Bad performance in large servers
peterbro1 opened this issue ยท 1 comments
Describe the bug
There seems to be a memory leak involving network markers that becomes critical at long server uptimes.
To Reproduce
Steps to reproduce the behavior:
- Fresh 1.12.2 server
- Add protocol lib
- 200+ players on/off for 10+ hours
Expected behavior
Protocol lib uses a relatively constant amount of memory
Screenshots
https://gyazo.com/c7046e4929b230fd040b0122f7cadb99
https://gyazo.com/9fa36aa2b92d355635b364f594f6fc4d
https://gyazo.com/155f3bbac5ac49d6c4eaade4c64bfcd9
Version Info
https://pastebin.com/vEJrEsZ5
Additional context
I don't have a lot of time to work and study how the plugin works, but I believe it to be an issue with weak references and packet hashcoding. The long cpu time can almost certainly be attributed to the packetMarker WeakHashMap in ChannelInjector.java. It is using the Object class's default hashcode function, which is very slow. Whatever the value half of the entry is inside the packerMarker map, it should have a wrapper and overridden hashcode function to preserve performance. The memory leak could be anything- I have no idea, but am not getting a whole lot of progress looking through the code. Hopefully you can help!
