TPS drop when using B_linear #146
Description
I recently migrated a server map from MCA to LinearV1 using this converter, then installed it on the server using the b_linear format. However, I made a mistake and forgot to change the compression level from 1 to 6 (the map was originally formatted with level 6).
About 12 hours after the server was up and running, this bug first appeared – the server console started filling up with errors like this:
[20:55:15 WARN]: Exception in thread "BufferedLinearRegionFile I/O Worker 20452" java.lang.RuntimeException: java.io.IOException: Failed to sync to master file!
[20:55:15 WARN]: at me.earthme.luminol.utils.BufferedLinearRegionFileFlusher.lambda$run$0(BufferedLinearRegionFileFlusher.java:97)
[20:55:15 WARN]: at java.base/java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1144)
[20:55:15 WARN]: at java.base/java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642)
[20:55:15 WARN]: at java.base/java.lang.Thread.run(Thread.java:1583)
[20:55:15 WARN]: Caused by: java.io.IOException: Failed to sync to master file!
[20:55:15 WARN]: at me.earthme.luminol.data.BufferedLinearRegionFile.syncToMasterFile(BufferedLinearRegionFile.java:162)
[20:55:15 WARN]: at me.earthme.luminol.data.BufferedLinearRegionFile.syncIfNeeded(BufferedLinearRegionFile.java:142)
[20:55:15 WARN]: at me.earthme.luminol.utils.BufferedLinearRegionFileFlusher.lambda$run$0(BufferedLinearRegionFileFlusher.java:95)
[20:55:15 WARN]: ... 3 more
[20:55:15 WARN]: Caused by: net.jpountz.lz4.LZ4Exception: Malformed input at 13
[20:55:15 WARN]: at net.jpountz.lz4.LZ4JavaUnsafeFastDecompressor.decompress(LZ4JavaUnsafeFastDecompressor.java:78)
[20:55:15 WARN]: at net.jpountz.lz4.LZ4FastDecompressor.decompress(LZ4FastDecompressor.java:109)
[20:55:15 WARN]: at net.jpountz.lz4.LZ4DecompressorWithLength.decompress(LZ4DecompressorWithLength.java:185)
[20:55:15 WARN]: at net.jpountz.lz4.LZ4DecompressorWithLength.decompress(LZ4DecompressorWithLength.java:204)
[20:55:15 WARN]: at me.earthme.luminol.data.BufferedLinearRegionFile$CompressingOps.fromCommitedSection(BufferedLinearRegionFile.java:691)
[20:55:15 WARN]: at me.earthme.luminol.data.BufferedLinearRegionFile.readChunkDataRaw(BufferedLinearRegionFile.java:449)
[20:55:15 WARN]: at me.earthme.luminol.data.BufferedLinearRegionFile$LinearMasterFileFrameParser.writeMainFile(BufferedLinearRegionFile.java:969)
[20:55:15 WARN]: at me.earthme.luminol.data.BufferedLinearRegionFile.syncToMasterFile(BufferedLinearRegionFile.java:157)
[20:55:15 WARN]: ... 5 more
Meanwhile, the thread number in each new exception (the number after "Exception in thread "BufferedLinearRegionFile I/O Worker") increased, reaching several million.
After a few hours, the TPS began to gradually decrease, dropping to 3. A restart helped resolve the issue, but only for 3-6 hours. I suspect this is related to a memory leak and heap overflow (since the number of GC calls increases over time).
I already thought the map had been corrupted, but no — players weren't reporting missing chunks or entities. I started digging further.
After trying to convert the format back to MCA using the same tools, I was getting an error about an invalid Superblock for some regions, causing those regions to be skipped. After further investigation, I noticed that the "corrupted" region files had a version of 2, while the valid ones had a version of 1. The signature was also different (0xffffdff7eddafd97, while the standard signature should be 0xc3ff13183cca9d9a). Perhaps B_linear changes it? I don't know.
I'll repeat again that there have been no issues with map corruption. Not a single player has reported missing files yet. anything. So, it's most likely a Luminol or format software issue, so I'm asking for help here.
I'm happy to provide any additional information you need.
Thanks in advance.