Tracker gets very big for long conversations

When a user conversation gets very long, the tracker will get huge and might slow down everything. Possible solutions:

• implement some sort of session concept for the tracker
• only return the last x events

Potential Difficulties:

• we need at least max_history action events for the predictions, but there can be revert events in between, which requires us to pull more events from the database
• users might need more than max_history events in the action server to access certain information

Things do when tackling this issue:

• Creating an implementation proposal
• discuss the proposal
• implementation

[souvikg10]

Just one question, the tracker depends on the max history parameter of the training policies right? maybe it should be possible to do hot/cold switch to reduce the footprint. two KV stores perhaps with the hot store emitting x events which depends on the max history size.

I would hate to lose the conversations in the tracker store if deleted after a user session without proper archiving as such. I am assuming most folks are using RDMS to store the tracker and later use it for analytics

[teremterem]

@benos It’s always useful to share findings, even if the thread is old. This information may then be googled by someone later. This is why I chose to comment.