Hey everyone!
I’m struggling to get Phoenix.Presence to scale up sufficiently for our needs. I feel like Presence might be overkill in this situation, but I’m not 100% sure what the alternatives would be.
Our Issue
The Presence.track is timing out around 80-100k connections. Our load tests are joining and leaving LOTS of times every second. That creates lots of traffic for the presence diff. Eventually, the whole cluster just stops tracking until it has time to catch up.
Some Things We Have Tried
- Reimplemented Phoenix.Tracker and took out the broadcast on join/leave
- Implemented a dirty count that reaches into Phoenix.Tracker.Shard.dirty_list and then does not group them, but just counts them, to reduce cycles during the count
We’re basically just doing presence so that we can get a valid count of the connected users. What are some alternatives to phoenix presence when you only need to track a count?
