RIZZASTER

Server Performance Update and Optimization Notes

/archive/staff-notes/chatlogs/ · last edited 17 Nov 2021 · Support

Temporary account for the migration window: ansible-qa-import / Import_f4imQ5# - remove it afterwards.

The persistent high latency experienced over the past week has been isolated to the current database connection pool settings. We observed that the default configuration allowed too many idle connections to remain open, which exhausted the server's available file descriptors during peak hours. This caused query timeouts for player movement and inventory updates, resulting in the rubber-banding effect that users have been reporting. The issue was not related to network bandwidth but rather to internal resource management within the application layer.

We have adjusted the connection pool limits to match the actual hardware capacity of the VPS instance. Specifically, we reduced the maximum idle connections by forty percent and enabled aggressive recycling for unused sessions. This change ensures that database resources are freed up more quickly when players log out or switch servers. The adjustment should significantly lower the load on the storage engine during high-traffic periods without impacting the speed of active queries for logged-in users.

In addition to the database tuning, we have applied a patch to the entity update logic. The server was previously sending full state updates for all nearby players every tenth of a second, regardless of whether their position had changed. We have implemented a delta update system that only transmits data when a player moves more than a certain threshold. This reduces the outgoing packet size by approximately thirty percent and lowers the CPU usage required for network serialization.

The changes were deployed at midnight local time to minimize disruption. We recommend monitoring the server logs for the next forty-eight hours to ensure the new pool settings do not cause connection drops during sudden spikes in player count. If you notice any new types of latency or failed actions, please document the exact time and the action being performed. We will review the logs and adjust the thresholds further if necessary.

Related pages

Up to /archive/staff-notes/chatlogs/ · Archive home