redis scaling

[hweihwang]

Summary

• Strengthened multi-node orchestration so presentation, viewport, recording, timer, and voting stay in sync across nodes and late joiners always hydrate current state via Redis Streams.
• Hardened cluster sweep and shared state cleanup to drop stale presentation/recording/timer/voting entries before notifying rooms, preventing ghost sessions.
• Improved node failover handling so syncer reassignment, presenter disconnects, and recorder moves remain consistent during node shutdowns or restarts.

Rationale

• Ensure every collaborative feature survives node changes, reconnects, and late joins without divergence.
• Keep shared state authoritative in Redis so any node can serve correct context immediately.
• Reduce operational risk in clustered deployments by cleaning stale data and making failover deterministic.

Manual Test Cases

• Presentation: start/stop across nodes; viewport updates reach followers; late joiner sees active presentation; presenter disconnect stops session.
• Recording: start and stop from different nodes; recorder reconnect preserves metadata; stale recording entries removed on join.
• Timer: start/pause/resume/extend/reset across nodes; late joiner sees current timer; state cleared when room empties.
• Voting: start vote, cast votes from multiple nodes, end vote; late joiner sees active vote; state cleared when room empties.
• Cluster sweep: seed stale presentation/recording/timer/voting entries, run sweep, confirm stale entries removed before room notifications.
• Node failover: shutdown syncer node and see reassignment; shutdown presenter node and verify stop broadcast; recorder node shutdown reassigns syncer.
• Restart resilience: restart a peer node and verify shared session state persists and late joiners still get active presentation/timer/vote state.

[hweihwang]

Currently focus on rebasing and align the logic for Timer and Voting features there, update the progress and description cases above soon

cc @juliusknorr

[hweihwang]

Updated there @juliusknorr !!!

[juliusknorr]

Thanks, I'll have a look 👍

[juliusknorr]

This may hide a critical error, @benjaminfrueh suggested that we may still throw here instead

[benjaminfrueh]

I added console.error in these cases of silent failures of write operations for now.

[juliusknorr]

Re-tested with and without redis and multiple websocket server nodes.