The inflow of Wikitext events into eventlogging is to high for the system to the point that yesterday vanadium stopped processing events due to the VE inflow. Intake of events is 200 events per second about 150 of those events (per second) are invalid.
This can be fixing by sampling events, which is done for pretty much any incoming schema. Since VE needs all events for a session the sampling should happen around token, via bucketing. For example, if session token is between 1 and 10 we can send events it such if session token is <= 5, that would be a sampling of 50%.