larger amounts of stake despite these Os having almost the same performance.

There has been some speculation that this commit in go-livepeer discovery: Include Pending Os in discovery (#2188) · livepeer/go-livepeer@9ab7a88 · GitHub could be the cause. The core team doesn’t currently believe that this is the cause because while this commit changed the default behavior of broadcasters, there are different discovery implementations that a broadcaster can use and the Livepeer Inc. broadcasters currently uses a discovery implementation that relies on data from the Livepeer subgraph which just considers Os that are active (although this will change soon at least for the Confluence upgrade to ensure a smoother upgrade).

The core team is currently focused on the upcoming Confluence upgrade which is an all hands on deck effort so we do not expect to investigate further until the upgrade is complete. Once the upgrade is complete, we will follow up here and provide more information about what was/is happening here.


Thank you for acknowledging this. I for one am quite happy with this strange behavior but I understand the massive shift in work distribution wasn’t intended. I really hope that when this is looked into and resolved, smaller high performing O’s don’t lose ALL their traffic.

Sooo this pretty much happened. Oh well, it was fun while it lasted!