Buffers in the boot loader now go to a separate directory from the map.
Better handling of UID allocation during boot loader. For unique objects, directly allocate a UID and avoid going through the XID -> UID mapping.
Keep the mapper buffers in memory. No need to write to disk.
Use zstd for compressing map files. This produces files roughly half the size compared to snappy. zstd is pretty fast too.
TODO is to switch to using DataDog/zstd entirely, as it consumes much less memory.
WIP: Attempt to skip keys with millions of instances, aka Jupiter Keys. These keys are causing the reducer to fail due to buffers exceeding 128 GiB in size.
Bench Tool:
Update the benchmark tool to include JSON-RPC endpoints and the new 0xfast stream service. Add logs for all these services.
Boot Loader:
Bench Tool:
Badger:
This change is